May 13, 2025 • 64 mins
Most freight tech stacks are a tangled mess of tools that don’t talk to each other. Ezequiel Peralta, VP of Technology at SPI Logistics, is working to fix that.
This episode strips away the hype and digs into what it really takes to build functional, secure, and scalable freight systems. Peralta shares how SPI is moving beyond the TMS-first mindset by introducing a flexible integration layer, leveraging AI to reduce email chaos, and automating back-office tasks without sidelining the people who matter most.
Freight leaders looking to create tech that works for humans, not the other way around, will get a lot out of this one.
Key Takeaways:
- Freight tech success depends on modeling your business process before adopting any tool.
- SPI Logistics is moving beyond the TMS as the tech stack centerpiece with a dedicated integration layer.
- AI tools are only valuable when paired with structured data and business-aligned goals.
- Automation is transforming back-office operations, reducing manual tasks for better auditing and analysis.
- Everyone—not just IT—must be involved in managing freight fraud and data security.
LINKS:
Feedback? Ideas for a future episode? Shoot us a text here to let us know.
-----------------------------------------
THANK YOU TO OUR SPONSORS!
Are you experienced in freight sales or already an independent freight agent? Listen to our Freight Agent Trenches interview series powered by SPI Logistics to hear from the company's agents on how they took the leap and found a home with SPI freight agent program.
CargoRex – Your Logistics Hub. Explore, discover, and evolve with the all-in-one platform connecting you to the top logistics tools, services, and industry voices. Whether you're a leader, researcher, or creator, CargoRex helps you stay ahead. Explore Now
Digital Dispatch maximizes your #1 sales tool with a website that establishes trust and builds rock-solid relationships with your leads and customers. Check out our website services here.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Eze Peralta (00:05):
When you're doing something by yourself, if I'm
trying to solve a problem and Idon't know how to do it, then
maybe I Google, maybe I use AIor whatever, but I'm still using
my brain to solve the problem.
Once I solve it, I learnedsomething, right? Same with
human interaction, maybe someemployee making a mistake also
(00:27):
reveals that there is a processthat needs to be improved,
right? So then you can so Ithink error, and it's part of
life, and it's also part oflearning. And I think we are
entering to this notion of we,we have these agents doing all
(00:49):
the work for you, but then afteryou see the work done, have you
learned something from thatprocess?
Blythe Brumleve (01:01):
The freight tech hype cycle is real, but
behind the buzzwords are folksactually building the plumbing
that keeps your freight offsrunning as a Peralta is one of
them, and he is back with SPIlogistics, VP of technology on
the everything is logisticspodcast, in order to break it
all down for us, I'm your host,Blake Milligan, and of course,
(01:21):
we were presented by SPIlogistics, and I was just
telling Eze his last appearanceon the podcast a couple years
ago, I believe is our mostpopular SPI logistics episode.
So sorry to everyone you knowwithin SPI that's come on the
show, but we will love Eze. We
Eze Peralta (01:40):
We are good friends. So I so it's all good.
I think.
Blythe Brumleve (01:44):
no competition.
So let's, let's set the stage.
So for SPI, for folks who maynot be familiar with your role
within SPI, can you kind of giveus a high level overview of what
you do on maybe a day to daybasis, a weekly basis. What does
I guess a week in the life of Zalook like?
Eze Peralta (02:06):
Well, there's we have two main areas. So one is
the things that we are buildinginternally. The other area is
the things that we need tomaintain or integrate or are in
relation to tech vendors. And sothere's a bit of swing between
(02:28):
between those two aspects. So alot of the time is meetings,
meetings with vendors trying toalign the road maps of what
we're trying to do with whatthey're trying to do, trying to
keep up with all the the newfeatures that the new vendors
(02:48):
that come up. It's it's a lot oftime, because evaluating tools
and evaluating next steps andtrying to sometimes might
maintain the roadmap, but at thesame time being flexible enough
to pivot. So a lot of a lot ofthe time is, is that kind of
more strategic point of view?
And then there's a lot of timethat is just coding or, you
(03:11):
know, jumping into the theactual implementation, a lot of
modeling and diagramming withthe with the business users as
well. A great deal of the ofsoftware development. A lot of
people focus on on the codingside with software development,
but in my experience, softwaredevelopment is a lot about
(03:35):
modeling and understanding whatyou need to build before
starting to build. Right? It'slike, for example, you're going
to build a house. You're notgoing to start, you know,
cutting wood and laying bricksright away. You need some sort
of blueprint. You need some sortof understanding of what needs
(03:56):
to be built. You need tounderstand, even, you know the
terrain where you're going towork with and, and, and a lot of
other factors. So a lot of thetime goes into that as well. And
Blythe Brumleve (04:09):
so for folks who may not be aware, you know,
SPI Logistics is a freightagency, and so you don't have
any some freight agencies willhave an in house brokerage team
or an in house trucking team.
SPI doesn't do that because theytheir internal team is set up
and dedicated just to supporttheir agents. And so you guys
are, I believe and correct me ifI'm wrong, but you guys have a
(04:31):
general tech stack that you use,but then you also accommodate
the freight agents, and maybe ifthey have, like, a particular
tech tool that they like to use,you'll incorporate that into
sort of, I guess, the mothershipof SPI Is that accurate?
Eze Peralta (04:48):
Yes, yes, correct.
So we try to embrace also thatdiversity in our agent agents
network. And you're correct.
Yes, we you. Where we spendquite, quite, quite a bit of
effort on understanding thoseneeds and also offering the
(05:08):
tools that we have to enhancethe existing operations when,
when agents on board with us. Sobecause maybe they already have
certain workflows that that,that they already applying, and
that's effective for them. So wedon't force anybody to, okay,
this is the way how you shoulddo it, but we do offer the tech
(05:29):
stack we have, but also open toincorporate and integrate other
technologies that agents mightbring when, when they on board
and, yeah, trying, trying toaugment and enhance their
operations and and then thisgoes in line we're saying
before, of understanding what isneeded before going into the how
(05:51):
we're going to do it right. Oncewe know what is needed and what
is the most helpful for ouragents, then we can decide,
okay, you know, how we implementit sometimes, that is getting a
new vendor. Sometimes is, youknow, building some
customization on an existingtool. It could take any shape,
(06:12):
basically.
Blythe Brumleve (06:14):
Are there any sets of tools that your agents
are using every day?
Eze Peralta (06:21):
Well, the TMS is what our agents use every day.
And I think the operations wehave the track and trace
capabilities that that sometimessome agents would go directly
into the tracking portalsbecause they have specific
needs, or some other agents willuse them. Use these tracking
(06:44):
capabilities within the TMS.
Similarly with posting,sometimes people would use, you
know, posting directly to theboards. Sometimes people
wouldn't do it through the TMS,but we try to have a layer of
control of all that data andwhere it goes, where it's coming
from. So we can also offer someinsights and analytics on on
that data as well. And then wehave, yeah, for example, tools
(07:08):
for procurement, capacity orrisk management that we created
internally. And a lot of thetimes, what we do is try to
aggregate that very vastlandscape of data into more
actionable information. So youknow, instead of our agent
having to go into five differentportals to make sure that their
(07:28):
carrier is not fraudulent, thenwe kind of like consolidate all
that data. We say, hey, yeah,this is, this is a good to go
carrier, or this is flagged forthese reasons, yeah, trying to
also combat that tiredness thatcomes sometimes people drop the
(07:48):
ball on certain things relatedto risk, because it's just too
cumbersome going into differentplaces to having to check in 10
different portals To make surethat you know your your provider
is good to go. So people stopdoing it right, people, people
stop checking because it's justcumbersome. So we try to also
(08:09):
think about that, and, like,make sure that we make it easy.
So, so those tools are, areused. Oh, that's super
interesting.
Blythe Brumleve (08:17):
So you essentially are, you know, using
different maybe, and I'm justspitballing here, but like a
highway and a carrier, sure. Andyou know, some of these, these
vendors that do are in the samerealm, but do things a little
bit differently, but they allhelp give a fuller picture. And
so you're able to build thatinto a dashboard for your
(08:38):
agents,
Eze Peralta (08:39):
yeah. And sometimes, even aggregating it.
We have tools where the resultis, you know, you're good,
you're not good and up to thatpoint. And then if you want to
see more details, you can gointo the portals, or you can
expand on those details. Butyes, we do that, like with all
the compliance vendors we have,because sometimes, you know,
maybe highway has some insightsor flags about specific things,
(09:03):
like, for example, where thesepeople, these carriers, are
accessing the system from. Butthen maybe freight guard, they
have very you know, if peoplewho are reporting these carriers
but right care for one, forexample, so you need both right?
Because one will give you, youknow, certain insights. The
(09:24):
other one is going to give you,I don't know, highway for
example, give you insuranceinformation, inspection
information, but if someone gotburned by by a bad actor, the
first thing they going to do is,most likely, is reporting it to
401, so we also want to, youknow, action on those, those
reports as well.
Blythe Brumleve (09:45):
So one of, and I heard you talking about this
with Chris jolly on the freightfreight coach podcast that you
guys were at technovations,Tia's, technovations in in late
2024 I'm sorry I missed that. Itwas actually it. Jacksonville,
Florida, but I was gettingmarried that week, so I
couldn't, I couldn't sneak awayto go to that to go to that
(10:06):
conference, but I heard that AIwas just everywhere, as it is
with most you know, freightconferences right now, what's
what's your? I hear you laughingthere. What's your, I guess,
overall take on, is it hype? Isit a buzzword? Is it? Yeah, is
it valuable? Is it kind of allof the above?
Eze Peralta (10:26):
I think, I think it's all of the above. And I
think AI is a very vast field,so it's not the same. Saying,
You know what? We are using thehype, this current hype wave of
AI is most related to largelanguage models, which is, you
(10:47):
know, the chatgpt. Butartificial intelligence is
something that existed for awhile. Now, the computing has
gone, you know, it's moreavailable and for companies to
use these large language modelsand the training techniques have
improved, but before we hadmachine learning or other types
(11:08):
of of artificial intelligenceapplied to analytics or
different aspects, likeforecasting and these kind of
things. So I do find it veryvaluable for a lot of use cases.
We use it internally a lot inour like our dev team uses this.
(11:28):
Use it quite a bit. But also wewe are cautious to not try to
chase AI or implementation orany tool like AI or any other
tool, just just for, justbecause everyone else is doing
it, or just because, you know,it appears to be the solution,
(11:52):
magic solution, to all these usecases we we are of the idea that
alignment between business andtechnology is key. You know,
integration and governance ofall the integrations is key.
Governance of data is key. Sothose things what we focus on,
(12:14):
and if there is, you know, if AIcan help us to achieve that, to
achieve the business outcomesthat we want to achieve, and
improve the KPI we want toachieve. Then we evaluate it as
we evaluate any other tool, andthen we we apply it, right? But
we are, you know, we try to keepfocus on what is important for
(12:38):
our agents, what is importantfor for the business as an
organization. And, yeah, we douse AI and in many ways, and we
are adding more use cases to it.
But at the same time, I, I thinkthere is this notion that is
going to come and solveeverything for everyone, and
(12:59):
suddenly it's going to replaceworkers and all this thing and
like, that's that's notsomething that, I mean, we don't
see it that way as much. Yeah,you can, you can automate phone
calls and these things. But whatabout the when things go wrong,
nobody wants to talk with amachine. And most of the
(13:23):
logistic business is managingrisk right. And are you going to
let a machine to manage thatrisk for you? Are you going to
let the machine do thatnegotiation or do that, you
know, relationship building foryou. So it's yeah, if, if you
have a mature enough system,then you can connect with your
(13:53):
shipper via API, and then youdon't need to have like a voice
agent, you know, calling ashipper or but at the same time,
there are certain, there arecertain use cases that, yes, you
know, for support, we had theseautoresponders for phone for
years, where you say, hey, youknow, give me your load number.
(14:13):
You just type it in the phone.
And so it's not very No, it'snot that different than that.
And then, and then you have alsothe ethics aspect, which is, I
was reading a LinkedIn post. Idon't remember who it was from,
but it was something in the inthe lines of, I was talking with
(14:34):
an AI rep, and I asked if it wasan AI. They said no, but then
they found out that it actuallywas an AI, but it seems that
this AI was trained to lie,right, to say that it was not an
AI. So I think, you know, weshould keep being honest. So
that's non negotiable, right?
Blythe Brumleve (14:56):
No, that's a very good point, and I think
that that's something. That's aconcern that we've raised in the
past and in previous episodes onwhen do you disclose that you're
talking to an AI agent? And Ithink most people are fine
talking to an AI agent when theyneed an answer quickly about
something, where is my load?
Something like that, but for adeeper maybe customer
(15:20):
relationship. There is a levelof you should probably disclose
you know that that informationas quickly as possible, so then
that way the person can makethat preference on if they want
to talk to somebody real or talkto a virtual agent. I would be
curious to know what other maybeuse cases are you seeing? You
(15:41):
know, maybe two different usecases. One where, you know, it's
machine learning that's actuallydoing the job and you're using
that. But then, on the flipside, a large language model.
I'm curious about the use casesfor each of those that you're
seeing.
Eze Peralta (15:58):
So machine learning is mostly used on, you know,
analytics and forecasting orfinding anomalies or things like
this. So you have a you know,data from the last, let's say,
10 years, and then you try tofind, you know, seasonality. You
try to find, like, certainpatterns that would be more
(16:18):
difficult to find just manually,you're trying to dig into all
that data, so and so for thatis, is, that's a good, good use
case. And you can also train itto find trends, you know, oh,
we're, you know, when these,when these metrics are, how
(16:38):
metrics relate, oh, when thesetwo metrics are going down, it
means that your business istrending up, or, I don't know,
or you start finding thoserelationships. And so for
analyzing those huge amounts ofof data, it's, it's machine
learning, it's, it's, it'svaluable. And for large language
(16:59):
models, I think the importantpart to understand is that no
matter what the what the vendorssay, they don't reason right.
Reasoning is a different thing.
They just predict the nexttoken, the next the next symbol
that's this should be thereaccording to their calculation
or their training. So what Ifind really interesting for as a
use case is categorization andthen interpreting natural
(17:25):
language text, for example,emails, tagging an email saying
you can plug it into to yourinbox and say, Okay, I have 20
call requests. So in a day Ihave 20 quote request. I have 10
track and trace requests, orwhere's my load kind of
(17:45):
questions. Then I have 15carrier carriers asking for
posting. And then I have threecomplaints and and then you can,
you know, categorize all ofthat, and eventually you're
going to have enough data tounderstand your your own inbox,
for example, much better, right?
(18:09):
And then from there, you canautomate, oh, when, when these
type of request comes in, orthis type of email comes in,
then you can auto generate aresponse, as long as it fits
with the with that category,right? So categorizing, I would
say, understanding topics orsubject on on text, and then
(18:33):
creating some sort of simple,simple responses or template or
responses for certain type ofinquiries. Those are, I think,
really good use cases. And
Blythe Brumleve (18:47):
then I with, as you were talking, I was thinking
about, you know, is there any, Iguess, safety issues that go on
with, you know, using a largelanguage model that, you know,
maybe agents should be a littlecautious of, of putting, you
know, personal information maybeinto some of these systems. Or
is it that concern maybe largelyoverblown? No,
Eze Peralta (19:10):
I think, I think that it's a it all depends on,
on the architecture that youhave behind it, right? So, for
example, we were talking withour team about cases where you
see people doing this that theycall now vive coding, which is
like,
Unknown (19:27):
I hate that phrase.
Eze Peralta (19:30):
It is, what it is.
And yeah, and then people wereable to hack applications built
in this way in 30 minutes, justbecause the credentials were all
pasted over in the code. So youcould just go into the
repository, public repository,and see all the credentials for,
for, for different APIs andthings. Because, yeah, if, if
(19:56):
you. Know what you're doing, youwouldn't do that, right? So if
you don't know what you'redoing, then you maybe think it's
fine. You trust that the AI willdo it okay for you, and then you
just ask the AI, oh, make itsecure. That just doesn't work.
You can't just tell the Oh, makemake this application for me.
Make it secure. I wish it waslike that, but, you know,
(20:21):
there's so much more to it. So Ithink security is one of the
big, big topics to think aboutwhen thinking about AI
Blythe Brumleve (20:32):
for sure, what about from the email side of
things? Because I personallystruggle with sorting through my
email on a daily basis, and Ican't imagine what you know a
typical freight agent is goingthrough with their email inbox.
What kind of tools are youseeing that's helping to
categorize those emails? Is itall strictly done through the
(20:55):
TMS, or is it, you know, a bunchof different tools that that can
help, you know, categorize thoseemails? Maybe, you know, auto
respond to them. You know, getrid of the junk. Is it a bunch
of tools, or is it one or tworeally good tools?
Eze Peralta (21:08):
So, I mean, there are a lot of tools and probably
missing a lot here, but so arethe PMs that we use ravinova,
they are adding now featuresthat natively allow you to to
use AI and then connect youremail to it and, and they will
(21:29):
It will create loads for you.
You will create truck postingsfor you, and, and, and also,
with all the automation thatexists already available in
Salesforce, you can create theseresponses back and connected to
all the the the other existingwork. So, so for us, that's
that's a really good one, andlevity, I think, is doing a lot
of a lot of work on that aswell. I I personally like the
(21:51):
what they have there. And thenwhat I'm seeing also is that
even the existing client emailclients are starting to add all
these features internally, like,if you have, for example,
Outlook, you could grab, youknow, power automate, which is
their little automation tool,and you could have little
(22:13):
workflows that categorize emailsand use the AI Within the
Microsoft environment. And thatshould be at least, you know, if
you already trusting outlook,then you know you can keep your
your existing data alreadythere, and it will not leave
that, leave that environment andthat that might be interesting.
Gmail, I think it's also addingall of that. And I think
(22:38):
eventually all the gmss aregoing to start doing that. I
know that other GMs for brokersas well, have already that. I
know that. Yeah, tools like praythe also have some email
automation going on. I haven'ttested it freely, but yeah, I
(22:58):
think all of those tools shoulddo a good job on, on, on, you
know, categorizing emails. It'snot at the same time I've seen
more and more orchestrationtools, or the tools that allow
you, like, to create your ownautomations. And a lot of people
are starting to build their ownautomations around email. I was
(23:21):
talking with Sarah from otherlogistics, and she was telling
me about some AI tools and thatthey're using for some use
cases. And I think it's calledsola solar, solar AI. And so any
tool should work again. It allcomes down. I want to bring it
(23:42):
back to have you modeled yourbusiness process. Do you know
what your business process is?
Do you know how you're going tomeasure if your process is
working or not? Have you Do youknow what's the outcome that
you're trying to achieve?
Because if you don't know that,then you're going to go into it.
How the implementation withoutknowing the what, and then you
(24:02):
know it might fail, and thenyou're going to think that the
problem is the tool, that theproblem is, oh, this AI is not
working, or this doesn't work,but, but how? How are you even
measuring if it's working ornot? Because you didn't know
what your performance wasbefore, you don't know the
metric that you need to look atto know what your if your
(24:23):
performance improved. So thenit's like, it's all going to be
gut feeling. And of course, thegut feeling at the beginning of
implementing a new technology,the gut feeling is going to be
frustration, most likely. But ifyou have the metrics in place
and the business processesmodeled properly, then you're
going to realize that, oh, well,I'm frustrated because I don't
(24:47):
know how to use this tool, butobjectively, these metrics are
improving, right? So I likealways taking it back to that,
like the business process. Needsto be solid and needs to be
documented, needs to be refined.
And it means because, if not,you end up with the vendor
(25:11):
deciding for you how you shouldbe working. And that's, I think,
advice for anyone evaluatingtech. You know, tech vendors
will generally tell you thatthey can do everything. You're
going to ask them, Oh, can youdo this? They're going to say
yes, because generally, you'retalking with a project manager
or a sales a salesrepresentative. And then they're
(25:34):
going to go to the developers,like teams, like my team, and
they're going to say, Hey, Isold this. I sold this new
feature. Now you need to buildit, and then when you roll it
out, then you're like, well, butlike, how does this match with
my existing workflow? What? Andthere's a lot of friction. So if
(25:55):
you don't know exactly whatyou're looking for in terms of
business processes, it's goingto be harder for a vendor to to
be effective, right? And it's,sometimes it's not even a
problem of the vendor, because,you know, these are tools, and
if you don't know what to dowith the tool, then also it's,
(26:18):
you know, it needs to be on bothsides, a commitment on both
sides.
Blythe Brumleve (26:21):
Yeah, no one wants to do the the boring and
really underrated, challengingpart is documenting your
processes, because you have tobe prepared to rip them all out
if they're not working. But justsimply documenting them is, I
find it personal experience,just very like draining on my
brain, but once I get a dud,then I'm so relieved, because
(26:45):
then I can figure out, okay,what do I need to do? What can I
outsource? What can a tool do?
And I'm curious as as to how youhow do you approach helping
agents document their process?
Eze Peralta (27:01):
So something that we do is visiting agents,
because being on site with withthe people who are executing the
tasks and executing theprocesses, I'm able to
understand much better all thenuances, right? And also, you
(27:21):
know, I don't like this idea ofthe tech team being, like,
isolated, and then, you know,requirements come in, and
people, you know, the dev teamdoesn't know what the process
is, and we try to all of ourteam is involved in, in these
modeling sessions. And wegenerally we try, because we
(27:46):
know also that agents are verybusy, and we want to also take
out too much time or effort onthat. But when we do this visit,
we try to document as much as wecan about what we see, and then
we come up with a plan of, okay,this is this is the process that
we see. These are the points,the critical points that we
(28:06):
could maybe improve. We havethese tools available to help
you on this particular point andasking a lot of questions. I
think listening and asking it'skey, because, again, if we in a
way, the tech team is like avendor for the agents, right, in
(28:29):
a way, like it's an internalvendor, but we need to provide
something that makes sense, andin order to be able to do that,
we need to understand what theproblems are. And so yeah, a lot
of Yeah, listening and going onvisits, because on a zoom call,
you can get an idea, butsometimes you just need to be
(28:50):
there to really to hear thephone ringing and, oh, how many
you know, how much of the timeis being spent on answering this
call? Oh, what if we have a toolhere that can filter out these
calls and like, you know, tellyou which, which call you should
take first. And how do you, youknow, how does that impact your
(29:13):
overall time to cover a low forexample? Just, just to give you
one example, something that wedid was also one of our agents
was having a lot of inquiriesfor for posted loads from from
boards. And what we were theywere copying and pasting a
template to respond to theseinquiries. We're like, Oh, what
(29:36):
if we try to automate that? Sowe just set it said, you just
set up all the your pricing inthe TMS, and then we can just
find the load that you're askingfor and then give them back that
information. And if it'scovered, instead of telling
them, oh, it's covered, we findsimilar loads in similar lanes,
so we think, Oh, that load iscovered, but we have these other
(29:56):
loads, right? But we. Needed togo visit the agent to really
understand that, right?
Blythe Brumleve (30:06):
How do you approach, you know, when you're
you're thinking about, you know,onboarding a new agent, or maybe
it's a current agent that is, islooking at a new tool. How do
you decide what to say yes toand what to maybe caution them
on?
Eze Peralta (30:24):
Yeah, I don't want to be super repetitive, but I
bring it back to what is the usecase that you're trying to
resolve? What is the businessprocess that this fits in?
Because sometimes you realizethat by moving a few pieces in a
different part of the board,right then you are also, you
(30:46):
know, changing the tension onthe part that you're interested
in. So sometimes it's okay,let's try to quickly understand
at least the skeleton of theprocess that you're doing, and
understand if, if it's a tool,what you need, or maybe it's a
(31:07):
shift on some other structure,right? That might be, maybe it's
a tool that we already have andthat they don't know that we
have. Maybe this a trainingissue. Maybe it's a lot of times
comes down to training. It comesdown to knowing the the tools
(31:28):
that we have available, becausethere are so many that sometimes
it's like, well, if you're anagent who is doing truckload
only, maybe you don't even knowthat we have an LTL program, or
that we have a customer portalthat so they maybe are looking
for a vendor that might givethem a portal for LTL
(31:49):
integrations. I don't know. Andthen, hey, we, we already have
that. We, you know, we, we canjust set it up for you and say,
a lot of times, goes back tothat understanding, where does
that fit in the overallbusiness, business process, and
if it's needed, then we're goingto talk with the vendor and make
sure that that can be sharedwith the entire network. And
(32:15):
sometimes we set it up for aspecific office, and we help
that specific office to get thatone vendor that they need.
Because reality is that thereare certain vendors that make
sense when you have a specifictype of office or specific type
of shippers, and there are, soit's like, why other agents
should go through the process ofonboarding that if, if they
(32:38):
don't need it, right? But, yeah,it'll go long story short, it
comes back down to businessprocess, and where does that
fit. And technology should be anenabler of that, not just
something that you bolt ontrying to just, you know, fix it
on itself, a square peg into around hole, almost. Yeah, yeah.
(33:01):
And in my experience, those typeof involvements have not gone
very well. A lot of times it's asolution in in search of a
process, right? So let's try tounderstand the problem first and
see if this is a good solution.
Blythe Brumleve (33:17):
So with all of the you know, remote offices,
the tools we've talked about.
How do you approach thecybersecurity elephant in the
room?
Eze Peralta (33:28):
Well, we are, we're cloud. Our TMS is cloud based.
So one a lot of is delegating,the main, the main operational
system that is the TMS. They'regetting a lot of that security
to to a company like Salesforcethat has, you know, invest
(33:49):
billions of dollars in insecurity and and so we don't
hold, we don't have anything onprem, on on servers, on prem.
Then for the tools that we buildinternally, we are cloud based,
but also use containerizedworkloads that are kind of like
(34:13):
disposable so if like it's weuse, yeah, so it's not that we
have the one big virtual machinewith everything in it. We have
each capability has its owncontainers, and then we have
event driven system thatconnects everything together. So
(34:35):
we try to incorporate thesecurity on on the design
itself, of the solutions that wedo so that's on the aspect of
more like infrastructuresecurity. You know, security
access to your database. Wedon't have one central database
where everything is there. Imean, we do that have a data
warehouse, but the operationalsystems you have, each feature
(34:57):
will have its own little. World,and if that, if that data goes
away, we could reconstruct it,for example, right? So we try to
incorporate all that security bydesign. We use for all of
infrastructure for the technerds out there, we use
(35:20):
infrastructure as code. So inorder to deploy infrastructure
like databases, Virtual Machinesand these kind of things, we are
not just going into a UI ordepending on a person doing
their job right. Everything istemplated and has all the
security measures already inthose templates. So when we
(35:42):
deploy it, for example, ourdatabases don't have access to
Internet. In order to access ourdatabase, you need to go through
some other some other steps, andonly our containers and virtual
machines can access thosedatabases. They are not open,
right? Yeah, and we tried tocontrol our APIs in that way as
(36:07):
well. Where if something doesn'tneed to go over the internet, it
doesn't, it doesn't go we onlyexpose to the internet the
things that like have to beexposed. So
Blythe Brumleve (36:19):
you guys are building fortress over there is
essentially what you say withseveral boats to protect the
fortress.
Eze Peralta (36:26):
Yeah, we have in our team. We have Valentin. He's
very, very strict on that aspectas well, and he knows quite a
bit on that front ofinfrastructure security. And so
we tried to incorporate allthat, you know, from the get go,
it's like we we don't deploysomething if it doesn't have
(36:47):
these characteristics, becauseit will be irresponsible. And
also, in our case, we are agentbased, so data needs to be
siloed, so we cannot expose, youknow, customer data from one
agent to another agent. So allof that is controlled in the
TMS, but also we have our ownseparate user pool and security
(37:09):
settings that make sure that allthe tools that we build on top
of the TMS and the tools that webuild custom also have that
characteristic right
Blythe Brumleve (37:21):
in your experience, or maybe what you've
seen at other, you know, agency,companies or brokerages, how I
guess, involved is the internalIT team when it comes to, like,
freight fraud. Are you guys, youknow, in there, you know, trying
to help combat it from a digitalperspective, like digital
warfare, or is that sort of aseparate team that, you know,
(37:44):
maybe closely works, you know,with the IT team, I'm just
trying to, I guess, maybeunderstand where it kind of
plays a role when it comes tothe just the dramatic increase
in freight fraud across theindustry, I think
Eze Peralta (37:59):
in order to have an effective security strategy. You
need everyone to be involved.
And it needs the business usersto to to be the ones also
pushing the the risk managementinitiatives. In our case, we
have very lucky that everyone inour team are, it's very much
(38:20):
aware our, yeah, our VP ofOperations, VP of Finance, VP,
you know, Director of Careerprocurement, and everyone. It's
very, very much aware of all ofthis. And they come up with the
initiatives to us, then we tellthem what is possible, then we
(38:41):
discuss what ways we could bevulnerable. But the IT team is
very aligned with the rest ofthe business in these
initiatives, and we all know andunderstand, and we try to
emphasize that with that team aswell, that what the scenario is,
what we are fighting against andbut I think it needs to be. It
(39:05):
cannot be just delegated to it,because the business user know a
lot more than they they thinksometimes about security,
because they know how a businessprocess work. They know, for
example, with when you see theseschemes of fake paperwork, or
these carriers buying MC numbersand doing all these maneuvers,
(39:30):
you know, they know the businessuser. Know the intricate
intricacies of that from from,from the business perspective,
and it can, can put in placemeasures to but we need to know
how that works sometimes, and weneed to be, yeah, in
collaboration constantly. Ithink in order to be effective,
you just, you need that. Youneed everyone on board. It's not
(39:52):
going to work if just delegatedto it, because it's just going
to be a very. Partial view, andyou're going to be covering this
side, but then they're going toattack you on the other side.
And a lot of the a lot of thethreats are social engineering.
So there's only so much you cando on the IT side against social
engineering, because if peopleare giving away their passwords
(40:16):
because someone just trickedthem into doing it,
Blythe Brumleve (40:21):
you know, is that sort of, the crazier side
of it that you've seen is justpeople willingly giving up that
information. Yeah, our
Eze Peralta (40:30):
James or CFO always says it don't click the link.
Like, if you get an email with alink that just don't click it,
like, you know, you can replyback. You can call the person.
You can, you know, try to, butdon't, don't click it, because
most of the times when somethinghappens that someone gains
(40:50):
access to an email or somethinglike that, is because someone
clicked the link, or someone gota call, got tricked into going
to a website that was not theactual website. And there's
always something like that, andso, yeah, I think definitely the
most scary part, and the mostdifficult to combat is the
(41:12):
social engineering threats.
Blythe Brumleve (41:15):
Are there any examples that you've seen that
have like, maybe at othercompanies or something, what
they're dealing with that hasthe creativity of a fraud scheme
has almost impressed you.
Eze Peralta (41:31):
What impressed me is when the quality of the
emails that people are craftingand sometimes even how they are
putting malicious code inside ofPDFs, for example, so you're not
even clicking a link. You'regetting, you're getting an email
from your boss saying, Hey, canyou check this report for me?
(41:54):
And and it's looks exactly likethem, and maybe it is coming
actually from there, from theirinbox, because they got access
to that inbox and and they say,Oh, I'm not taking any link,
even if it's it's a file, right?
So you click on the file andthen, and then you're done. So I
think, yeah. Or when at Tia, thekeynote speaker was was talking
(42:21):
about how, with deep fakes, youcan have someone, some AI,
talking to you as if it's yourboss, for example, asking you to
put money on an account, orthings like this. And in this
case, the CEO was talking withthe CFO on a video call. But it
was not the CEO, it was just adeep fake. And they asked them
(42:43):
to put money on an account forfor an acquisition they were
going to do. And they did it andlike, how can you combat that?
Yeah, it's, it's very it's veryhard, right?
Blythe Brumleve (43:00):
Yeah, I think on the podcasting side of
things, I have warned my familythat I'm like, 30 seconds of
audio is all it takes to deepfake me. And so if you get a
call from me at you know, firstof all, I don't call them, I
text them. Probably know thatsomething was up. But if you get
(43:20):
a call for me, we've developed,like, a safe word within the
family that you know, you'regoing to use in the event of
something like that happening,where we would call and, you
know, ask for money or somethinglike that. We have a family safe
word. It kind of sounds likemaybe we need to do that at the
corporate level
Eze Peralta (43:35):
too. Oh, that's definitely, that's multi factor
authentication. Like, When?
When? When people do their multifactor authentication on their
phone asking you it's that it'sa shared secret that only the
people that it's securing thatsecret know, and then it can be
used to to to unlock, right,whatever you need to unlock. And
in this case, the this, thatsafe word is acting as that, you
(43:59):
know, multi factor that are youactually you? It's like, so,
yeah. But first things first, ifyou don't have multi factor
authentication on in yourorganization by default for
absolutely everything, then youshould be looking into doing
that first, because that'sanother thing, is that we're
(44:22):
talking about really socialengineering schemes and
everything. But then people justdon't have their multi factor
codes on, on, on, on theirphones. So you're basically
that, you know, hackers and thebad actors. They try to find the
easiest path for getting it. So,yeah, just start there that that
(44:45):
was that solves like 90% plus ofthe of the phishing attacks. Is
Blythe Brumleve (44:53):
there any freight tech that doesn't exist
that you wish
Eze Peralta (44:58):
existed? Yeah?
Yeah, wow. I I think maybe allthe integration landscape in the
in the freight tech world isit's kind of all over the place.
So I think some sort ofstandardization of what we all
(45:20):
think a load is, what, what,what kind of events are relevant
to, to a load or so, I thinksome sort of, I don't know if
it's a software itself, somesort of agreement or
standardization on how we'regoing to communicate these
(45:41):
systems, between systems. EDIwas an attempt to to achieve
that, and I think it is stillbeing used big in part, because
of that, because it was somehowsomewhat successful in in you
know this creating a basestandard that then everyone need
(46:02):
to derail from that standard,and ended up doing like each
area its only world, but withAPIs, for example, it's
happening that every API isdifferent, and you need to, you
have all these point To pointintegrations, and there's no,
yeah, there's no standard. And Ithink there's, it generates a
(46:23):
lot of waste for a lot oforganizations, having to
maintain software that is just apipe to put data from A to B,
right? So that, I would say,yeah, a lot of the time, even
for tech vendor, even forvendors, it's like, it's not
valuable work. It's justsomething that you need to do in
order to get data. And then youstart, you know, adding value
(46:46):
with your product.
Blythe Brumleve (46:51):
Now, last few questions here for a potential
agent that's thinking aboutmaking the switch, is there
anything that they need to do onthe tech side of things, on
their end of things, to betterprepare them for, for making the
jump to, you know, maybe,hopefully, SBI, but maybe
another company. Is there any, Iguess, maybe standardization of
(47:16):
of what you would recommend foran agent to get ready to make a
switch.
Eze Peralta (47:21):
I think, I think asking the questions to to the
to the agent network that thatthey're going to they're
evaluating about what's theirtech stack. But no, no, no, not
as as listing the vendors thatthey use, because that you could
have a really good list ofvendors, but not have them
(47:42):
working together. Well, right?
So it's okay, can, can thisnetwork give me all fulfill all
the use cases that I need tofulfill? For example, we offer,
we push data to shippers.
Sometimes they needed as a CSVfile with a very specific
(48:03):
format. Sometimes they neededthe API. Sometimes they need to
be Adi. Sometimes they need itas they come. Sometimes they
need it in a schedule, right? Soif you have shippers that are
requiring this type oftechnology, then you need to ask
very clearly at the beginning,hey, can you push data to my
shippers in the way they needit. How long it takes for you to
(48:23):
complete the new integration ifa new shipper needs it? For
example, can you automaticallyquote, because I have a shipper
that asked me to respond to aquote within a minute? Can you
do that like and same for you?
Know what carrier fraud, carriervetting and compliance is like,
asking the questions, you know,but trying to get to the proper
(48:46):
depth of the question. It'slike, Oh, do you use one
question could be Oh, do you usehighway? Or do you use RMS, or
do you use my care packets?
That's one way of askinganother. Another way of asking
is like, how do you handle witha carrier trying to do this? How
would you prevent this fromhappening, or how do you use
(49:10):
those tools in order to protectme or protect my business? And I
think, trying to find the depthin the questions beyond, oh, we
use this vendor, okay, but howyou use it? Why you use it? Why
that one or not the other one?
Like trying to get a bit moredeep into into the into the ask,
into the asking.
Blythe Brumleve (49:33):
Is there any?
Well, actually, one more quickquestion that just popped in my
head. AI agents, not necessarilylike the call ones, but the the
ones that are promising, youknow, to fix, you know, be your
internal marketing department,or be your internal sales
department. It Do you see therise of sort of a AGI agents
that you know, you have a littleyou know, minions that are doing
(49:55):
all of your work for you? Yeah.
Eze Peralta (50:01):
So to be honest, like from, from what I've been
seeing so far, I think we're farfrom it still, whenever I try
to, I mean, I use coding toolthat has some sort of agent in
it and and it also has, like anauto complete feature. I end up
(50:26):
using the auto complete morethan the agent, because a lot of
times the agent is just going togo in a loop of trying to solve
a problem, but in a very kindof, like, naive way, and maybe
it's not understood for certaintasks. Yeah, I can see it
working for certain tasks, but Idon't see it happening at the
(50:49):
level that is being advertised.
Of, oh, now you're going to havethese digital workers that is
going to just do all of yourcalls for you, and it's going to
give you all, you know, allthese benefits, and I have yet
to see something that shows methat that's possible and that's
and also, one aspect that we, Iwas talking with my team
yesterday, is what you get, youknow, when you're in when you're
(51:13):
doing something by yourself,like, if I'm trying to solve a
problem and I don't know how toDo it, then maybe I Google,
maybe I use a AI or whatever,but I'm still using my brain to
solve the problem. Once I solveit, I learn something right. And
same with human interaction,maybe some employee making a
(51:34):
mistake also reveals that thereis a process that needs to be
improved, right? So then you canso I think error and and it's
part of life, and it's also partof learning. And I think we are
entering to this notion of we,we're going to have these agents
(51:58):
doing all the work for you, butthen after you see the work
done, have you learned somethingfrom that process? So I see one
of the possibilities is that ourlearning curve, it goes up, up,
up, up, but then now we are justdelegating all this resolution
to these automations, but wedon't know how they work,
(52:21):
because nobody knows what an AIagent is thinking, and nobody is
even if we could, no one isgoing back and say, Oh, how did
you reach this conclusion? Wasit accurate? Was it not what can
I learn from the process ofreaching the conclusion? So I am
a believer that that process ofreaching a conclusion is really
(52:44):
valuable, and learning fromthese processes, learning from
mistakes, is valuable. And Ithink sometimes the promotion of
the type of tools forget thatpart, and they're just too
focused on the solution. Insteadof trying to understand how we
get to a solution, how theproblem works, why the problem
(53:07):
even exists, could we definethis problem out of scope so we
don't even have to solve itbecause it doesn't exist
anymore? You know, trying tofind more depth, I think, I
think it comes down to depth. Ithink a lot of these tools are
being promoted with not enoughdepth in mind, just very
tactically, and we're notthinking about, you know, maybe
(53:31):
you needed to, maybe you neededto walk the path. Make a little
mistake, that is, as long as itdoesn't destroy you, right? You
needed to walk the path. Makethat mistake, learn from it,
improve and keep going and andyou can still do that while
(53:53):
you're using AI agent, but ifyou just delegate everything to
to and then you don't thinkabout it anymore, then we're
becoming kind of like thinkinglazy, right? Like we're becoming
lazy at thinking and Ipersonally don't like that. I
Blythe Brumleve (54:11):
completely agree, and I echo that
statement, because I on one sideof the coin, I will say that,
you know these even like Groxdeep research tool. I love it,
but I use that inside of chatGPT in order to help me come up
with an interview flow. Andit'll I did it for this
(54:31):
interview. For example, I tookour previous episode, I took the
transcript. I said, Give me alandscape of, you know, the
current freight trends andfreight tech trends and
technology and things that areon the market, analyze the
transcript and then come up withan interview flow based on that.
And I would say probably 60% ofthe questions were pretty good.
(54:51):
But if I didn't do the activelistening to our previous
episode and also to otherepisodes you've been on. I
wouldn't have been able to, Ithink, craft a better interview
for the sake of thisconversation. And so it's it. I
don't know that that's somethingthat you can replace. Is that
(55:14):
active learning by doing it youhave to still do the thing. And,
you know, maybe there's someways where you can automate the
boring stuff, but there's still,you know, a creative aspect that
you create, a problem solvingand learning that I think is
still needs to be prioritized.
Yeah,
Eze Peralta (55:31):
and in this case, what the process you're
describing you were driving theprocess you were not delegating
to an agent that was driving theprocess, an AI agent that will
drive it. But you were drivingthe process. You were using a
tool to categorize to give you,you know, some help, but you
were in the in the driving law,and I think that's important and
(55:53):
and also it's like, it's, it'sfun to be a human, and it's fun
to connect with humans. Big fanof humans. So
Blythe Brumleve (56:01):
don't tell the AI that it
Eze Peralta (56:05):
already knows.
Probably that's a topic foranother, another podcast. But
like, then there is that aspectof like, do you want to be
working with an AI agent? Or youwant to, you know, are we
working for? What are we workingfor humans? Are working for
ourselves, for human species,for that's more philosophical,
maybe, but that comes into playtoo. When we're talking about,
(56:29):
you know, replacing people, it'slike,
Blythe Brumleve (56:35):
I don't know, I want to do more of the things
that I like doing and less ofthat. And I think that that's,
you know, sort of the the commoncomplaint I hear with AI and
automation is like it's takingaway, or some people feel as if
it's taking away from the thingsthat we want to do more of
instead of the things that wedon't want to do more of, which
is, like, I don't know, cleanthe bathroom, or, you know, fold
(56:56):
your clothes. Like nobody wantsto do that. But, you know, AI is
not fixing that yet, and that'swhat we want it to fix. We don't
want it to take away the thingsyou know, that do make us human,
creative problem solving, youknow, creative indent
adventures, things like that,talking to people, developing
relationships with people. Ithink all of those things are
really important. You kind ofhit the nail on the head for the
(57:20):
rest of you know, I guess say2025, and beyond. Are there any,
you know, new tech solutionsthat you guys are working on, or
integrations that you couldshare with us, or or things
you're thinking about that youthink the audience should know
Eze Peralta (57:35):
we're working continue improving on, on the
capacity, capacity tools that wehave. We know that the you know,
market still somehow lose onthat sense. But we know, you
know, being prepared for, forfor different market conditions
as as they show up. So addingmore sources of capacity to our
(57:57):
capacity hub tool continueadding on, you know, more data
into our risk management tools.
Also we're adding these morefeatures on the AI email, you
know, creating loads from emailsand and these categorizing email
requests, working also onautomation of of bidding to the
(58:21):
shipper side so and, and also alot of work on on the back
office, because we, we need to,we need to provide, you know,
best service in the office,because our agents, we work a
(58:42):
bit differently than maybe otherother agent models, where agents
just do the operation, theycomplete the load, and then we
take over the entire AR and APcycle, right? So in order for
that to be effective, we wouldneed to have a high level of
automation, high level ofefficiency on the back office.
(59:05):
So a lot of work on 2025 is alsogoing to go into continuing the
already, already highlyautomated back office is going
to continue being more automatedand providing our staff with
tools to, yeah, maybe to reducesome of their data entry and so
(59:25):
they can focus on being moreanalytical, on on the audit and
these things so, so yeah, that'sthose are things we're working
on. A
Blythe Brumleve (59:39):
lot of work you guys are got ahead of you for
this year and beyond, but I'msure you know a lot of the
agents are extremely thankfulfor the investments that you
guys continue to make into theplatform and the program itself
as a where can folks find you,follow more of your work, maybe
connect with you at a futureconference.
Eze Peralta (59:58):
Yeah, so my LinkedIn. I am sa Peralta. You
can find me there. My email is EPeralta at SPI three, bpl.com
you can email me and I'll try toreply as soon as you can. Yes,
no, I don't use email bots, soif I take time, it's because I'm
actually responding. And if yousee me at any conference,
(01:00:22):
generally, I go to Tia andcapital ideas and technovations
and maybe a few other more. Idon't have anything planned as
of now. For for conferences, wehave our annual agent conference
this week. Yeah, exactly. It'sgonna be fun. I be fun. But
(01:00:44):
yeah, if you see me at anyconference, just Yeah, would
love to connect. Well,
Blythe Brumleve (01:00:49):
perfect. This was a great discussion.
Hopefully people you know, likethis one just as much, hopefully
more than than the previousepisode that we did together.
But I'll be sure to add allthose links into the show notes.
Just make it easy for folks, butas a this is great. Thank you so
much for your perspectiveinsight. Thank
Eze Peralta (01:01:05):
you very much. Need for having me. Absolutely
Blythe Brumleve (01:01:12):
thanks for tuning in to another episode of
everything this logistics, wherewe talk all things supply chain
for the thinkers in freight, ifyou like this episode, there's
plenty more where that camefrom. Be sure to follow or
subscribe on your favoritepodcast app so you never miss a
conversation. The show is alsoavailable in video format over
on YouTube, just by searchingeverything as logistics. And if
(01:01:32):
you're working in freightlogistics or supply chain
marketing, check out my company,digital dispatch, we help you
build smarter websites andmarketing systems that actually
drive results, not just vanitymetrics. Additionally, if you're
trying to find the right freighttech tools or partners without
getting buried in buzz words,head on over to cargo rex.io
where we're building the largestdatabase of logistics services
(01:01:55):
and solutions, all the links youneed are in the show notes. I'll
catch you in the Next episode inGo, Jags, you,
When you're doing something by yourself, if I'm
trying to solve a problem and Idon't know how to do it, then
maybe I Google, maybe I use AIor whatever, but I'm still using
my brain to solve the problem.
Once I solve it, I learnedsomething, right? Same with
human interaction, maybe someemployee making a mistake also
(00:27):
reveals that there is a processthat needs to be improved,
right? So then you can so Ithink error, and it's part of
life, and it's also part oflearning. And I think we are
entering to this notion of we,we have these agents doing all
(00:49):
the work for you, but then afteryou see the work done, have you
learned something from thatprocess?
Blythe Brumleve (01:01):
The freight tech hype cycle is real, but
behind the buzzwords are folksactually building the plumbing
that keeps your freight offsrunning as a Peralta is one of
them, and he is back with SPIlogistics, VP of technology on
the everything is logisticspodcast, in order to break it
all down for us, I'm your host,Blake Milligan, and of course,
(01:21):
we were presented by SPIlogistics, and I was just
telling Eze his last appearanceon the podcast a couple years
ago, I believe is our mostpopular SPI logistics episode.
So sorry to everyone you knowwithin SPI that's come on the
show, but we will love Eze. We
Eze Peralta (01:40):
We are good friends. So I so it's all good.
I think.
Blythe Brumleve (01:44):
no competition.
So let's, let's set the stage.
So for SPI, for folks who maynot be familiar with your role
within SPI, can you kind of giveus a high level overview of what
you do on maybe a day to daybasis, a weekly basis. What does
I guess a week in the life of Zalook like?
Eze Peralta (02:06):
Well, there's we have two main areas. So one is
the things that we are buildinginternally. The other area is
the things that we need tomaintain or integrate or are in
relation to tech vendors. And sothere's a bit of swing between
(02:28):
between those two aspects. So alot of the time is meetings,
meetings with vendors trying toalign the road maps of what
we're trying to do with whatthey're trying to do, trying to
keep up with all the the newfeatures that the new vendors
(02:48):
that come up. It's it's a lot oftime, because evaluating tools
and evaluating next steps andtrying to sometimes might
maintain the roadmap, but at thesame time being flexible enough
to pivot. So a lot of a lot ofthe time is, is that kind of
more strategic point of view?
And then there's a lot of timethat is just coding or, you
(03:11):
know, jumping into the theactual implementation, a lot of
modeling and diagramming withthe with the business users as
well. A great deal of the ofsoftware development. A lot of
people focus on on the codingside with software development,
but in my experience, softwaredevelopment is a lot about
(03:35):
modeling and understanding whatyou need to build before
starting to build. Right? It'slike, for example, you're going
to build a house. You're notgoing to start, you know,
cutting wood and laying bricksright away. You need some sort
of blueprint. You need some sortof understanding of what needs
(03:56):
to be built. You need tounderstand, even, you know the
terrain where you're going towork with and, and, and a lot of
other factors. So a lot of thetime goes into that as well. And
Blythe Brumleve (04:09):
so for folks who may not be aware, you know,
SPI Logistics is a freightagency, and so you don't have
any some freight agencies willhave an in house brokerage team
or an in house trucking team.
SPI doesn't do that because theytheir internal team is set up
and dedicated just to supporttheir agents. And so you guys
are, I believe and correct me ifI'm wrong, but you guys have a
(04:31):
general tech stack that you use,but then you also accommodate
the freight agents, and maybe ifthey have, like, a particular
tech tool that they like to use,you'll incorporate that into
sort of, I guess, the mothershipof SPI Is that accurate?
Eze Peralta (04:48):
Yes, yes, correct.
So we try to embrace also thatdiversity in our agent agents
network. And you're correct.
Yes, we you. Where we spendquite, quite, quite a bit of
effort on understanding thoseneeds and also offering the
(05:08):
tools that we have to enhancethe existing operations when,
when agents on board with us. Sobecause maybe they already have
certain workflows that that,that they already applying, and
that's effective for them. So wedon't force anybody to, okay,
this is the way how you shoulddo it, but we do offer the tech
(05:29):
stack we have, but also open toincorporate and integrate other
technologies that agents mightbring when, when they on board
and, yeah, trying, trying toaugment and enhance their
operations and and then thisgoes in line we're saying
before, of understanding what isneeded before going into the how
(05:51):
we're going to do it right. Oncewe know what is needed and what
is the most helpful for ouragents, then we can decide,
okay, you know, how we implementit sometimes, that is getting a
new vendor. Sometimes is, youknow, building some
customization on an existingtool. It could take any shape,
(06:12):
basically.
Blythe Brumleve (06:14):
Are there any sets of tools that your agents
are using every day?
Eze Peralta (06:21):
Well, the TMS is what our agents use every day.
And I think the operations wehave the track and trace
capabilities that that sometimessome agents would go directly
into the tracking portalsbecause they have specific
needs, or some other agents willuse them. Use these tracking
(06:44):
capabilities within the TMS.
Similarly with posting,sometimes people would use, you
know, posting directly to theboards. Sometimes people
wouldn't do it through the TMS,but we try to have a layer of
control of all that data andwhere it goes, where it's coming
from. So we can also offer someinsights and analytics on on
that data as well. And then wehave, yeah, for example, tools
(07:08):
for procurement, capacity orrisk management that we created
internally. And a lot of thetimes, what we do is try to
aggregate that very vastlandscape of data into more
actionable information. So youknow, instead of our agent
having to go into five differentportals to make sure that their
(07:28):
carrier is not fraudulent, thenwe kind of like consolidate all
that data. We say, hey, yeah,this is, this is a good to go
carrier, or this is flagged forthese reasons, yeah, trying to
also combat that tiredness thatcomes sometimes people drop the
(07:48):
ball on certain things relatedto risk, because it's just too
cumbersome going into differentplaces to having to check in 10
different portals To make surethat you know your your provider
is good to go. So people stopdoing it right, people, people
stop checking because it's justcumbersome. So we try to also
(08:09):
think about that, and, like,make sure that we make it easy.
So, so those tools are, areused. Oh, that's super
interesting.
Blythe Brumleve (08:17):
So you essentially are, you know, using
different maybe, and I'm justspitballing here, but like a
highway and a carrier, sure. Andyou know, some of these, these
vendors that do are in the samerealm, but do things a little
bit differently, but they allhelp give a fuller picture. And
so you're able to build thatinto a dashboard for your
(08:38):
agents,
Eze Peralta (08:39):
yeah. And sometimes, even aggregating it.
We have tools where the resultis, you know, you're good,
you're not good and up to thatpoint. And then if you want to
see more details, you can gointo the portals, or you can
expand on those details. Butyes, we do that, like with all
the compliance vendors we have,because sometimes, you know,
maybe highway has some insightsor flags about specific things,
(09:03):
like, for example, where thesepeople, these carriers, are
accessing the system from. Butthen maybe freight guard, they
have very you know, if peoplewho are reporting these carriers
but right care for one, forexample, so you need both right?
Because one will give you, youknow, certain insights. The
(09:24):
other one is going to give you,I don't know, highway for
example, give you insuranceinformation, inspection
information, but if someone gotburned by by a bad actor, the
first thing they going to do is,most likely, is reporting it to
401, so we also want to, youknow, action on those, those
reports as well.
Blythe Brumleve (09:45):
So one of, and I heard you talking about this
with Chris jolly on the freightfreight coach podcast that you
guys were at technovations,Tia's, technovations in in late
2024 I'm sorry I missed that. Itwas actually it. Jacksonville,
Florida, but I was gettingmarried that week, so I
couldn't, I couldn't sneak awayto go to that to go to that
(10:06):
conference, but I heard that AIwas just everywhere, as it is
with most you know, freightconferences right now, what's
what's your? I hear you laughingthere. What's your, I guess,
overall take on, is it hype? Isit a buzzword? Is it? Yeah, is
it valuable? Is it kind of allof the above?
Eze Peralta (10:26):
I think, I think it's all of the above. And I
think AI is a very vast field,so it's not the same. Saying,
You know what? We are using thehype, this current hype wave of
AI is most related to largelanguage models, which is, you
(10:47):
know, the chatgpt. Butartificial intelligence is
something that existed for awhile. Now, the computing has
gone, you know, it's moreavailable and for companies to
use these large language modelsand the training techniques have
improved, but before we hadmachine learning or other types
(11:08):
of of artificial intelligenceapplied to analytics or
different aspects, likeforecasting and these kind of
things. So I do find it veryvaluable for a lot of use cases.
We use it internally a lot inour like our dev team uses this.
(11:28):
Use it quite a bit. But also wewe are cautious to not try to
chase AI or implementation orany tool like AI or any other
tool, just just for, justbecause everyone else is doing
it, or just because, you know,it appears to be the solution,
(11:52):
magic solution, to all these usecases we we are of the idea that
alignment between business andtechnology is key. You know,
integration and governance ofall the integrations is key.
Governance of data is key. Sothose things what we focus on,
(12:14):
and if there is, you know, if AIcan help us to achieve that, to
achieve the business outcomesthat we want to achieve, and
improve the KPI we want toachieve. Then we evaluate it as
we evaluate any other tool, andthen we we apply it, right? But
we are, you know, we try to keepfocus on what is important for
(12:38):
our agents, what is importantfor for the business as an
organization. And, yeah, we douse AI and in many ways, and we
are adding more use cases to it.
But at the same time, I, I thinkthere is this notion that is
going to come and solveeverything for everyone, and
(12:59):
suddenly it's going to replaceworkers and all this thing and
like, that's that's notsomething that, I mean, we don't
see it that way as much. Yeah,you can, you can automate phone
calls and these things. But whatabout the when things go wrong,
nobody wants to talk with amachine. And most of the
(13:23):
logistic business is managingrisk right. And are you going to
let a machine to manage thatrisk for you? Are you going to
let the machine do thatnegotiation or do that, you
know, relationship building foryou. So it's yeah, if, if you
have a mature enough system,then you can connect with your
(13:53):
shipper via API, and then youdon't need to have like a voice
agent, you know, calling ashipper or but at the same time,
there are certain, there arecertain use cases that, yes, you
know, for support, we had theseautoresponders for phone for
years, where you say, hey, youknow, give me your load number.
(14:13):
You just type it in the phone.
And so it's not very No, it'snot that different than that.
And then, and then you have alsothe ethics aspect, which is, I
was reading a LinkedIn post. Idon't remember who it was from,
but it was something in the inthe lines of, I was talking with
(14:34):
an AI rep, and I asked if it wasan AI. They said no, but then
they found out that it actuallywas an AI, but it seems that
this AI was trained to lie,right, to say that it was not an
AI. So I think, you know, weshould keep being honest. So
that's non negotiable, right?
Blythe Brumleve (14:56):
No, that's a very good point, and I think
that that's something. That's aconcern that we've raised in the
past and in previous episodes onwhen do you disclose that you're
talking to an AI agent? And Ithink most people are fine
talking to an AI agent when theyneed an answer quickly about
something, where is my load?
Something like that, but for adeeper maybe customer
(15:20):
relationship. There is a levelof you should probably disclose
you know that that informationas quickly as possible, so then
that way the person can makethat preference on if they want
to talk to somebody real or talkto a virtual agent. I would be
curious to know what other maybeuse cases are you seeing? You
(15:41):
know, maybe two different usecases. One where, you know, it's
machine learning that's actuallydoing the job and you're using
that. But then, on the flipside, a large language model.
I'm curious about the use casesfor each of those that you're
seeing.
Eze Peralta (15:58):
So machine learning is mostly used on, you know,
analytics and forecasting orfinding anomalies or things like
this. So you have a you know,data from the last, let's say,
10 years, and then you try tofind, you know, seasonality. You
try to find, like, certainpatterns that would be more
(16:18):
difficult to find just manually,you're trying to dig into all
that data, so and so for thatis, is, that's a good, good use
case. And you can also train itto find trends, you know, oh,
we're, you know, when these,when these metrics are, how
(16:38):
metrics relate, oh, when thesetwo metrics are going down, it
means that your business istrending up, or, I don't know,
or you start finding thoserelationships. And so for
analyzing those huge amounts ofof data, it's, it's machine
learning, it's, it's, it'svaluable. And for large language
(16:59):
models, I think the importantpart to understand is that no
matter what the what the vendorssay, they don't reason right.
Reasoning is a different thing.
They just predict the nexttoken, the next the next symbol
that's this should be thereaccording to their calculation
or their training. So what Ifind really interesting for as a
use case is categorization andthen interpreting natural
(17:25):
language text, for example,emails, tagging an email saying
you can plug it into to yourinbox and say, Okay, I have 20
call requests. So in a day Ihave 20 quote request. I have 10
track and trace requests, orwhere's my load kind of
(17:45):
questions. Then I have 15carrier carriers asking for
posting. And then I have threecomplaints and and then you can,
you know, categorize all ofthat, and eventually you're
going to have enough data tounderstand your your own inbox,
for example, much better, right?
(18:09):
And then from there, you canautomate, oh, when, when these
type of request comes in, orthis type of email comes in,
then you can auto generate aresponse, as long as it fits
with the with that category,right? So categorizing, I would
say, understanding topics orsubject on on text, and then
(18:33):
creating some sort of simple,simple responses or template or
responses for certain type ofinquiries. Those are, I think,
really good use cases. And
Blythe Brumleve (18:47):
then I with, as you were talking, I was thinking
about, you know, is there any, Iguess, safety issues that go on
with, you know, using a largelanguage model that, you know,
maybe agents should be a littlecautious of, of putting, you
know, personal information maybeinto some of these systems. Or
is it that concern maybe largelyoverblown? No,
Eze Peralta (19:10):
I think, I think that it's a it all depends on,
on the architecture that youhave behind it, right? So, for
example, we were talking withour team about cases where you
see people doing this that theycall now vive coding, which is
like,
Unknown (19:27):
I hate that phrase.
Eze Peralta (19:30):
It is, what it is.
And yeah, and then people wereable to hack applications built
in this way in 30 minutes, justbecause the credentials were all
pasted over in the code. So youcould just go into the
repository, public repository,and see all the credentials for,
for, for different APIs andthings. Because, yeah, if, if
(19:56):
you. Know what you're doing, youwouldn't do that, right? So if
you don't know what you'redoing, then you maybe think it's
fine. You trust that the AI willdo it okay for you, and then you
just ask the AI, oh, make itsecure. That just doesn't work.
You can't just tell the Oh, makemake this application for me.
Make it secure. I wish it waslike that, but, you know,
(20:21):
there's so much more to it. So Ithink security is one of the
big, big topics to think aboutwhen thinking about AI
Blythe Brumleve (20:32):
for sure, what about from the email side of
things? Because I personallystruggle with sorting through my
email on a daily basis, and Ican't imagine what you know a
typical freight agent is goingthrough with their email inbox.
What kind of tools are youseeing that's helping to
categorize those emails? Is itall strictly done through the
(20:55):
TMS, or is it, you know, a bunchof different tools that that can
help, you know, categorize thoseemails? Maybe, you know, auto
respond to them. You know, getrid of the junk. Is it a bunch
of tools, or is it one or tworeally good tools?
Eze Peralta (21:08):
So, I mean, there are a lot of tools and probably
missing a lot here, but so arethe PMs that we use ravinova,
they are adding now featuresthat natively allow you to to
use AI and then connect youremail to it and, and they will
(21:29):
It will create loads for you.
You will create truck postingsfor you, and, and, and also,
with all the automation thatexists already available in
Salesforce, you can create theseresponses back and connected to
all the the the other existingwork. So, so for us, that's
that's a really good one, andlevity, I think, is doing a lot
of a lot of work on that aswell. I I personally like the
(21:51):
what they have there. And thenwhat I'm seeing also is that
even the existing client emailclients are starting to add all
these features internally, like,if you have, for example,
Outlook, you could grab, youknow, power automate, which is
their little automation tool,and you could have little
(22:13):
workflows that categorize emailsand use the AI Within the
Microsoft environment. And thatshould be at least, you know, if
you already trusting outlook,then you know you can keep your
your existing data alreadythere, and it will not leave
that, leave that environment andthat that might be interesting.
Gmail, I think it's also addingall of that. And I think
(22:38):
eventually all the gmss aregoing to start doing that. I
know that other GMs for brokersas well, have already that. I
know that. Yeah, tools like praythe also have some email
automation going on. I haven'ttested it freely, but yeah, I
(22:58):
think all of those tools shoulddo a good job on, on, on, you
know, categorizing emails. It'snot at the same time I've seen
more and more orchestrationtools, or the tools that allow
you, like, to create your ownautomations. And a lot of people
are starting to build their ownautomations around email. I was
(23:21):
talking with Sarah from otherlogistics, and she was telling
me about some AI tools and thatthey're using for some use
cases. And I think it's calledsola solar, solar AI. And so any
tool should work again. It allcomes down. I want to bring it
(23:42):
back to have you modeled yourbusiness process. Do you know
what your business process is?
Do you know how you're going tomeasure if your process is
working or not? Have you Do youknow what's the outcome that
you're trying to achieve?
Because if you don't know that,then you're going to go into it.
How the implementation withoutknowing the what, and then you
(24:02):
know it might fail, and thenyou're going to think that the
problem is the tool, that theproblem is, oh, this AI is not
working, or this doesn't work,but, but how? How are you even
measuring if it's working ornot? Because you didn't know
what your performance wasbefore, you don't know the
metric that you need to look atto know what your if your
(24:23):
performance improved. So thenit's like, it's all going to be
gut feeling. And of course, thegut feeling at the beginning of
implementing a new technology,the gut feeling is going to be
frustration, most likely. But ifyou have the metrics in place
and the business processesmodeled properly, then you're
going to realize that, oh, well,I'm frustrated because I don't
(24:47):
know how to use this tool, butobjectively, these metrics are
improving, right? So I likealways taking it back to that,
like the business process. Needsto be solid and needs to be
documented, needs to be refined.
And it means because, if not,you end up with the vendor
(25:11):
deciding for you how you shouldbe working. And that's, I think,
advice for anyone evaluatingtech. You know, tech vendors
will generally tell you thatthey can do everything. You're
going to ask them, Oh, can youdo this? They're going to say
yes, because generally, you'retalking with a project manager
or a sales a salesrepresentative. And then they're
(25:34):
going to go to the developers,like teams, like my team, and
they're going to say, Hey, Isold this. I sold this new
feature. Now you need to buildit, and then when you roll it
out, then you're like, well, butlike, how does this match with
my existing workflow? What? Andthere's a lot of friction. So if
(25:55):
you don't know exactly whatyou're looking for in terms of
business processes, it's goingto be harder for a vendor to to
be effective, right? And it's,sometimes it's not even a
problem of the vendor, because,you know, these are tools, and
if you don't know what to dowith the tool, then also it's,
(26:18):
you know, it needs to be on bothsides, a commitment on both
sides.
Blythe Brumleve (26:21):
Yeah, no one wants to do the the boring and
really underrated, challengingpart is documenting your
processes, because you have tobe prepared to rip them all out
if they're not working. But justsimply documenting them is, I
find it personal experience,just very like draining on my
brain, but once I get a dud,then I'm so relieved, because
(26:45):
then I can figure out, okay,what do I need to do? What can I
outsource? What can a tool do?
And I'm curious as as to how youhow do you approach helping
agents document their process?
Eze Peralta (27:01):
So something that we do is visiting agents,
because being on site with withthe people who are executing the
tasks and executing theprocesses, I'm able to
understand much better all thenuances, right? And also, you
(27:21):
know, I don't like this idea ofthe tech team being, like,
isolated, and then, you know,requirements come in, and
people, you know, the dev teamdoesn't know what the process
is, and we try to all of ourteam is involved in, in these
modeling sessions. And wegenerally we try, because we
(27:46):
know also that agents are verybusy, and we want to also take
out too much time or effort onthat. But when we do this visit,
we try to document as much as wecan about what we see, and then
we come up with a plan of, okay,this is this is the process that
we see. These are the points,the critical points that we
(28:06):
could maybe improve. We havethese tools available to help
you on this particular point andasking a lot of questions. I
think listening and asking it'skey, because, again, if we in a
way, the tech team is like avendor for the agents, right, in
(28:29):
a way, like it's an internalvendor, but we need to provide
something that makes sense, andin order to be able to do that,
we need to understand what theproblems are. And so yeah, a lot
of Yeah, listening and going onvisits, because on a zoom call,
you can get an idea, butsometimes you just need to be
(28:50):
there to really to hear thephone ringing and, oh, how many
you know, how much of the timeis being spent on answering this
call? Oh, what if we have a toolhere that can filter out these
calls and like, you know, tellyou which, which call you should
take first. And how do you, youknow, how does that impact your
(29:13):
overall time to cover a low forexample? Just, just to give you
one example, something that wedid was also one of our agents
was having a lot of inquiriesfor for posted loads from from
boards. And what we were theywere copying and pasting a
template to respond to theseinquiries. We're like, Oh, what
(29:36):
if we try to automate that? Sowe just set it said, you just
set up all the your pricing inthe TMS, and then we can just
find the load that you're askingfor and then give them back that
information. And if it'scovered, instead of telling
them, oh, it's covered, we findsimilar loads in similar lanes,
so we think, Oh, that load iscovered, but we have these other
(29:56):
loads, right? But we. Needed togo visit the agent to really
understand that, right?
Blythe Brumleve (30:06):
How do you approach, you know, when you're
you're thinking about, you know,onboarding a new agent, or maybe
it's a current agent that is, islooking at a new tool. How do
you decide what to say yes toand what to maybe caution them
on?
Eze Peralta (30:24):
Yeah, I don't want to be super repetitive, but I
bring it back to what is the usecase that you're trying to
resolve? What is the businessprocess that this fits in?
Because sometimes you realizethat by moving a few pieces in a
different part of the board,right then you are also, you
(30:46):
know, changing the tension onthe part that you're interested
in. So sometimes it's okay,let's try to quickly understand
at least the skeleton of theprocess that you're doing, and
understand if, if it's a tool,what you need, or maybe it's a
(31:07):
shift on some other structure,right? That might be, maybe it's
a tool that we already have andthat they don't know that we
have. Maybe this a trainingissue. Maybe it's a lot of times
comes down to training. It comesdown to knowing the the tools
(31:28):
that we have available, becausethere are so many that sometimes
it's like, well, if you're anagent who is doing truckload
only, maybe you don't even knowthat we have an LTL program, or
that we have a customer portalthat so they maybe are looking
for a vendor that might givethem a portal for LTL
(31:49):
integrations. I don't know. Andthen, hey, we, we already have
that. We, you know, we, we canjust set it up for you and say,
a lot of times, goes back tothat understanding, where does
that fit in the overallbusiness, business process, and
if it's needed, then we're goingto talk with the vendor and make
sure that that can be sharedwith the entire network. And
(32:15):
sometimes we set it up for aspecific office, and we help
that specific office to get thatone vendor that they need.
Because reality is that thereare certain vendors that make
sense when you have a specifictype of office or specific type
of shippers, and there are, soit's like, why other agents
should go through the process ofonboarding that if, if they
(32:38):
don't need it, right? But, yeah,it'll go long story short, it
comes back down to businessprocess, and where does that
fit. And technology should be anenabler of that, not just
something that you bolt ontrying to just, you know, fix it
on itself, a square peg into around hole, almost. Yeah, yeah.
(33:01):
And in my experience, those typeof involvements have not gone
very well. A lot of times it's asolution in in search of a
process, right? So let's try tounderstand the problem first and
see if this is a good solution.
Blythe Brumleve (33:17):
So with all of the you know, remote offices,
the tools we've talked about.
How do you approach thecybersecurity elephant in the
room?
Eze Peralta (33:28):
Well, we are, we're cloud. Our TMS is cloud based.
So one a lot of is delegating,the main, the main operational
system that is the TMS. They'regetting a lot of that security
to to a company like Salesforcethat has, you know, invest
(33:49):
billions of dollars in insecurity and and so we don't
hold, we don't have anything onprem, on on servers, on prem.
Then for the tools that we buildinternally, we are cloud based,
but also use containerizedworkloads that are kind of like
(34:13):
disposable so if like it's weuse, yeah, so it's not that we
have the one big virtual machinewith everything in it. We have
each capability has its owncontainers, and then we have
event driven system thatconnects everything together. So
(34:35):
we try to incorporate thesecurity on on the design
itself, of the solutions that wedo so that's on the aspect of
more like infrastructuresecurity. You know, security
access to your database. Wedon't have one central database
where everything is there. Imean, we do that have a data
warehouse, but the operationalsystems you have, each feature
(34:57):
will have its own little. World,and if that, if that data goes
away, we could reconstruct it,for example, right? So we try to
incorporate all that security bydesign. We use for all of
infrastructure for the technerds out there, we use
(35:20):
infrastructure as code. So inorder to deploy infrastructure
like databases, Virtual Machinesand these kind of things, we are
not just going into a UI ordepending on a person doing
their job right. Everything istemplated and has all the
security measures already inthose templates. So when we
(35:42):
deploy it, for example, ourdatabases don't have access to
Internet. In order to access ourdatabase, you need to go through
some other some other steps, andonly our containers and virtual
machines can access thosedatabases. They are not open,
right? Yeah, and we tried tocontrol our APIs in that way as
(36:07):
well. Where if something doesn'tneed to go over the internet, it
doesn't, it doesn't go we onlyexpose to the internet the
things that like have to beexposed. So
Blythe Brumleve (36:19):
you guys are building fortress over there is
essentially what you say withseveral boats to protect the
fortress.
Eze Peralta (36:26):
Yeah, we have in our team. We have Valentin. He's
very, very strict on that aspectas well, and he knows quite a
bit on that front ofinfrastructure security. And so
we tried to incorporate allthat, you know, from the get go,
it's like we we don't deploysomething if it doesn't have
(36:47):
these characteristics, becauseit will be irresponsible. And
also, in our case, we are agentbased, so data needs to be
siloed, so we cannot expose, youknow, customer data from one
agent to another agent. So allof that is controlled in the
TMS, but also we have our ownseparate user pool and security
(37:09):
settings that make sure that allthe tools that we build on top
of the TMS and the tools that webuild custom also have that
characteristic right
Blythe Brumleve (37:21):
in your experience, or maybe what you've
seen at other, you know, agency,companies or brokerages, how I
guess, involved is the internalIT team when it comes to, like,
freight fraud. Are you guys, youknow, in there, you know, trying
to help combat it from a digitalperspective, like digital
warfare, or is that sort of aseparate team that, you know,
(37:44):
maybe closely works, you know,with the IT team, I'm just
trying to, I guess, maybeunderstand where it kind of
plays a role when it comes tothe just the dramatic increase
in freight fraud across theindustry, I think
Eze Peralta (37:59):
in order to have an effective security strategy. You
need everyone to be involved.
And it needs the business usersto to to be the ones also
pushing the the risk managementinitiatives. In our case, we
have very lucky that everyone inour team are, it's very much
(38:20):
aware our, yeah, our VP ofOperations, VP of Finance, VP,
you know, Director of Careerprocurement, and everyone. It's
very, very much aware of all ofthis. And they come up with the
initiatives to us, then we tellthem what is possible, then we
(38:41):
discuss what ways we could bevulnerable. But the IT team is
very aligned with the rest ofthe business in these
initiatives, and we all know andunderstand, and we try to
emphasize that with that team aswell, that what the scenario is,
what we are fighting against andbut I think it needs to be. It
(39:05):
cannot be just delegated to it,because the business user know a
lot more than they they thinksometimes about security,
because they know how a businessprocess work. They know, for
example, with when you see theseschemes of fake paperwork, or
these carriers buying MC numbersand doing all these maneuvers,
(39:30):
you know, they know the businessuser. Know the intricate
intricacies of that from from,from the business perspective,
and it can, can put in placemeasures to but we need to know
how that works sometimes, and weneed to be, yeah, in
collaboration constantly. Ithink in order to be effective,
you just, you need that. Youneed everyone on board. It's not
(39:52):
going to work if just delegatedto it, because it's just going
to be a very. Partial view, andyou're going to be covering this
side, but then they're going toattack you on the other side.
And a lot of the a lot of thethreats are social engineering.
So there's only so much you cando on the IT side against social
engineering, because if peopleare giving away their passwords
(40:16):
because someone just trickedthem into doing it,
Blythe Brumleve (40:21):
you know, is that sort of, the crazier side
of it that you've seen is justpeople willingly giving up that
information. Yeah, our
Eze Peralta (40:30):
James or CFO always says it don't click the link.
Like, if you get an email with alink that just don't click it,
like, you know, you can replyback. You can call the person.
You can, you know, try to, butdon't, don't click it, because
most of the times when somethinghappens that someone gains
(40:50):
access to an email or somethinglike that, is because someone
clicked the link, or someone gota call, got tricked into going
to a website that was not theactual website. And there's
always something like that, andso, yeah, I think definitely the
most scary part, and the mostdifficult to combat is the
(41:12):
social engineering threats.
Blythe Brumleve (41:15):
Are there any examples that you've seen that
have like, maybe at othercompanies or something, what
they're dealing with that hasthe creativity of a fraud scheme
has almost impressed you.
Eze Peralta (41:31):
What impressed me is when the quality of the
emails that people are craftingand sometimes even how they are
putting malicious code inside ofPDFs, for example, so you're not
even clicking a link. You'regetting, you're getting an email
from your boss saying, Hey, canyou check this report for me?
(41:54):
And and it's looks exactly likethem, and maybe it is coming
actually from there, from theirinbox, because they got access
to that inbox and and they say,Oh, I'm not taking any link,
even if it's it's a file, right?
So you click on the file andthen, and then you're done. So I
think, yeah. Or when at Tia, thekeynote speaker was was talking
(42:21):
about how, with deep fakes, youcan have someone, some AI,
talking to you as if it's yourboss, for example, asking you to
put money on an account, orthings like this. And in this
case, the CEO was talking withthe CFO on a video call. But it
was not the CEO, it was just adeep fake. And they asked them
(42:43):
to put money on an account forfor an acquisition they were
going to do. And they did it andlike, how can you combat that?
Yeah, it's, it's very it's veryhard, right?
Blythe Brumleve (43:00):
Yeah, I think on the podcasting side of
things, I have warned my familythat I'm like, 30 seconds of
audio is all it takes to deepfake me. And so if you get a
call from me at you know, firstof all, I don't call them, I
text them. Probably know thatsomething was up. But if you get
(43:20):
a call for me, we've developed,like, a safe word within the
family that you know, you'regoing to use in the event of
something like that happening,where we would call and, you
know, ask for money or somethinglike that. We have a family safe
word. It kind of sounds likemaybe we need to do that at the
corporate level
Eze Peralta (43:35):
too. Oh, that's definitely, that's multi factor
authentication. Like, When?
When? When people do their multifactor authentication on their
phone asking you it's that it'sa shared secret that only the
people that it's securing thatsecret know, and then it can be
used to to to unlock, right,whatever you need to unlock. And
in this case, the this, thatsafe word is acting as that, you
(43:59):
know, multi factor that are youactually you? It's like, so,
yeah. But first things first, ifyou don't have multi factor
authentication on in yourorganization by default for
absolutely everything, then youshould be looking into doing
that first, because that'sanother thing, is that we're
(44:22):
talking about really socialengineering schemes and
everything. But then people justdon't have their multi factor
codes on, on, on, on theirphones. So you're basically
that, you know, hackers and thebad actors. They try to find the
easiest path for getting it. So,yeah, just start there that that
(44:45):
was that solves like 90% plus ofthe of the phishing attacks. Is
Blythe Brumleve (44:53):
there any freight tech that doesn't exist
that you wish
Eze Peralta (44:58):
existed? Yeah?
Yeah, wow. I I think maybe allthe integration landscape in the
in the freight tech world isit's kind of all over the place.
So I think some sort ofstandardization of what we all
(45:20):
think a load is, what, what,what kind of events are relevant
to, to a load or so, I thinksome sort of, I don't know if
it's a software itself, somesort of agreement or
standardization on how we'regoing to communicate these
(45:41):
systems, between systems. EDIwas an attempt to to achieve
that, and I think it is stillbeing used big in part, because
of that, because it was somehowsomewhat successful in in you
know this creating a basestandard that then everyone need
(46:02):
to derail from that standard,and ended up doing like each
area its only world, but withAPIs, for example, it's
happening that every API isdifferent, and you need to, you
have all these point To pointintegrations, and there's no,
yeah, there's no standard. And Ithink there's, it generates a
(46:23):
lot of waste for a lot oforganizations, having to
maintain software that is just apipe to put data from A to B,
right? So that, I would say,yeah, a lot of the time, even
for tech vendor, even forvendors, it's like, it's not
valuable work. It's justsomething that you need to do in
order to get data. And then youstart, you know, adding value
(46:46):
with your product.
Blythe Brumleve (46:51):
Now, last few questions here for a potential
agent that's thinking aboutmaking the switch, is there
anything that they need to do onthe tech side of things, on
their end of things, to betterprepare them for, for making the
jump to, you know, maybe,hopefully, SBI, but maybe
another company. Is there any, Iguess, maybe standardization of
(47:16):
of what you would recommend foran agent to get ready to make a
switch.
Eze Peralta (47:21):
I think, I think asking the questions to to the
to the agent network that thatthey're going to they're
evaluating about what's theirtech stack. But no, no, no, not
as as listing the vendors thatthey use, because that you could
have a really good list ofvendors, but not have them
(47:42):
working together. Well, right?
So it's okay, can, can thisnetwork give me all fulfill all
the use cases that I need tofulfill? For example, we offer,
we push data to shippers.
Sometimes they needed as a CSVfile with a very specific
(48:03):
format. Sometimes they neededthe API. Sometimes they need to
be Adi. Sometimes they need itas they come. Sometimes they
need it in a schedule, right? Soif you have shippers that are
requiring this type oftechnology, then you need to ask
very clearly at the beginning,hey, can you push data to my
shippers in the way they needit. How long it takes for you to
(48:23):
complete the new integration ifa new shipper needs it? For
example, can you automaticallyquote, because I have a shipper
that asked me to respond to aquote within a minute? Can you
do that like and same for you?
Know what carrier fraud, carriervetting and compliance is like,
asking the questions, you know,but trying to get to the proper
(48:46):
depth of the question. It'slike, Oh, do you use one
question could be Oh, do you usehighway? Or do you use RMS, or
do you use my care packets?
That's one way of askinganother. Another way of asking
is like, how do you handle witha carrier trying to do this? How
would you prevent this fromhappening, or how do you use
(49:10):
those tools in order to protectme or protect my business? And I
think, trying to find the depthin the questions beyond, oh, we
use this vendor, okay, but howyou use it? Why you use it? Why
that one or not the other one?
Like trying to get a bit moredeep into into the into the ask,
into the asking.
Blythe Brumleve (49:33):
Is there any?
Well, actually, one more quickquestion that just popped in my
head. AI agents, not necessarilylike the call ones, but the the
ones that are promising, youknow, to fix, you know, be your
internal marketing department,or be your internal sales
department. It Do you see therise of sort of a AGI agents
that you know, you have a littleyou know, minions that are doing
(49:55):
all of your work for you? Yeah.
Eze Peralta (50:01):
So to be honest, like from, from what I've been
seeing so far, I think we're farfrom it still, whenever I try
to, I mean, I use coding toolthat has some sort of agent in
it and and it also has, like anauto complete feature. I end up
(50:26):
using the auto complete morethan the agent, because a lot of
times the agent is just going togo in a loop of trying to solve
a problem, but in a very kindof, like, naive way, and maybe
it's not understood for certaintasks. Yeah, I can see it
working for certain tasks, but Idon't see it happening at the
(50:49):
level that is being advertised.
Of, oh, now you're going to havethese digital workers that is
going to just do all of yourcalls for you, and it's going to
give you all, you know, allthese benefits, and I have yet
to see something that shows methat that's possible and that's
and also, one aspect that we, Iwas talking with my team
yesterday, is what you get, youknow, when you're in when you're
(51:13):
doing something by yourself,like, if I'm trying to solve a
problem and I don't know how toDo it, then maybe I Google,
maybe I use a AI or whatever,but I'm still using my brain to
solve the problem. Once I solveit, I learn something right. And
same with human interaction,maybe some employee making a
(51:34):
mistake also reveals that thereis a process that needs to be
improved, right? So then you canso I think error and and it's
part of life, and it's also partof learning. And I think we are
entering to this notion of we,we're going to have these agents
(51:58):
doing all the work for you, butthen after you see the work
done, have you learned somethingfrom that process? So I see one
of the possibilities is that ourlearning curve, it goes up, up,
up, up, but then now we are justdelegating all this resolution
to these automations, but wedon't know how they work,
(52:21):
because nobody knows what an AIagent is thinking, and nobody is
even if we could, no one isgoing back and say, Oh, how did
you reach this conclusion? Wasit accurate? Was it not what can
I learn from the process ofreaching the conclusion? So I am
a believer that that process ofreaching a conclusion is really
(52:44):
valuable, and learning fromthese processes, learning from
mistakes, is valuable. And Ithink sometimes the promotion of
the type of tools forget thatpart, and they're just too
focused on the solution. Insteadof trying to understand how we
get to a solution, how theproblem works, why the problem
(53:07):
even exists, could we definethis problem out of scope so we
don't even have to solve itbecause it doesn't exist
anymore? You know, trying tofind more depth, I think, I
think it comes down to depth. Ithink a lot of these tools are
being promoted with not enoughdepth in mind, just very
tactically, and we're notthinking about, you know, maybe
(53:31):
you needed to, maybe you neededto walk the path. Make a little
mistake, that is, as long as itdoesn't destroy you, right? You
needed to walk the path. Makethat mistake, learn from it,
improve and keep going and andyou can still do that while
(53:53):
you're using AI agent, but ifyou just delegate everything to
to and then you don't thinkabout it anymore, then we're
becoming kind of like thinkinglazy, right? Like we're becoming
lazy at thinking and Ipersonally don't like that. I
Blythe Brumleve (54:11):
completely agree, and I echo that
statement, because I on one sideof the coin, I will say that,
you know these even like Groxdeep research tool. I love it,
but I use that inside of chatGPT in order to help me come up
with an interview flow. Andit'll I did it for this
(54:31):
interview. For example, I tookour previous episode, I took the
transcript. I said, Give me alandscape of, you know, the
current freight trends andfreight tech trends and
technology and things that areon the market, analyze the
transcript and then come up withan interview flow based on that.
And I would say probably 60% ofthe questions were pretty good.
(54:51):
But if I didn't do the activelistening to our previous
episode and also to otherepisodes you've been on. I
wouldn't have been able to, Ithink, craft a better interview
for the sake of thisconversation. And so it's it. I
don't know that that's somethingthat you can replace. Is that
(55:14):
active learning by doing it youhave to still do the thing. And,
you know, maybe there's someways where you can automate the
boring stuff, but there's still,you know, a creative aspect that
you create, a problem solvingand learning that I think is
still needs to be prioritized.
Yeah,
Eze Peralta (55:31):
and in this case, what the process you're
describing you were driving theprocess you were not delegating
to an agent that was driving theprocess, an AI agent that will
drive it. But you were drivingthe process. You were using a
tool to categorize to give you,you know, some help, but you
were in the in the driving law,and I think that's important and
(55:53):
and also it's like, it's, it'sfun to be a human, and it's fun
to connect with humans. Big fanof humans. So
Blythe Brumleve (56:01):
don't tell the AI that it
Eze Peralta (56:05):
already knows.
Probably that's a topic foranother, another podcast. But
like, then there is that aspectof like, do you want to be
working with an AI agent? Or youwant to, you know, are we
working for? What are we workingfor humans? Are working for
ourselves, for human species,for that's more philosophical,
maybe, but that comes into playtoo. When we're talking about,
(56:29):
you know, replacing people, it'slike,
Blythe Brumleve (56:35):
I don't know, I want to do more of the things
that I like doing and less ofthat. And I think that that's,
you know, sort of the the commoncomplaint I hear with AI and
automation is like it's takingaway, or some people feel as if
it's taking away from the thingsthat we want to do more of
instead of the things that wedon't want to do more of, which
is, like, I don't know, cleanthe bathroom, or, you know, fold
(56:56):
your clothes. Like nobody wantsto do that. But, you know, AI is
not fixing that yet, and that'swhat we want it to fix. We don't
want it to take away the thingsyou know, that do make us human,
creative problem solving, youknow, creative indent
adventures, things like that,talking to people, developing
relationships with people. Ithink all of those things are
really important. You kind ofhit the nail on the head for the
(57:20):
rest of you know, I guess say2025, and beyond. Are there any,
you know, new tech solutionsthat you guys are working on, or
integrations that you couldshare with us, or or things
you're thinking about that youthink the audience should know
Eze Peralta (57:35):
we're working continue improving on, on the
capacity, capacity tools that wehave. We know that the you know,
market still somehow lose onthat sense. But we know, you
know, being prepared for, forfor different market conditions
as as they show up. So addingmore sources of capacity to our
(57:57):
capacity hub tool continueadding on, you know, more data
into our risk management tools.
Also we're adding these morefeatures on the AI email, you
know, creating loads from emailsand and these categorizing email
requests, working also onautomation of of bidding to the
(58:21):
shipper side so and, and also alot of work on on the back
office, because we, we need to,we need to provide, you know,
best service in the office,because our agents, we work a
(58:42):
bit differently than maybe otherother agent models, where agents
just do the operation, theycomplete the load, and then we
take over the entire AR and APcycle, right? So in order for
that to be effective, we wouldneed to have a high level of
automation, high level ofefficiency on the back office.
(59:05):
So a lot of work on 2025 is alsogoing to go into continuing the
already, already highlyautomated back office is going
to continue being more automatedand providing our staff with
tools to, yeah, maybe to reducesome of their data entry and so
(59:25):
they can focus on being moreanalytical, on on the audit and
these things so, so yeah, that'sthose are things we're working
on. A
Blythe Brumleve (59:39):
lot of work you guys are got ahead of you for
this year and beyond, but I'msure you know a lot of the
agents are extremely thankfulfor the investments that you
guys continue to make into theplatform and the program itself
as a where can folks find you,follow more of your work, maybe
connect with you at a futureconference.
Eze Peralta (59:58):
Yeah, so my LinkedIn. I am sa Peralta. You
can find me there. My email is EPeralta at SPI three, bpl.com
you can email me and I'll try toreply as soon as you can. Yes,
no, I don't use email bots, soif I take time, it's because I'm
actually responding. And if yousee me at any conference,
(01:00:22):
generally, I go to Tia andcapital ideas and technovations
and maybe a few other more. Idon't have anything planned as
of now. For for conferences, wehave our annual agent conference
this week. Yeah, exactly. It'sgonna be fun. I be fun. But
(01:00:44):
yeah, if you see me at anyconference, just Yeah, would
love to connect. Well,
Blythe Brumleve (01:00:49):
perfect. This was a great discussion.
Hopefully people you know, likethis one just as much, hopefully
more than than the previousepisode that we did together.
But I'll be sure to add allthose links into the show notes.
Just make it easy for folks, butas a this is great. Thank you so
much for your perspectiveinsight. Thank
Eze Peralta (01:01:05):
you very much. Need for having me. Absolutely
Blythe Brumleve (01:01:12):
thanks for tuning in to another episode of
everything this logistics, wherewe talk all things supply chain
for the thinkers in freight, ifyou like this episode, there's
plenty more where that camefrom. Be sure to follow or
subscribe on your favoritepodcast app so you never miss a
conversation. The show is alsoavailable in video format over
on YouTube, just by searchingeverything as logistics. And if
(01:01:32):
you're working in freightlogistics or supply chain
marketing, check out my company,digital dispatch, we help you
build smarter websites andmarketing systems that actually
drive results, not just vanitymetrics. Additionally, if you're
trying to find the right freighttech tools or partners without
getting buried in buzz words,head on over to cargo rex.io
where we're building the largestdatabase of logistics services
(01:01:55):
and solutions, all the links youneed are in the show notes. I'll
catch you in the Next episode inGo, Jags, you,
Popular Podcasts
Cold Case Files: Miami
Joyce Sapp, 76; Bryan Herrera, 16; and Laurance Webb, 32—three Miami residents whose lives were stolen in brutal, unsolved homicides. Cold Case Files: Miami follows award‑winning radio host and City of Miami Police reserve officer Enrique Santos as he partners with the department’s Cold Case Homicide Unit, determined family members, and the advocates who spend their lives fighting for justice for the victims who can no longer fight for themselves.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.