November 10, 2025 • 53 mins
Text us your thoughts on the episode or the show!
In this episode of OpsCast, hosted by Michael Hartmann and powered by MarketingOps.com, we are joined by Tracey Fudge, AI Operations Architect and Agentic Workflow Designer at AI By Thrive.
Tracey has spent the past several years working hands-on with language models, automation systems, and what she calls agentic workflows. She helps marketing and operations teams move past AI buzzwords and turn technology into practical tools that drive better results.
The discussion focuses on how to build real, usable AI systems that enhance creativity, improve efficiency, and deliver measurable business outcomes. Tracey explains what agentic workflows are, how they differ from traditional automation, and how teams can start integrating AI into everyday work in a thoughtful, scalable way.
In this episode, you will learn:
- How to apply AI and automation in practical marketing and operations use cases
- What agentic workflows are and how they create intelligent systems
- Techniques for prompting and choosing the right AI tools for each task
- Ways to balance human creativity with AI assistance for better outcomes
This episode is ideal for marketing and operations professionals who want to make AI an integrated part of their workflow without losing the human touch.
Episode Brought to You By MO Pros
The #1 Community for Marketing Operations Professionals
Ops Cast is brought to you in partnership with Emmie Co, an incredible group of consultants leading the top brands in all things Marketing Operations. Check the mount at Emmieco.com
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Michael Hartmann (00:24):
Hello everyone, welcome to another
episode of OpsCast, brought toyou by MarketingOps.com and
powered by all the MoPros outthere.
I am your host, MichaelHartman.
Flying solo as a host this weekbecause we are recording this
during Mopsapalooza 2025.
So if you're there andlistening to this later, we hope
you had a good time there.
(00:45):
Uh, Mike and Yumi, I'm surewe'll be back soon.
In fact, we've got planned arecap, so watch for that when
coming into your inbox.
So today I am joined by TracyFudge.
She is an AI operationsarchitect and agentic workflow
designer, and uh, she's doingthat on her own with her
consultancy AI by Thrive.
Tracy has spent the last fewyears deep in the AI trenches
(01:06):
experimenting with languagemodels, automation frameworks,
and what she calls agenticworkflows.
She helps marketing andoperations teams not just
understand technology, butactually put it to work.
So today we're going to betalking about how to move beyond
the AI buzzwords, not thatthose are out there, to build
real usable systems that savetime, spark creativity, and
drive better business outcomes.
So, Tracy, welcome to the show.
Tracy Fudge (01:29):
Thank you so much, Michael.
I'm really excited, and I'm Iam, I'm I'm missing MAPSA, but
uh I'm there virtually and in myheart, a lot of my friends are
there.
Michael Hartmann (01:39):
Me too.
I have serious FOMO going onright now.
Well, I'm glad, but I am I'mgrateful that we've been able to
stitch this together.
I was going back to some of theconversations that you and I
had to try to get this started.
It went all the way back to itis now late October.
Uh, and I think it's we startedin February sometime.
So um good things come to thosewho wait, right?
Tracy Fudge (02:02):
Yes, exactly.
Michael Hartmann (02:04):
All right.
Well, why don't we let's startwith this?
Like, let's start with maybe alittle bit of like a short
version of your journey.
Um in particular, like the lastcouple of years, like what led
you from marketing andoperations, uh, kind of the
traditional one we were justchatting about this before we
started recording about how somany people are still focused on
platforms, um, and getting youinto AI systems, agentic
(02:26):
workflows, things like that.
Tracy Fudge (02:29):
Yeah, so it was it was an easy transition.
Well, one, I've been in tech mywhole career.
Um, you know, I started insoftware sales uh before I
really understood what softwarewas.
You know, my explanation was,well, what's software?
And I was like, I don't know, Ithink it goes on hardware.
So that's how far back it goes.
Michael Hartmann (02:46):
You're not gonna start talking about floppy
disk drives, are you?
Tracy Fudge (02:49):
You know, there was a day when we when software was
a new word.
Um, and you know, so then fellinto mops.
Um, you know, I was managing uhsocial media platform for
somebody, and they're like, hey,can you go manage this email
platform we have?
And that's sort of, and mostpeople fall into mops by
accident.
I'll start.
You know, there's no collegecourses or anything like that.
Michael Hartmann (03:11):
Not yet.
Tracy Fudge (03:12):
Not yet, yeah.
But the um the the AI part ofit was just something that was
just intriguing to me.
And you know, systems areconnected via API.
So the agentic part of it wasalready being done.
We just weren't calling itthat, right?
So when I um decided it to godeep and really learn, you know,
(03:34):
on what agentic platforms were,you know, I had known Zapier
and Zapier, you know, nobodyreally knows quite it.
Michael Hartmann (03:41):
They told us that we've had people on from
there.
It is Zapier, like happierobviously fingers.
Not that we're like, we're notthis is not a promotional thing,
but I think I now have thatthat rhyme in my head.
Tracy Fudge (03:54):
And I don't know how far back you've been using
it, but I was using it beforepeople did correct me on what to
call it.
Um, and uh so that's been gosh,and I mean well over 10 years.
But um anyway, the the whole AIpart of it, the agentic part of
it was sort of an easiertransition to sort of understand
(04:15):
what this all meant.
And obviously I was curiousabout well, what's an LLM?
And oh, there's other types ofLLMs.
Oh, what do these frontiermodels mean?
And then putting them alltogether and making the
intelligence layer over theconnections.
Um, you know, and that's justthat's much more complex than
(04:37):
people realize because you gointo the conversations around
data.
And when you're inside of anorganization, sometimes
marketing doesn't have all thedata or the systems.
You know, I it systems sit overthere.
So that's a big uh gap that'sstill there.
And teams are starting to cometogether and realize that AI
kind of goes over the wholething.
Um, and AI needs to work on thesame set of data, uh, not the
(05:04):
same fields per se, but let'sall talk about the same data.
Let's make sure that thepipeline reflects in the CRM, et
cetera, et cetera.
Uh, so that's really wherewhere I got into it.
And it's it's still a new, youknow, people uh frankly don't
understand it.
They think it's just workflows.
Oh, just flow build sixworkflows, and that makes it,
(05:28):
you know, and that's not iteither.
Um it's part of it, butcertainly when you're looking at
a business, you've got toconnect it all together.
Uh and finance needs to havethe same initiatives as RevOps,
as Mops.
I mean, it it needs to all flowtogether.
Michael Hartmann (05:42):
Yeah, it's I think that's the that's the uh
the goal, right?
The only thing you you saidsomething that uh marketing
sometimes doesn't have all thedata.
And I I I would push back a biton that and say that I think
most marketing teams have moredata than they know what to do
with, um, and probably lackskills and knowledge to be able
to do any much with it is islike to me that's the bigger
(06:05):
gap.
It's not data itself, but theability to understand it.
Tracy Fudge (06:10):
Yeah, um, and I think you I would say that you
are correct in that aspect.
When I think of data, I meanthe the ability to action on it,
to add some predictiveanalytics, right?
Uh to be able to hold the datain a sandbox and just kind of
play around with it, apply somemachine learning, some
prediction.
And that way your leadscoring's a little bit better.
(06:30):
But if you're still sayingbecause person A watched Webinar
X, let's give them five points.
And it doesn't take intoaccount all the other things
that said prospect or lead didalong the way.
Michael Hartmann (06:44):
Sure.
Yeah.
Yeah.
That's interesting.
Yeah.
Um well, so you you kind ofhinted at this when you were
talking about this, but yeah,something you said to me, which
I've been thinking about, isthat you said AI isn't really
new, that it's yeah, automationand data taken to another level,
I think is the way you put it.
Like, what do you like?
What do you mean by that?
Break that down a little bit.
Tracy Fudge (07:05):
So hooking up, let's say HubSpot to Clay or
HubSpot to Salesforce, thathappens for the most part, out
of the box, right?
API, API, field match, fieldmatch.
Um so that's already sort of aningenic portion connecting
APIs.
I think uh, you know, when youadd in the intelligence layer,
(07:29):
for instance, if you were todrop the data that's moving over
to the marketing automationplatform, drop that into a
knowledge play, uh, an LLM nodeto do something with the data,
make decisions, apply tags, dowhatever.
And maybe that when you drop itinto say N8M, and I'm just I'm
making I'm making reference tothings that we all know or to a
(07:51):
Zapier.
But the point is thatconnection doesn't have to be
direct anymore.
You pause it, you do thingswith it, and then you map it.
Michael Hartmann (08:00):
Right.
Tracy Fudge (08:00):
Um, and then you iterate, and that mapping can
change.
So it's that part of it that isnew, but the actual technical
API stuff, even installing aplatform like an N8M, things
things are so easy, uh easier tounderstand once you've
(08:21):
developed on one platform, youcan do it on another.
One CRM largely looks likeanother, you know, minus some
capabilities, but the conceptsare the same.
Michael Hartmann (08:33):
Yeah.
It's it's interesting becauseyou um it feels like well, one
thing is like some of theseterms get thrown around as if
they're the same, right?
And I like in my head, I tryI've started to try to break
down like um AI does notnecessarily mean automation,
(08:56):
right?
Uh AI also I think generallygets conflated with LLMs, where
it feels like there's a wholelot more that's under the AI
umbrella.
LLM is sort of one type of AIum modeling kind of thing,
right?
Where you've got you'vementioned machine learning.
I think predictive analyticscould be a part of it, or it's a
component.
But it feels like people kindof mix those up.
(09:18):
And to your point, I think somepeople are just still have
maybe not blinders on, but liketheir their focus is on like
small relatively like let's dothe replace this small thing
that we're doing with somethingthat has AI as in quotes about
it.
But um when they start to openup the the aperture a little
bit, they start to getoverwhelmed, right?
(09:39):
So for those people who arelike, hey, do you feel free to
(10:44):
disagree with my my assertionthere about like the mixing of
terminology, which is I thinkvery common in a scenario like
this, but also like anyrecommendations for those people
out there in marketing,marketing ops who are maybe
feeling a little overwhelmedabout like where to even start
with this AI journey, if youwill?
Tracy Fudge (11:05):
Uh well, most people that I know that did it,
let's say within the past threeyears, because that's really
where when I went in.
Michael Hartmann (11:14):
We it's crazy that it's been three years for
stuff, Rick.
It feels like it was just likethe other day.
Tracy Fudge (11:20):
I um I actually started, there was somebody
something kept coming in myfeed, right?
It was this no code, low code.
And I was like, okay, thismight be part of it.
And I think um that is easierto start because I'm not a
software developer.
I can I've done Python, youknow, I did a serious analytics
(11:41):
course where I had to writePython code, but then there's
stuff that does it for you.
So, you know, that sort ofskill set goes away.
But the the whole idea of um AIjust being connections, it
could be um, you know, an LLMplay.
But in terms of starting, Idon't start with the start with
(12:03):
the low-code, no-code platforms.
And I personally started onMaven and I stayed there for the
whole time.
The the courses uh people thenuh did more courses and more
courses.
And plus you're with a cohort.
Michael Hartmann (12:17):
And I think um Maven is a like a training.
Tracy Fudge (12:22):
Yeah, uh not the platform.
Okay, uh I'm just saying Mavenwas where I started.
Michael Hartmann (12:28):
Sure.
Tracy Fudge (12:29):
Um and they do uh, or at the time were really the
only ones giving true how tostart with AI classes.
Okay.
Um I think that that's one wayto start, just real basic.
What is AI?
And now, fast forward two and ahalf, three years, you could
(12:50):
get the same thing on a YouTubevideo.
I mean, honestly, just break itdown.
I think the hard part is this.
Our background is marketing andmarketing apps.
And so we're flooded withplatform.
Oh, Salesforce has AI.
Look at these Salesforceagents, right?
Spot has AI.
Oh, everything's got AI.
(13:10):
So I think there's a falsesense of being AI when it's not
AI.
It is just the platform'sversion of AI.
You have no control over theprompt, uh, little control over
the knowledge base, um, and it'scontext, and you don't know
what model is in there.
Um, a true AI person that wantsto add the knowledge layer has
(13:35):
all the levers to temperatureand knowledge base and prompting
and system instructions andprompt chaining, and that's the
limitation.
So you might, and then anotherpart of it is some of the AI in
these things, it's just fillingout fields, right?
It's um it's just doing, oh,that field needs this data or
(13:57):
that field needs this data.
And you can do that now with alot of platforms, you know,
Airtable off is their uh AIcomponent, Zapier's got one,
they've all got one, but it'skind of filling out the
architecture for you, which ispart of the part of part of what
AI is.
But once the scaffolding isbuilt, where's the knowledge?
(14:20):
You still have to create that.
And that cannot be done yetwithout a human that knows what
the hell they're doing.
Yeah.
I think unless you've youunderstand data, and I I always
I've always said this is a dataproblem.
Um, understand not just wherethe data is, but what the data
(14:42):
is, uh, what data matters, uh ifthe data changes what's the
story it's telling.
Michael Hartmann (14:47):
Yeah.
Tracy Fudge (14:47):
Exactly.
And we're not having enoughdata conversations.
Michael Hartmann (14:51):
Um well, they this goes back to like one of my
absolute strong beliefs rightnow is that we just lack in
marketing in general, in ops inparticular, um the skill set to
really do much with data, right?
I think there's just not a deepunderstanding about how to do
it.
Um I want to go like somethingthat's just sort of clicked with
(15:14):
me that I think maybe is partof the uh the challenge for some
people in this, in that we'reso used to having what I would
call, I guess, deterministickinds of solutions.
So, you know, I get a bunch ofjob titles, I'm gonna put
something in there that I definethe rules to say this is the
job level, right?
So that which is sort ofinferred from the titles.
(15:37):
And yeah, I think an a veryobvious use case for probably
LLM in this case, right, as partof a flow is to use that to
categorize jobs, people with jobtitles into categories that
make sense and use someintelligence on that, maybe with
a little bit of humanoversight.
But I think, yeah, what youdescribed, right, in some of
(15:59):
these other ones, there's alittle bit of this trade-off,
right?
You could either do thatyourself and you have a little
more control over what'shappening inside the black box,
or you take advantage of some ofthese tools that are being uh
embedded within other platforms.
You mentioned a few, right?
But it becomes even lessapparent, like what's happening
within that black box.
And so there's this lack of, Ithink there's this like, do I
(16:22):
can I trust it?
Can I not trust it?
Yeah, how can I tweak it?
Can I tweak it enough?
And it feels like maybe that'spart of the like the resistance
for people who are strugglingwith it, where they, yeah,
you've got less like, oh, I wantto take advantage of the stuff
that's being provided in theplatforms that I'm used to
working in every day, but itdoesn't really give me, like, I
don't know enough about it.
So somebody starts questioningit, I can't answer it, which
(16:43):
puts them in a position of notfeeling comfortable about it.
And then at the other part islike the other struct struggle
is like I still have to, like,how do I fit in time to go
beyond that?
And that's I think it feelslike that's maybe the rub that's
the or the the the the thingsthat are making it difficult for
some people to go, like, howcan I step in and take a step
forward in this?
So does that sound right toyou?
Tracy Fudge (17:06):
Yeah, it's a gap.
And I have friends at verylarge organizations, and um,
they don't have time to go learnthis.
And I think when you're a W-2,uh you don't have a lot of time.
We're all overworked.
I had, I chose to take timeoff.
(17:26):
Well, I have a side business,and that's what's been
supporting me, to be honest.
Um, but I've taken uh a lot oftime and taken very uh a few
projects, but it's not, I'm notworking 40 hours a week for
someone else.
Uh, it's probably 20, and thenI work the other, you know, even
(17:47):
weekends to learn this stuff.
And I think it is it's a realum miss of an opportunity uh
because you've got really smartpeople that are working really
hard in these huge organizationsthat are not getting upskilled
in this.
Because you and it's and here,here, here's the other element
(18:11):
to this.
When you go in and design aworkflow, it's not something you
touch today for an hour, youcome back next Tuesday and you
finish it.
No, it's deep work, it's 12hours sitting down, getting up
just to go pee or eat and doingit end-to-end.
And there's no, it's heavy,deep can't outsource that part
(18:32):
to AI, right?
You can't.
And it is you you've got toremember where you are and
you're deep into that.
You're living inside theworkflow.
I mean, I dream in spreadsheetsand workflows because you know,
and not everybody has thatability to say, okay, you wake
up at seven and you realize thatit's you're gonna work on this
until seven at night, and youdon't stop.
(18:54):
You don't answer the phone, youdon't, you don't get on social,
nothing.
You're just inside thatworkflow.
Um, and hopefully in that timeyou you have enough good data,
good parameters to test it andit worked out.
But some of that can just be,you know, beating your head
against the wall to try to makeit work.
And um, you've still got to getit reviewed for the client, you
(19:17):
know, for the client or foryourself or whoever to see your
art because you're the one thatdesigned it.
Michael Hartmann (19:24):
And here's the that's the main right, and
that's hard to to test.
Tracy Fudge (19:31):
Nobody's done this before.
Michael Hartmann (19:32):
Well, and because if you're using so in my
head, one of the distinctionsI've started making is there's
the AI comp AI can be uh thereare AI components could be used
in some sort of automation orworkflow or agentic model,
right?
So pick the tool.
But there are other like thereare other more deterministic
(19:52):
ones.
And I think what again I getback to like when you're trying
to validate, right?
It's it becomes less of a umdid it do what it was supposed
to in in an exact way, or did itproduce an output that I is
expected even if I don'tunderstand how it came to that?
And that's a very differentvalidation.
And and then if you're notcomfortable, if you're the kind
(20:13):
of person who's not ascomfortable with these sort of
um what's wrong, right?
Like less, like unless like,oh, this is this directionally
correct, right?
If you're comfortable with thatas a quality assurance um level
of quality or completeness,then you're gonna really
(20:35):
struggle with this stuff, Ithink, right?
Because a lot of times whatI've seen, even when I'm just
using, say, uh an LLM, Chat GPT,Claude, whatever, right?
The output I get from my backand forth in terms of prompts,
right?
Sometimes I go, I could Iscratch my head, I go, like,
where did it come up with that?
Right.
It we've all heard about thehallucination thing, but also
(20:58):
sometimes it just doesn'tfreaking follow the directions,
right?
I remember working on somethingwhere I was like, I want you to
draft a document.
I had we'd gone back and forth,I had a lot of stuff, dropped
this document and came up withsomething.
I was like, I like that, but Idon't like this one piece of it,
right?
It was like I think I had donea header image or something.
I said, Yeah, just go fix that.
And it completely redid thewhole fucking thing.
Tracy Fudge (21:19):
So yeah, and I was like, no, like that cost us
down, right?
Like that that you can't putthe whole prompt in one place.
Michael Hartmann (21:28):
Well, and I hadn't, like I had gone, but
like I had produced somethingthat was probably 80% of what I
wanted.
It was in line.
I'd gone through like as manysteps.
It was just like this last one,like the formatting, the layout
of this document wasn't what Iquite wanted.
And so I asked to make onesimple change and it completely
like it redid the whole thing.
(21:49):
And I was just couldn'tunderstand why.
Tracy Fudge (21:51):
Yeah.
Michael Hartmann (21:52):
And that's the kind of stuff.
So if you take that, right,where I'm like going back and
forth and it took me time to doit, and you say, like, I'm gonna
do something like that, maybenot exactly, and I'm gonna be
make that as a part of anoverall flow, then like, how do
you test that and evaluate it?
That's like that's where theuncomfortable like people get
uncomfortable with how do I fitthis into something where people
(22:13):
need to trust the overalloutput of that, not just one
piece of it.
Tracy Fudge (22:20):
Yeah.
I was introduced to a platformcalled Cassidy.
Um, it is it's a pretty coolplatform.
I use it mainly for uh doc gen.
So creating um a workflow tocreate an RFP process or a
workflow to create uh apresentation or talking point or
(22:43):
or something like that.
It's to do one thing is eightdifferent agents or assistants,
right?
One of which is the JSON, andthe JSON is gonna be how that
document is formatted, and youtack that on to the end and you
go, okay, I want the header tobe this,
da-da-da-da-da-da-da-da.
(23:04):
And then and then it goesthrough that node, which is a
code node, um, and then it'sformatting.
And that uh had I not gonethrough Cassidy workflow on a
simple doc generation thing,yeah, I wouldn't have realized
that no, no, we we first have tohave a research agent.
(23:24):
Oh, and then we oh, first weneed to have the brand voice
agent, you know, or wherever youput that in, how do I sound?
So you've got eight differentassistants to do one eight-page
output.
And I know that sounds possiblygratuitous, but it's how you do
it.
Michael Hartmann (23:44):
Yeah.
Well, I think so going back tomy original question, to you,
like, where should people start?
It feels like the way to startis because you can't do all this
in one sitting.
If you're a kind of personwho's busy with their day job
and you can't do it all in onesitting, you could start to
experiment with pieces of whatmight be a bigger process,
right?
(24:04):
And you like figure that out.
And then because I it I I spenta lot of time trying to learn
how to just use I have a biastowards ChatGPT, not for any
good reason other than it's justthe one like was one of the
first ones.
I've spent more time with it,and I'm open to the other ones.
But I've done lots ofindividual things.
It wasn't until I'd been doingthat for a while where I sort of
(24:26):
lifted my head up and said,okay, I keep hearing about this
agentic right in quote stuff,you know, what does that really
mean?
And I played around withdifferent tools.
And I don't think that if I Ithink if I'd started with the
N8Ns of the world, or like I II've had pretty good luck with
relay.app, right?
There are others, I know thereare plenty of others, but if I
(24:48):
had started with those asopposed to learning how the LLMs
work, I think I would havestruggled even more because I
would have tried to do what yousaid, like I would have tried to
do like a whole bunch of stuffin one uh AI node of a process
as opposed to breaking it downinto smaller chunks, which does
feel on the surface to beinefficient, but it feels like
(25:11):
like now I understand that uh inorder to maintain a level of
quality or consistency, that'skind of what you have to do.
Tracy Fudge (25:21):
Yeah.
Michael Hartmann (25:21):
Which means then you have to understand like
which it should be a strengthof ops people, right?
What's the process that I gothrough?
That what's the well, how can Ibreak down this overall process
that I want to automate, haveagents do into smaller steps so
that I can trust the output atthe end.
Yeah.
Tracy Fudge (25:40):
Yeah, it's it's and you evolve to that point.
We all start, we're alldifferent learners.
And for me, I like being partof a cohort, a learning cohort.
And I didn't want to, I kind ofpaused in my mops world
intentionally because I didn't,I felt like mops wasn't where I
wanted to go.
So there were a few mops peoplein the community where I where
(26:04):
I ended up, uh, and I still am,but not many.
It's CEOs are in there.
You know, it's just differentpeople having different
questions.
And I personally would beexposed to more use cases than
just hooking up a CRM to amarketing automation plan.
You know what I mean?
Like I it and that's wherethat's where um so I went, you
(26:29):
know, that's where the decisionI made to get out of the mops
mentality and the mops worldbecause it's all mops.
And while mops is veryrelevant, take the blinders off,
right?
Yeah, you just need to go andunderstand it.
And it is it's overwhelming.
It's overwhelming.
I cannot tell you the it's justit's it's it changes the way
(26:56):
you see everything.
Michael Hartmann (26:57):
Yeah.
Tracy Fudge (26:58):
It's not just AI, it's it changes everything.
I mean, I look at things sodifferently now uh in the world,
and um it it changeseverything.
And I don't know where to tellpeople to start that.
I don't, I for okay, so fastforward today.
If I were somebody gettingstarted, I would say go to N8N,
(27:22):
do one of their startertemplates, and just build it.
Just at least get the the clickand the link and all that
stuff, just understand it.
Yeah, and mops people willunderstand that because most
mops people, I mean, actually, Iwouldn't say most mops people.
There's a set of mops peoplethat don't hook systems up.
You know, they're not on theAPI side and they're not on the
(27:42):
data side.
But the ones that are will getthat part.
Um and I would start there.
N8N's gotten a lot better.
I was an NADN, I left.
I didn't, I was not a fan ofthe community.
Um, I didn't feel like therewas a lot of help there.
I thought it was a lot ofbravado and not things that were
(28:04):
true.
Uh, and it turned out that wasthe case.
A lot of people were saying,well, we'll come in and do all
this magic, and and theycouldn't, they could just build
the thing inside of NADN and itwas visually appealing, but it
didn't do anything.
So anyway, um, but now I thinkit's a lot better, and they do
have things for people to golearn quickly.
And I don't, and I think youjust do these P these bytes of
(28:25):
things, maybe an hour each, andjust build and then and then
move into the data layer.
And that's where Airtable andSuper Base uh or Superbase, you
know, that's another one.
Michael Hartmann (28:37):
Do you know how to say I have no idea?
Tracy Fudge (28:41):
But they're very good.
Uh and when I started to stepinto data and believe it or not,
uh into knowledge graphs.
So, and we could go into this,but AI has to know context and
it has to know how things aretagged so that it can go and
retreat it.
(29:01):
Um, this is different thanGoogle's algorithm looking for
keywords.
So AI has to have context.
So you've got to have a definedontology.
And that means, you know, thisis a rotter word for saying
tagging, you know, you've got toknow what things mean and then
how they're related.
Uh, and that's where triplescome in.
So AI can't move beyond thatlayer.
(29:25):
So if you don't have theknowledge graph and the nodes
and ontology, it can't work.
So when that light bulb wentoff in my head, I was like,
damn.
Michael Hartmann (29:37):
Yeah.
So that's like con that's likecon providing context to a whole
nother level, right?
Tracy Fudge (29:44):
Yeah, but it's getting it right.
It's so it doesn't hallucinate,right?
Right.
So you'll see, and they may notbe called knowledge graphs, but
it is.
Michael Hartmann (29:54):
It's gonna No, I it it's a good, I think like
me, it's a good mental model,like in my I I can already kind
of envision it what you mean,but ontology is a big word that
I think most people listening orwatching may not really fully
understand.
Tracy Fudge (30:06):
Um it's getting it is giving the relation of how
things are related.
Michael Hartmann (30:13):
Yeah, yeah.
Um it's like librarian, right?
Virtual librarian.
Um not that anybody goesactually goes to libraries
anymore, but um that's a wholeother topic.
Uh so I'm curious, like whatare what are some of the like
can you give us some examples?
Maybe uh going back to earlyon, like things that you've done
(30:35):
where you've kind of that werehelpful in your learning and
still were maybe productive,maybe weren't productive in
terms of an actual thing you canapply in the work, but it was
like a good learning experience.
And maybe something more recentthat you've done as you've kind
of learned and evolved and andwhat you've seen as possible.
Tracy Fudge (30:55):
Well, I was lucky enough to have my own business
to sort of act as a playground,right?
Michael Hartmann (31:00):
Sure.
Tracy Fudge (31:01):
So I automated um a lot of the front office, a lot
of the back office.
Michael Hartmann (31:07):
Um for your business, you mean?
Tracy Fudge (31:10):
That's where I practiced a lot of it, um, where
I learned um real life thingsthat that were impactful.
So um that's how I was able to.
To do things live and playaround with it.
Not everybody does that.
They're still so if you're inan organization, you've got
their platforms, you know,you've got their stuff, and it
(31:33):
may not, they may not have NADN,you know, they may not want you
playing on an agentic platformto understand it all.
But those the so theefficiency, so that's how I
learned a lot of it.
I just threw my own business,my own use case, I built my own
knowledge graph.
Um, I built my second brain,and that's all knowledge graph
work.
Um, and I'm actually buildingthe second brain for someone as
(31:55):
a client right now.
And that's actually allencompassing because uh a
person's second brain that's anevolved person, say a diplomat,
um, does a lot of differentthings.
So you have to connect a lot ofdifferent dots for that person.
Um, but from a workflowstandpoint, I've done lead
routing, uh, which was honestly,it was easy.
(32:19):
It was an easy problem for whenthey started talking, I knew
right away, okay, that's theproblem.
We need the lead routing to befixed.
And all it was on the salesside.
They couldn't keep theirterritories and their data and
Salesforce up to date.
And it was honestly, I createdan SOP, uh, I created an ideal
(32:41):
framework for it.
This is how we want it.
And they built it.
And then we cut lead routingdown from 48 hours to four
minutes.
Um, and then I did somepredictive lead scoring uh for
somebody, but that's on the mopside.
And I've other done otherboring things um that people
might find are boring, but it isum just workflows and filing.
(33:06):
Uh, you can build an ontologywith a workflow now.
Like to do it right, you'd haveto look go say you had a
thousand documents.
Well, someone would have tophysically go through all
thousand and apply an ontology,relate it, build the triples and
stuff like that.
So there's workflows.
Michael Hartmann (33:23):
So sorry, you've used the word triples now
multiple times, and I I wasjust gonna assume that it's a no
I don't know what you mean.
What does that mean?
Tracy Fudge (33:31):
So it is meaning that um Johnny kicked the ball
in the street.
You got Johnny ball street.
Michael Hartmann (33:39):
Okay.
Tracy Fudge (33:40):
Well, someone else could have kicked the same ball
into the ocean.
So it's it's the it's how it'srelated in triples, three
things.
And so long as you can matchthe three things, then
everything you can see howeverything is related.
Uh and I'm not doing a greatjob of explaining it, but you'd
have to see it.
(34:00):
It's just applyingrelationships by three.
Michael Hartmann (34:04):
Okay.
Interesting.
Okay.
I mean, that makes sense.
I mean, triangulation on stuffkind of makes sense.
Okay.
I'm gonna be uh you're gonnasee my heads like you're gonna
see my eyes going back and forthbecause I'm still thinking
about this.
Keep going.
I'm sorry.
Tracy Fudge (34:19):
No, that was it.
It is, and I'm not anontologist, so there are people,
you know, as I stepped intothis world, I'm in the other
cohorts and taking more classes.
Um, you know, I am I'm I'm kindof addicted to taking classes,
um, to be honest, because Idon't want to do it myself.
I want to sign up for a classthat I've invested in and it's
got beginning and an end date,and it keeps me accountable.
(34:41):
Plus, you meet people along theway, right?
Um, which is a great way.
And and community andnetworking is gonna matter now
more than ever.
Um it's gonna matter now morethan ever because it's gonna be
who you know.
Michael Hartmann (34:53):
Sure.
Yeah.
Tracy Fudge (34:55):
And at my point you are how many degrees you have.
Um, so I don't know if Ianswered that question.
Michael Hartmann (35:02):
No, I think I think I think if you did, I
think I'm trying to think aboutlike if I was someone listening,
watching whatever, and um Ididn't have my own side
business, what could I do?
And I think about some of thethings that I've done, because
most of mine has been uh more ona personal level.
So interesting.
So I've done automationactually related to the podcast,
where um I have now a link forpeople to schedule this
(35:25):
recording, right?
And part of what I have done,this doesn't involve AI at all,
which is interesting.
But I did play with anautomation platform that goes
this looks for when uh that newthing is added to my calendar,
it goes looks for new ones andit does a couple of things with
it, right?
Adds a couple of people to it.
Um I've been thinking aboutlike how could I take that even
a step further?
So I still do a manual effortof prepping.
(35:48):
So I've taught uh in this caseGPT, uh how I like to do the
preparation document.
So for people listening, right,we actually do a fair amount of
preparation for this.
It's not a total wing-it thingas much as we try to make it
feel organic.
But you know, that's still afairly manual process based on
some.
(36:08):
You've everybody knows we wehave initial discussions about
these things, and and uh youknow, I've got recordings.
So I would love to get theprocess if I could set set it up
where like it can go back andgo find my notes and transcripts
from our planning discussionsto generate the document, right?
That would be an automationthat would would benefit for me.
This still doesn't have takenme as much that much time
(36:30):
manually, so it hasn't been atop of mind.
But I have done other things umthat are more focused on still
manual but recurring things thatI'm doing, or you kind of
hinted at this, like doingresearch now.
Um, I find it really, reallyvaluable in doing research on
stuff.
Um to put like I literally hadtoday, had people here, service
(36:54):
repair people.
We have an old uh cooktophaving problems with it.
And so I did a bunch ofresearch on is this stuff I
could fix myself, right?
Did it I did it with uh we hada deep freezer that went bad
over the summer.
Um, and I did a bunch ofresearch like what are the
possible causes and is otherthings that as someone who's not
super handy could could fix.
(37:16):
And it's a it it made gave me asolution that was a $20
solution as opposed to a replaceit for a few hundred dollars,
right?
Which is beneficial.
And I probably could have donethat research on my own, but
having the LLM do a bunch ofthat for me, come back with
stuff, and then I could refineit.
(37:37):
Um I've even done stuff ifyou're a kind of person who
likes to cook at home.
We took photos of all the stufffrom the pantry and said, come
up with some recipes.
Take a picture.
That's what we did.
Yeah, it's really actuallyquite good at that.
I found it's not so good atgenerating photos, but it's or
pictures, but it's really goodat deciphering what's in photos.
(37:59):
I I've been sort of blown away.
I didn't believe it would workwell, and it's been really good.
Yeah, so we actually did thatand it generated a number of
recipes, and we tried one and itwas terrific, right?
No, no, will everything be thatway?
Right?
Tracy Fudge (38:11):
I doubt it, but um so I think walking into your
kitchen and all you have to dois talk to your kitchen, yeah.
Michael Hartmann (38:22):
So I think I think I wouldn't yeah, go ahead.
I think if people think aboutlike how could I try to learn
some of the stuff even if Idon't have a site, like it
doesn't have to be a businesscontext.
I think you can learn a lotfrom doing things, and there are
like I've played withconnecting like my email to one
of these automation platformsusing an AI kind of engine to
(38:43):
automate drafting replies andthings like that, or
categorizing things as importantor not important.
Um honestly, I'm a little bitof a like I'm paranoid and like
just like the amount of accessthat those tools currently have
to have to things like yourinbox, not comfortable with
personally.
So, but I know it's possible.
(39:08):
Like, and it's definitely uhlike that's a very common use
case that most of theseautomation platforms will give
you as a starting point.
Like open up your email,classify the emails, draft the
reply to those ones that youthink are important, and then
you can go in and and then sendyou a summary on a daily basis.
Pretty common thing you coulddo.
And um, like I think there arethings that people, if they
thought about it, even ifthey're not doing a work-related
(39:30):
one, because there are concernsabout you know uh security and
proprietary, you know,confidential information, things
like that, that would probablymake it harder.
Um, but there are ways you canlearn.
So I think that would be highencouragement is to do that kind
of thing.
So I oh I get so one of thethings you have I think
(39:53):
encouraged me, um, and I haveyet to to really do that, and
I've already mentioned rightthat I have a bias towards Chat
PT, GPT because I'm mostfamiliar with that, and then I
particularly like um Grokbecause you can put on the uh
what do they call it, a persona,I think?
You can do the unhingedcomedian one, and it's like I
(40:13):
just that's more of anentertainment thing than
anything else, just to see whatit comes up with.
But um how do you like whatwhat could you share about what
you know about some of the let'skeep into the LLMs with this, I
guess at this point, like whatlike what do you see from them?
What do you think that they'relike what are their strengths
and weaknesses?
Because it feels like they allhave sort of various things, and
(40:35):
I know it's evolving, so likeby the time this gets out, it's
probably gonna change a littlebit.
But in general, what's yourtake on those?
Tracy Fudge (40:42):
I um so I I don't really use grok.
I have in the past, but not tothe extent.
So my main ones are ChatGPT.
I just pay the 20, and thenClaude.
I I think I did bump it up tothe higher one because Claude um
has some memory and contextissues, and I hate having to
start over in the middle.
Michael Hartmann (41:01):
Uh yeah, yeah.
Tracy Fudge (41:02):
So, which is very irritating.
Um, and then Claude'slimitation was not being able to
hold the context inside offolders or projects.
You know, you can createfolders.
So I have a I have a hugefolder structure, and I was
actually arguing with ChatGPT.
Because it used to be able todrop into the folder and ask a
question and uh about anythingin that folder.
(41:26):
Well, today I found uh that itdidn't it didn't have the
context of a conversation that Ihad just had in that same
project.
Michael Hartmann (41:34):
Yeah, I've found that I've I've had to give
it explicit direction, go tothis other chat that's in the
same folder.
Tracy Fudge (41:41):
Yeah, and that was not the case.
So I don't know what's goingon.
Maybe they realize that theylet too much out with chat GPT
that this is a lot of powerrequirement to make your, you
know, the return of your requesthappen.
Um but I use both.
I actually um I intentionallydid not do Claude Desktop until
(42:04):
a month ago or two months agobecause I didn't want to get too
advanced because I knew thatthat would then uh I would build
MCP servers.
So I've evolved into ClaudeDesktop, MCP servers on my on my
desktop to go into thesesystems uh for me.
But that's not for everybody.
(42:26):
You still have to know what thehell you're doing.
Or what it's doing, you know.
Michael Hartmann (42:31):
Yeah.
Tracy Fudge (42:32):
Build, you know, build the JSON to be, you know,
uploaded into, you know, uh N8N.
I mean, you've still got toknow uh a few things about that.
But that that's what uh and Iuse Claude for that.
So but I still go back to chatGVT for creative.
I've I've started to write moreoften.
Um I love to write and I'vestarted to write more often on
(42:55):
LinkedIn and I've done it.
Michael Hartmann (42:56):
Do you find it do you find it better at
drafting initial stuff with alot of context or you drafting
stuff and now asking it to umgive you feedback on it?
Tracy Fudge (43:07):
I've been training it for a while.
Um, and assuming it's still gotthe same project context.
That was one of my things I wasfrustrated with uh because it
has my brand voice, it has mylast post, it knows sort of
where I left off last week.
Right because I get one loud.
I don't, I I I took last weekoff and I was like, I'm not
going on LinkedIn.
So I took a break.
(43:28):
Um, but it's still so that forthat I use it.
Um uh I use Chat GPT for that.
But Claude, I use for thetechnical stuff.
But Claude's gotten better uhwith the creative and the
writing um as well.
And my you know, there's somuch noise, and I know everyone
(43:49):
says this, there's so much AInoise, and now with all these
layoffs and things, and I just Iwant to post something that
matters.
It's just not more of an echoof what's already and sometimes
that's hard.
Michael Hartmann (44:02):
Yeah.
Tracy Fudge (44:03):
Because you want to call out the noise for what it
is, right or wrong, and then youknow, have your point of view.
And it's it's hard to get apoint of view sometimes when
when it's so noisy out there,you know.
You don't want to, you know,it's hard to not be afraid, you
know, these types of things.
Michael Hartmann (44:19):
So yeah.
It's interesting because likemy biggest, my latest challenge,
even though I've got somefolder structures inside
ChatGPT, because I'm also at apaid level, I think that's a you
have to have at least the basiclevel folders.
Yeah.
Um is that even within that,I've got like I'll I I I
(44:39):
actually did something with mywife too, for some she does in
her work, which there's ongoingdoing a lot of the same things,
and so having context is reallygood.
What I found is when I it goeson for a long time, it um chat
PT GPT tends to lock up, right?
If I if it's if the chat threadgets too long, and so that's my
(45:00):
latest one is like how do Iretain what is like I did a lot
of work training it to voice andstuff like that.
Um but the performance on newasks is still there, and that
that's my latest challenge.
But um it's what's a screenedthis week.
Tracy Fudge (45:17):
I don't know what it is.
Michael Hartmann (45:18):
Well, I've been noticing it for a few
weeks, so I don't think it'sanything specific to this week
um for me, but um and and Ithink it's just just that,
right?
The the threads are just solong now that uh it may just be
browser, right?
The browser rendering it takestoo long and it's just locking
up.
So uh but again, like this islike part of the learning
(45:39):
process that we go through, andit it would help me knowing
like, hey, if I'm building somesort of automation for a work
context and um I wanted thecontext to be there and
available over time, right?
I need to think about how tostructure it so that I don't run
into the same thing, anautomation step where I'm like,
hey, asking a whole bunch ofstuff to a say uh uh LLM engine,
(46:01):
and uh because it's like ifit's building it out over time,
is that gonna cause that to slowdown and not run correctly?
So things like that, right?
So I've kind of tied tying thattogether.
Um let's let's let's wrap uphere because you kind of alluded
to this.
Like, how how how can peoplethink about what's the right way
(46:25):
to start to incorporate AI,LLM, you know, automation slash
agentic stuff into their, let'skeep it to Mops, RevOps,
whatever, uh, into their worldwhere it's it's a complement uh
or a um I like the term I heardsomebody use, like it provides a
(46:46):
time dividend, right?
To like into their theirday-to-day um work effort,
right?
How did like what are some waysthey can do think about doing
that?
Tracy Fudge (46:56):
So are you asking from the perspective of the mops
individual or mops?
Michael Hartmann (47:02):
I think I think mops people as
individuals, right, but as partof a team, even if it's a team
of one or two people.
Tracy Fudge (47:10):
Mm-hmm.
Content is a big thing.
Um, and I think content'sreally relevant.
Uh, but you can't just put itinto Chat GPT and say, write an
article on this topic.
I mean, you've got to iterateit, you've got to give it
examples, you have tounderstand.
But I think getting content outfor the team um for um, you
(47:34):
know, email is another one.
I have not ventured into emailfor exactly what you said.
I want to go in and read myemails and things like that.
And I'm scared to, and I I useuh Google Workspace and I'm
scared to turn on Gemini and myemail.
I don't know what it's gonnado.
Right.
Um, but I do think that, youknow, find a few workflows that
(47:56):
are for you and build those out.
Uh for the business, if it's ifit's easy to show a return,
then do it.
And you know you could show itbecause you want to show ROI.
And that's gonna elevate you toleadership as well.
Look, I I shortened your leadrouting.
We could do this.
(48:16):
Look, let's um, and do somepredictive analytics, pull some
data out of sales if you can,you know, the last four
campaigns.
Uh, what else did those peopledo that converted to a sale?
Did they visit uh a certain andbesides the webinar, did they
do these other things?
Michael Hartmann (48:37):
Um looking for pattern, like looking for
patterns tied to exactly uh thedesired outcomes.
Tracy Fudge (48:44):
Most companies aren't capturing those events,
right?
So you've got to figure out,okay, can I pull in data on the
white paper they read, not justthe webinar they signed on?
Can I pull uh and we alreadyknow what emails are opened?
Michael Hartmann (48:57):
Um those you could provide it, you could
provide it not only the data,but also the content, right?
The examples of the content.
This here's this content, it'stied to this data point or this
kind of activity that's in thedata versus this other piece of
content.
And um you could then tell itto look for what are patterns in
(49:18):
the content that uh are tied toum better conversions, more
pipeline, like whatever that isthat you're trying to optimize
for.
Tracy Fudge (49:31):
Possibly look at the last thing they did, like
they looked at an ROI whitepaper or something.
Let's just make that up.
So then you have the AI notethat, right?
But then craft a draft emailaround RLI.
And it's something they justdid, so it's fresh in their
brain.
It may not be the thing thatpushes them over the to be a uh
(49:52):
cli uh a paying client, but it'ssomething that's relevant.
Um, that's something that a lotof people uh could easily do.
Michael Hartmann (50:01):
Yeah.
I mean, I think I think what Iwould do, like if I was in an
organization that had likeproduct marketing and or just
content marketing teams, and umwe had a sales team that had
some sort of call recordingplatform, I would be like, that
sounds like such great contextif you could capture the
transcripts from those salesconversations, then tie it back
(50:26):
to which of those and ended upturning into deals.
Mm-hmm.
And capture the language that'sin there, both client side and
sales side, to feed back intohow are you thinking about the
content you're publishing andthe material you're providing to
the sales teams, right?
Because you because like Ithink there's so much crap out
there on B2B websites inparticular that is like just
(50:50):
gobbledygook nonsense thatdoesn't really tell anybody
anything about what you actuallydo.
Like if you could drive, likeif you could do that, take that
back, feed that back into yourcontent engine, um, which then
hopefully is gonna drive moretraffic and awareness, right?
Like you could build a prettystrong flywheel, um, I think in
terms of that that's just simplybased on what are the actual
(51:11):
terms and uh things that uhcustomers, prospects actually
care about, and make sure thatyou've got that really well
honed on your public-facingwebsite and the content that
goes with it.
Like to me, that's like it'sit's not trivial, I get it, but
like it talks about like uh anROI over the long run, both
(51:32):
short term and long term.
I think that that one to me isa use case that I would love to
see more people trying out.
Tracy Fudge (51:38):
And then you could retarget people to go to that
new dynamically created blogpost that was created, you know.
So these things are dynamicallyum done because there's APIs in
there that are doing it.
Yeah.
So there's the API work, butabsolutely.
Michael Hartmann (51:53):
Yeah, so anyway, well, hey, I it feels
like as so many of these arethat we could go on and on and
on.
Uh, but unfortunately, we dohave to we do need to wrap up.
So, Tracy, first off, thank youfor sharing.
It's been a lot of fun.
I enjoyed the conversation.
I'm so glad we were able tomake this work uh while everyone
else is out there having a funtime at Mopspalooza.
(52:14):
But if uh if if folks want toconnect with you or learn more
about what you're doing or godeeper on this topic with you,
what's the best way for them todo that?
Tracy Fudge (52:24):
Uh find me on LinkedIn.
Um, my website's AI by Thrive,and I'm Tracy with an EY at AI
by Thrive, if they won't emailme.
Michael Hartmann (52:33):
Perfect.
Well, thank you again, Tracy.
Appreciate it.
Thanks always to our listenersand now watchers now that we're
uh going live with videos.
Well, we appreciate yoursupport.
And as always, if you haveideas for topics or guests or
want to be a guest, you canreach out to Mike, Naomi, or me.
We'd be happy to talk to youabout it.
Until next time.
Bye, everybody.
Tracy Fudge (52:54):
Thank you, bye bye.
Hello everyone, welcome to another
episode of OpsCast, brought toyou by MarketingOps.com and
powered by all the MoPros outthere.
I am your host, MichaelHartman.
Flying solo as a host this weekbecause we are recording this
during Mopsapalooza 2025.
So if you're there andlistening to this later, we hope
you had a good time there.
(00:45):
Uh, Mike and Yumi, I'm surewe'll be back soon.
In fact, we've got planned arecap, so watch for that when
coming into your inbox.
So today I am joined by TracyFudge.
She is an AI operationsarchitect and agentic workflow
designer, and uh, she's doingthat on her own with her
consultancy AI by Thrive.
Tracy has spent the last fewyears deep in the AI trenches
(01:06):
experimenting with languagemodels, automation frameworks,
and what she calls agenticworkflows.
She helps marketing andoperations teams not just
understand technology, butactually put it to work.
So today we're going to betalking about how to move beyond
the AI buzzwords, not thatthose are out there, to build
real usable systems that savetime, spark creativity, and
drive better business outcomes.
So, Tracy, welcome to the show.
Tracy Fudge (01:29):
Thank you so much, Michael.
I'm really excited, and I'm Iam, I'm I'm missing MAPSA, but
uh I'm there virtually and in myheart, a lot of my friends are
there.
Michael Hartmann (01:39):
Me too.
I have serious FOMO going onright now.
Well, I'm glad, but I am I'mgrateful that we've been able to
stitch this together.
I was going back to some of theconversations that you and I
had to try to get this started.
It went all the way back to itis now late October.
Uh, and I think it's we startedin February sometime.
So um good things come to thosewho wait, right?
Tracy Fudge (02:02):
Yes, exactly.
Michael Hartmann (02:04):
All right.
Well, why don't we let's startwith this?
Like, let's start with maybe alittle bit of like a short
version of your journey.
Um in particular, like the lastcouple of years, like what led
you from marketing andoperations, uh, kind of the
traditional one we were justchatting about this before we
started recording about how somany people are still focused on
platforms, um, and getting youinto AI systems, agentic
(02:26):
workflows, things like that.
Tracy Fudge (02:29):
Yeah, so it was it was an easy transition.
Well, one, I've been in tech mywhole career.
Um, you know, I started insoftware sales uh before I
really understood what softwarewas.
You know, my explanation was,well, what's software?
And I was like, I don't know, Ithink it goes on hardware.
So that's how far back it goes.
Michael Hartmann (02:46):
You're not gonna start talking about floppy
disk drives, are you?
Tracy Fudge (02:49):
You know, there was a day when we when software was
a new word.
Um, and you know, so then fellinto mops.
Um, you know, I was managing uhsocial media platform for
somebody, and they're like, hey,can you go manage this email
platform we have?
And that's sort of, and mostpeople fall into mops by
accident.
I'll start.
You know, there's no collegecourses or anything like that.
Michael Hartmann (03:11):
Not yet.
Tracy Fudge (03:12):
Not yet, yeah.
But the um the the AI part ofit was just something that was
just intriguing to me.
And you know, systems areconnected via API.
So the agentic part of it wasalready being done.
We just weren't calling itthat, right?
So when I um decided it to godeep and really learn, you know,
(03:34):
on what agentic platforms were,you know, I had known Zapier
and Zapier, you know, nobodyreally knows quite it.
Michael Hartmann (03:41):
They told us that we've had people on from
there.
It is Zapier, like happierobviously fingers.
Not that we're like, we're notthis is not a promotional thing,
but I think I now have thatthat rhyme in my head.
Tracy Fudge (03:54):
And I don't know how far back you've been using
it, but I was using it beforepeople did correct me on what to
call it.
Um, and uh so that's been gosh,and I mean well over 10 years.
But um anyway, the the whole AIpart of it, the agentic part of
it was sort of an easiertransition to sort of understand
(04:15):
what this all meant.
And obviously I was curiousabout well, what's an LLM?
And oh, there's other types ofLLMs.
Oh, what do these frontiermodels mean?
And then putting them alltogether and making the
intelligence layer over theconnections.
Um, you know, and that's justthat's much more complex than
(04:37):
people realize because you gointo the conversations around
data.
And when you're inside of anorganization, sometimes
marketing doesn't have all thedata or the systems.
You know, I it systems sit overthere.
So that's a big uh gap that'sstill there.
And teams are starting to cometogether and realize that AI
kind of goes over the wholething.
Um, and AI needs to work on thesame set of data, uh, not the
(05:04):
same fields per se, but let'sall talk about the same data.
Let's make sure that thepipeline reflects in the CRM, et
cetera, et cetera.
Uh, so that's really wherewhere I got into it.
And it's it's still a new, youknow, people uh frankly don't
understand it.
They think it's just workflows.
Oh, just flow build sixworkflows, and that makes it,
(05:28):
you know, and that's not iteither.
Um it's part of it, butcertainly when you're looking at
a business, you've got toconnect it all together.
Uh and finance needs to havethe same initiatives as RevOps,
as Mops.
I mean, it it needs to all flowtogether.
Michael Hartmann (05:42):
Yeah, it's I think that's the that's the uh
the goal, right?
The only thing you you saidsomething that uh marketing
sometimes doesn't have all thedata.
And I I I would push back a biton that and say that I think
most marketing teams have moredata than they know what to do
with, um, and probably lackskills and knowledge to be able
to do any much with it is islike to me that's the bigger
(06:05):
gap.
It's not data itself, but theability to understand it.
Tracy Fudge (06:10):
Yeah, um, and I think you I would say that you
are correct in that aspect.
When I think of data, I meanthe the ability to action on it,
to add some predictiveanalytics, right?
Uh to be able to hold the datain a sandbox and just kind of
play around with it, apply somemachine learning, some
prediction.
And that way your leadscoring's a little bit better.
(06:30):
But if you're still sayingbecause person A watched Webinar
X, let's give them five points.
And it doesn't take intoaccount all the other things
that said prospect or lead didalong the way.
Michael Hartmann (06:44):
Sure.
Yeah.
Yeah.
That's interesting.
Yeah.
Um well, so you you kind ofhinted at this when you were
talking about this, but yeah,something you said to me, which
I've been thinking about, isthat you said AI isn't really
new, that it's yeah, automationand data taken to another level,
I think is the way you put it.
Like, what do you like?
What do you mean by that?
Break that down a little bit.
Tracy Fudge (07:05):
So hooking up, let's say HubSpot to Clay or
HubSpot to Salesforce, thathappens for the most part, out
of the box, right?
API, API, field match, fieldmatch.
Um so that's already sort of aningenic portion connecting
APIs.
I think uh, you know, when youadd in the intelligence layer,
(07:29):
for instance, if you were todrop the data that's moving over
to the marketing automationplatform, drop that into a
knowledge play, uh, an LLM nodeto do something with the data,
make decisions, apply tags, dowhatever.
And maybe that when you drop itinto say N8M, and I'm just I'm
making I'm making reference tothings that we all know or to a
(07:51):
Zapier.
But the point is thatconnection doesn't have to be
direct anymore.
You pause it, you do thingswith it, and then you map it.
Michael Hartmann (08:00):
Right.
Tracy Fudge (08:00):
Um, and then you iterate, and that mapping can
change.
So it's that part of it that isnew, but the actual technical
API stuff, even installing aplatform like an N8M, things
things are so easy, uh easier tounderstand once you've
(08:21):
developed on one platform, youcan do it on another.
One CRM largely looks likeanother, you know, minus some
capabilities, but the conceptsare the same.
Michael Hartmann (08:33):
Yeah.
It's it's interesting becauseyou um it feels like well, one
thing is like some of theseterms get thrown around as if
they're the same, right?
And I like in my head, I tryI've started to try to break
down like um AI does notnecessarily mean automation,
(08:56):
right?
Uh AI also I think generallygets conflated with LLMs, where
it feels like there's a wholelot more that's under the AI
umbrella.
LLM is sort of one type of AIum modeling kind of thing,
right?
Where you've got you'vementioned machine learning.
I think predictive analyticscould be a part of it, or it's a
component.
But it feels like people kindof mix those up.
(09:18):
And to your point, I think somepeople are just still have
maybe not blinders on, but liketheir their focus is on like
small relatively like let's dothe replace this small thing
that we're doing with somethingthat has AI as in quotes about
it.
But um when they start to openup the the aperture a little
bit, they start to getoverwhelmed, right?
(09:39):
So for those people who arelike, hey, do you feel free to
(10:44):
disagree with my my assertionthere about like the mixing of
terminology, which is I thinkvery common in a scenario like
this, but also like anyrecommendations for those people
out there in marketing,marketing ops who are maybe
feeling a little overwhelmedabout like where to even start
with this AI journey, if youwill?
Tracy Fudge (11:05):
Uh well, most people that I know that did it,
let's say within the past threeyears, because that's really
where when I went in.
Michael Hartmann (11:14):
We it's crazy that it's been three years for
stuff, Rick.
It feels like it was just likethe other day.
Tracy Fudge (11:20):
I um I actually started, there was somebody
something kept coming in myfeed, right?
It was this no code, low code.
And I was like, okay, thismight be part of it.
And I think um that is easierto start because I'm not a
software developer.
I can I've done Python, youknow, I did a serious analytics
(11:41):
course where I had to writePython code, but then there's
stuff that does it for you.
So, you know, that sort ofskill set goes away.
But the the whole idea of um AIjust being connections, it
could be um, you know, an LLMplay.
But in terms of starting, Idon't start with the start with
(12:03):
the low-code, no-code platforms.
And I personally started onMaven and I stayed there for the
whole time.
The the courses uh people thenuh did more courses and more
courses.
And plus you're with a cohort.
Michael Hartmann (12:17):
And I think um Maven is a like a training.
Tracy Fudge (12:22):
Yeah, uh not the platform.
Okay, uh I'm just saying Mavenwas where I started.
Michael Hartmann (12:28):
Sure.
Tracy Fudge (12:29):
Um and they do uh, or at the time were really the
only ones giving true how tostart with AI classes.
Okay.
Um I think that that's one wayto start, just real basic.
What is AI?
And now, fast forward two and ahalf, three years, you could
(12:50):
get the same thing on a YouTubevideo.
I mean, honestly, just break itdown.
I think the hard part is this.
Our background is marketing andmarketing apps.
And so we're flooded withplatform.
Oh, Salesforce has AI.
Look at these Salesforceagents, right?
Spot has AI.
Oh, everything's got AI.
(13:10):
So I think there's a falsesense of being AI when it's not
AI.
It is just the platform'sversion of AI.
You have no control over theprompt, uh, little control over
the knowledge base, um, and it'scontext, and you don't know
what model is in there.
Um, a true AI person that wantsto add the knowledge layer has
(13:35):
all the levers to temperatureand knowledge base and prompting
and system instructions andprompt chaining, and that's the
limitation.
So you might, and then anotherpart of it is some of the AI in
these things, it's just fillingout fields, right?
It's um it's just doing, oh,that field needs this data or
(13:57):
that field needs this data.
And you can do that now with alot of platforms, you know,
Airtable off is their uh AIcomponent, Zapier's got one,
they've all got one, but it'skind of filling out the
architecture for you, which ispart of the part of part of what
AI is.
But once the scaffolding isbuilt, where's the knowledge?
(14:20):
You still have to create that.
And that cannot be done yetwithout a human that knows what
the hell they're doing.
Yeah.
I think unless you've youunderstand data, and I I always
I've always said this is a dataproblem.
Um, understand not just wherethe data is, but what the data
(14:42):
is, uh, what data matters, uh ifthe data changes what's the
story it's telling.
Michael Hartmann (14:47):
Yeah.
Tracy Fudge (14:47):
Exactly.
And we're not having enoughdata conversations.
Michael Hartmann (14:51):
Um well, they this goes back to like one of my
absolute strong beliefs rightnow is that we just lack in
marketing in general, in ops inparticular, um the skill set to
really do much with data, right?
I think there's just not a deepunderstanding about how to do
it.
Um I want to go like somethingthat's just sort of clicked with
(15:14):
me that I think maybe is partof the uh the challenge for some
people in this, in that we'reso used to having what I would
call, I guess, deterministickinds of solutions.
So, you know, I get a bunch ofjob titles, I'm gonna put
something in there that I definethe rules to say this is the
job level, right?
So that which is sort ofinferred from the titles.
(15:37):
And yeah, I think an a veryobvious use case for probably
LLM in this case, right, as partof a flow is to use that to
categorize jobs, people with jobtitles into categories that
make sense and use someintelligence on that, maybe with
a little bit of humanoversight.
But I think, yeah, what youdescribed, right, in some of
(15:59):
these other ones, there's alittle bit of this trade-off,
right?
You could either do thatyourself and you have a little
more control over what'shappening inside the black box,
or you take advantage of some ofthese tools that are being uh
embedded within other platforms.
You mentioned a few, right?
But it becomes even lessapparent, like what's happening
within that black box.
And so there's this lack of, Ithink there's this like, do I
(16:22):
can I trust it?
Can I not trust it?
Yeah, how can I tweak it?
Can I tweak it enough?
And it feels like maybe that'spart of the like the resistance
for people who are strugglingwith it, where they, yeah,
you've got less like, oh, I wantto take advantage of the stuff
that's being provided in theplatforms that I'm used to
working in every day, but itdoesn't really give me, like, I
don't know enough about it.
So somebody starts questioningit, I can't answer it, which
(16:43):
puts them in a position of notfeeling comfortable about it.
And then at the other part islike the other struct struggle
is like I still have to, like,how do I fit in time to go
beyond that?
And that's I think it feelslike that's maybe the rub that's
the or the the the the thingsthat are making it difficult for
some people to go, like, howcan I step in and take a step
forward in this?
So does that sound right toyou?
Tracy Fudge (17:06):
Yeah, it's a gap.
And I have friends at verylarge organizations, and um,
they don't have time to go learnthis.
And I think when you're a W-2,uh you don't have a lot of time.
We're all overworked.
I had, I chose to take timeoff.
(17:26):
Well, I have a side business,and that's what's been
supporting me, to be honest.
Um, but I've taken uh a lot oftime and taken very uh a few
projects, but it's not, I'm notworking 40 hours a week for
someone else.
Uh, it's probably 20, and thenI work the other, you know, even
(17:47):
weekends to learn this stuff.
And I think it is it's a realum miss of an opportunity uh
because you've got really smartpeople that are working really
hard in these huge organizationsthat are not getting upskilled
in this.
Because you and it's and here,here, here's the other element
(18:11):
to this.
When you go in and design aworkflow, it's not something you
touch today for an hour, youcome back next Tuesday and you
finish it.
No, it's deep work, it's 12hours sitting down, getting up
just to go pee or eat and doingit end-to-end.
And there's no, it's heavy,deep can't outsource that part
(18:32):
to AI, right?
You can't.
And it is you you've got toremember where you are and
you're deep into that.
You're living inside theworkflow.
I mean, I dream in spreadsheetsand workflows because you know,
and not everybody has thatability to say, okay, you wake
up at seven and you realize thatit's you're gonna work on this
until seven at night, and youdon't stop.
(18:54):
You don't answer the phone, youdon't, you don't get on social,
nothing.
You're just inside thatworkflow.
Um, and hopefully in that timeyou you have enough good data,
good parameters to test it andit worked out.
But some of that can just be,you know, beating your head
against the wall to try to makeit work.
And um, you've still got to getit reviewed for the client, you
(19:17):
know, for the client or foryourself or whoever to see your
art because you're the one thatdesigned it.
Michael Hartmann (19:24):
And here's the that's the main right, and
that's hard to to test.
Tracy Fudge (19:31):
Nobody's done this before.
Michael Hartmann (19:32):
Well, and because if you're using so in my
head, one of the distinctionsI've started making is there's
the AI comp AI can be uh thereare AI components could be used
in some sort of automation orworkflow or agentic model,
right?
So pick the tool.
But there are other like thereare other more deterministic
(19:52):
ones.
And I think what again I getback to like when you're trying
to validate, right?
It's it becomes less of a umdid it do what it was supposed
to in in an exact way, or did itproduce an output that I is
expected even if I don'tunderstand how it came to that?
And that's a very differentvalidation.
And and then if you're notcomfortable, if you're the kind
(20:13):
of person who's not ascomfortable with these sort of
um what's wrong, right?
Like less, like unless like,oh, this is this directionally
correct, right?
If you're comfortable with thatas a quality assurance um level
of quality or completeness,then you're gonna really
(20:35):
struggle with this stuff, Ithink, right?
Because a lot of times whatI've seen, even when I'm just
using, say, uh an LLM, Chat GPT,Claude, whatever, right?
The output I get from my backand forth in terms of prompts,
right?
Sometimes I go, I could Iscratch my head, I go, like,
where did it come up with that?
Right.
It we've all heard about thehallucination thing, but also
(20:58):
sometimes it just doesn'tfreaking follow the directions,
right?
I remember working on somethingwhere I was like, I want you to
draft a document.
I had we'd gone back and forth,I had a lot of stuff, dropped
this document and came up withsomething.
I was like, I like that, but Idon't like this one piece of it,
right?
It was like I think I had donea header image or something.
I said, Yeah, just go fix that.
And it completely redid thewhole fucking thing.
Tracy Fudge (21:19):
So yeah, and I was like, no, like that cost us
down, right?
Like that that you can't putthe whole prompt in one place.
Michael Hartmann (21:28):
Well, and I hadn't, like I had gone, but
like I had produced somethingthat was probably 80% of what I
wanted.
It was in line.
I'd gone through like as manysteps.
It was just like this last one,like the formatting, the layout
of this document wasn't what Iquite wanted.
And so I asked to make onesimple change and it completely
like it redid the whole thing.
(21:49):
And I was just couldn'tunderstand why.
Tracy Fudge (21:51):
Yeah.
Michael Hartmann (21:52):
And that's the kind of stuff.
So if you take that, right,where I'm like going back and
forth and it took me time to doit, and you say, like, I'm gonna
do something like that, maybenot exactly, and I'm gonna be
make that as a part of anoverall flow, then like, how do
you test that and evaluate it?
That's like that's where theuncomfortable like people get
uncomfortable with how do I fitthis into something where people
(22:13):
need to trust the overalloutput of that, not just one
piece of it.
Tracy Fudge (22:20):
Yeah.
I was introduced to a platformcalled Cassidy.
Um, it is it's a pretty coolplatform.
I use it mainly for uh doc gen.
So creating um a workflow tocreate an RFP process or a
workflow to create uh apresentation or talking point or
(22:43):
or something like that.
It's to do one thing is eightdifferent agents or assistants,
right?
One of which is the JSON, andthe JSON is gonna be how that
document is formatted, and youtack that on to the end and you
go, okay, I want the header tobe this,
da-da-da-da-da-da-da-da.
(23:04):
And then and then it goesthrough that node, which is a
code node, um, and then it'sformatting.
And that uh had I not gonethrough Cassidy workflow on a
simple doc generation thing,yeah, I wouldn't have realized
that no, no, we we first have tohave a research agent.
(23:24):
Oh, and then we oh, first weneed to have the brand voice
agent, you know, or wherever youput that in, how do I sound?
So you've got eight differentassistants to do one eight-page
output.
And I know that sounds possiblygratuitous, but it's how you do
it.
Michael Hartmann (23:44):
Yeah.
Well, I think so going back tomy original question, to you,
like, where should people start?
It feels like the way to startis because you can't do all this
in one sitting.
If you're a kind of personwho's busy with their day job
and you can't do it all in onesitting, you could start to
experiment with pieces of whatmight be a bigger process,
right?
(24:04):
And you like figure that out.
And then because I it I I spenta lot of time trying to learn
how to just use I have a biastowards ChatGPT, not for any
good reason other than it's justthe one like was one of the
first ones.
I've spent more time with it,and I'm open to the other ones.
But I've done lots ofindividual things.
It wasn't until I'd been doingthat for a while where I sort of
(24:26):
lifted my head up and said,okay, I keep hearing about this
agentic right in quote stuff,you know, what does that really
mean?
And I played around withdifferent tools.
And I don't think that if I Ithink if I'd started with the
N8Ns of the world, or like I II've had pretty good luck with
relay.app, right?
There are others, I know thereare plenty of others, but if I
(24:48):
had started with those asopposed to learning how the LLMs
work, I think I would havestruggled even more because I
would have tried to do what yousaid, like I would have tried to
do like a whole bunch of stuffin one uh AI node of a process
as opposed to breaking it downinto smaller chunks, which does
feel on the surface to beinefficient, but it feels like
(25:11):
like now I understand that uh inorder to maintain a level of
quality or consistency, that'skind of what you have to do.
Tracy Fudge (25:21):
Yeah.
Michael Hartmann (25:21):
Which means then you have to understand like
which it should be a strengthof ops people, right?
What's the process that I gothrough?
That what's the well, how can Ibreak down this overall process
that I want to automate, haveagents do into smaller steps so
that I can trust the output atthe end.
Yeah.
Tracy Fudge (25:40):
Yeah, it's it's and you evolve to that point.
We all start, we're alldifferent learners.
And for me, I like being partof a cohort, a learning cohort.
And I didn't want to, I kind ofpaused in my mops world
intentionally because I didn't,I felt like mops wasn't where I
wanted to go.
So there were a few mops peoplein the community where I where
(26:04):
I ended up, uh, and I still am,but not many.
It's CEOs are in there.
You know, it's just differentpeople having different
questions.
And I personally would beexposed to more use cases than
just hooking up a CRM to amarketing automation plan.
You know what I mean?
Like I it and that's wherethat's where um so I went, you
(26:29):
know, that's where the decisionI made to get out of the mops
mentality and the mops worldbecause it's all mops.
And while mops is veryrelevant, take the blinders off,
right?
Yeah, you just need to go andunderstand it.
And it is it's overwhelming.
It's overwhelming.
I cannot tell you the it's justit's it's it changes the way
(26:56):
you see everything.
Michael Hartmann (26:57):
Yeah.
Tracy Fudge (26:58):
It's not just AI, it's it changes everything.
I mean, I look at things sodifferently now uh in the world,
and um it it changeseverything.
And I don't know where to tellpeople to start that.
I don't, I for okay, so fastforward today.
If I were somebody gettingstarted, I would say go to N8N,
(27:22):
do one of their startertemplates, and just build it.
Just at least get the the clickand the link and all that
stuff, just understand it.
Yeah, and mops people willunderstand that because most
mops people, I mean, actually, Iwouldn't say most mops people.
There's a set of mops peoplethat don't hook systems up.
You know, they're not on theAPI side and they're not on the
(27:42):
data side.
But the ones that are will getthat part.
Um and I would start there.
N8N's gotten a lot better.
I was an NADN, I left.
I didn't, I was not a fan ofthe community.
Um, I didn't feel like therewas a lot of help there.
I thought it was a lot ofbravado and not things that were
(28:04):
true.
Uh, and it turned out that wasthe case.
A lot of people were saying,well, we'll come in and do all
this magic, and and theycouldn't, they could just build
the thing inside of NADN and itwas visually appealing, but it
didn't do anything.
So anyway, um, but now I thinkit's a lot better, and they do
have things for people to golearn quickly.
And I don't, and I think youjust do these P these bytes of
(28:25):
things, maybe an hour each, andjust build and then and then
move into the data layer.
And that's where Airtable andSuper Base uh or Superbase, you
know, that's another one.
Michael Hartmann (28:37):
Do you know how to say I have no idea?
Tracy Fudge (28:41):
But they're very good.
Uh and when I started to stepinto data and believe it or not,
uh into knowledge graphs.
So, and we could go into this,but AI has to know context and
it has to know how things aretagged so that it can go and
retreat it.
(29:01):
Um, this is different thanGoogle's algorithm looking for
keywords.
So AI has to have context.
So you've got to have a definedontology.
And that means, you know, thisis a rotter word for saying
tagging, you know, you've got toknow what things mean and then
how they're related.
Uh, and that's where triplescome in.
So AI can't move beyond thatlayer.
(29:25):
So if you don't have theknowledge graph and the nodes
and ontology, it can't work.
So when that light bulb wentoff in my head, I was like,
damn.
Michael Hartmann (29:37):
Yeah.
So that's like con that's likecon providing context to a whole
nother level, right?
Tracy Fudge (29:44):
Yeah, but it's getting it right.
It's so it doesn't hallucinate,right?
Right.
So you'll see, and they may notbe called knowledge graphs, but
it is.
Michael Hartmann (29:54):
It's gonna No, I it it's a good, I think like
me, it's a good mental model,like in my I I can already kind
of envision it what you mean,but ontology is a big word that
I think most people listening orwatching may not really fully
understand.
Tracy Fudge (30:06):
Um it's getting it is giving the relation of how
things are related.
Michael Hartmann (30:13):
Yeah, yeah.
Um it's like librarian, right?
Virtual librarian.
Um not that anybody goesactually goes to libraries
anymore, but um that's a wholeother topic.
Uh so I'm curious, like whatare what are some of the like
can you give us some examples?
Maybe uh going back to earlyon, like things that you've done
(30:35):
where you've kind of that werehelpful in your learning and
still were maybe productive,maybe weren't productive in
terms of an actual thing you canapply in the work, but it was
like a good learning experience.
And maybe something more recentthat you've done as you've kind
of learned and evolved and andwhat you've seen as possible.
Tracy Fudge (30:55):
Well, I was lucky enough to have my own business
to sort of act as a playground,right?
Michael Hartmann (31:00):
Sure.
Tracy Fudge (31:01):
So I automated um a lot of the front office, a lot
of the back office.
Michael Hartmann (31:07):
Um for your business, you mean?
Tracy Fudge (31:10):
That's where I practiced a lot of it, um, where
I learned um real life thingsthat that were impactful.
So um that's how I was able to.
To do things live and playaround with it.
Not everybody does that.
They're still so if you're inan organization, you've got
their platforms, you know,you've got their stuff, and it
(31:33):
may not, they may not have NADN,you know, they may not want you
playing on an agentic platformto understand it all.
But those the so theefficiency, so that's how I
learned a lot of it.
I just threw my own business,my own use case, I built my own
knowledge graph.
Um, I built my second brain,and that's all knowledge graph
work.
Um, and I'm actually buildingthe second brain for someone as
(31:55):
a client right now.
And that's actually allencompassing because uh a
person's second brain that's anevolved person, say a diplomat,
um, does a lot of differentthings.
So you have to connect a lot ofdifferent dots for that person.
Um, but from a workflowstandpoint, I've done lead
routing, uh, which was honestly,it was easy.
(32:19):
It was an easy problem for whenthey started talking, I knew
right away, okay, that's theproblem.
We need the lead routing to befixed.
And all it was on the salesside.
They couldn't keep theirterritories and their data and
Salesforce up to date.
And it was honestly, I createdan SOP, uh, I created an ideal
(32:41):
framework for it.
This is how we want it.
And they built it.
And then we cut lead routingdown from 48 hours to four
minutes.
Um, and then I did somepredictive lead scoring uh for
somebody, but that's on the mopside.
And I've other done otherboring things um that people
might find are boring, but it isum just workflows and filing.
(33:06):
Uh, you can build an ontologywith a workflow now.
Like to do it right, you'd haveto look go say you had a
thousand documents.
Well, someone would have tophysically go through all
thousand and apply an ontology,relate it, build the triples and
stuff like that.
So there's workflows.
Michael Hartmann (33:23):
So sorry, you've used the word triples now
multiple times, and I I wasjust gonna assume that it's a no
I don't know what you mean.
What does that mean?
Tracy Fudge (33:31):
So it is meaning that um Johnny kicked the ball
in the street.
You got Johnny ball street.
Michael Hartmann (33:39):
Okay.
Tracy Fudge (33:40):
Well, someone else could have kicked the same ball
into the ocean.
So it's it's the it's how it'srelated in triples, three
things.
And so long as you can matchthe three things, then
everything you can see howeverything is related.
Uh and I'm not doing a greatjob of explaining it, but you'd
have to see it.
(34:00):
It's just applyingrelationships by three.
Michael Hartmann (34:04):
Okay.
Interesting.
Okay.
I mean, that makes sense.
I mean, triangulation on stuffkind of makes sense.
Okay.
I'm gonna be uh you're gonnasee my heads like you're gonna
see my eyes going back and forthbecause I'm still thinking
about this.
Keep going.
I'm sorry.
Tracy Fudge (34:19):
No, that was it.
It is, and I'm not anontologist, so there are people,
you know, as I stepped intothis world, I'm in the other
cohorts and taking more classes.
Um, you know, I am I'm I'm kindof addicted to taking classes,
um, to be honest, because Idon't want to do it myself.
I want to sign up for a classthat I've invested in and it's
got beginning and an end date,and it keeps me accountable.
(34:41):
Plus, you meet people along theway, right?
Um, which is a great way.
And and community andnetworking is gonna matter now
more than ever.
Um it's gonna matter now morethan ever because it's gonna be
who you know.
Michael Hartmann (34:53):
Sure.
Yeah.
Tracy Fudge (34:55):
And at my point you are how many degrees you have.
Um, so I don't know if Ianswered that question.
Michael Hartmann (35:02):
No, I think I think I think if you did, I
think I'm trying to think aboutlike if I was someone listening,
watching whatever, and um Ididn't have my own side
business, what could I do?
And I think about some of thethings that I've done, because
most of mine has been uh more ona personal level.
So interesting.
So I've done automationactually related to the podcast,
where um I have now a link forpeople to schedule this
(35:25):
recording, right?
And part of what I have done,this doesn't involve AI at all,
which is interesting.
But I did play with anautomation platform that goes
this looks for when uh that newthing is added to my calendar,
it goes looks for new ones andit does a couple of things with
it, right?
Adds a couple of people to it.
Um I've been thinking aboutlike how could I take that even
a step further?
So I still do a manual effortof prepping.
(35:48):
So I've taught uh in this caseGPT, uh how I like to do the
preparation document.
So for people listening, right,we actually do a fair amount of
preparation for this.
It's not a total wing-it thingas much as we try to make it
feel organic.
But you know, that's still afairly manual process based on
some.
(36:08):
You've everybody knows we wehave initial discussions about
these things, and and uh youknow, I've got recordings.
So I would love to get theprocess if I could set set it up
where like it can go back andgo find my notes and transcripts
from our planning discussionsto generate the document, right?
That would be an automationthat would would benefit for me.
This still doesn't have takenme as much that much time
(36:30):
manually, so it hasn't been atop of mind.
But I have done other things umthat are more focused on still
manual but recurring things thatI'm doing, or you kind of
hinted at this, like doingresearch now.
Um, I find it really, reallyvaluable in doing research on
stuff.
Um to put like I literally hadtoday, had people here, service
(36:54):
repair people.
We have an old uh cooktophaving problems with it.
And so I did a bunch ofresearch on is this stuff I
could fix myself, right?
Did it I did it with uh we hada deep freezer that went bad
over the summer.
Um, and I did a bunch ofresearch like what are the
possible causes and is otherthings that as someone who's not
super handy could could fix.
(37:16):
And it's a it it made gave me asolution that was a $20
solution as opposed to a replaceit for a few hundred dollars,
right?
Which is beneficial.
And I probably could have donethat research on my own, but
having the LLM do a bunch ofthat for me, come back with
stuff, and then I could refineit.
(37:37):
Um I've even done stuff ifyou're a kind of person who
likes to cook at home.
We took photos of all the stufffrom the pantry and said, come
up with some recipes.
Take a picture.
That's what we did.
Yeah, it's really actuallyquite good at that.
I found it's not so good atgenerating photos, but it's or
pictures, but it's really goodat deciphering what's in photos.
(37:59):
I I've been sort of blown away.
I didn't believe it would workwell, and it's been really good.
Yeah, so we actually did thatand it generated a number of
recipes, and we tried one and itwas terrific, right?
No, no, will everything be thatway?
Right?
Tracy Fudge (38:11):
I doubt it, but um so I think walking into your
kitchen and all you have to dois talk to your kitchen, yeah.
Michael Hartmann (38:22):
So I think I think I wouldn't yeah, go ahead.
I think if people think aboutlike how could I try to learn
some of the stuff even if Idon't have a site, like it
doesn't have to be a businesscontext.
I think you can learn a lotfrom doing things, and there are
like I've played withconnecting like my email to one
of these automation platformsusing an AI kind of engine to
(38:43):
automate drafting replies andthings like that, or
categorizing things as importantor not important.
Um honestly, I'm a little bitof a like I'm paranoid and like
just like the amount of accessthat those tools currently have
to have to things like yourinbox, not comfortable with
personally.
So, but I know it's possible.
(39:08):
Like, and it's definitely uhlike that's a very common use
case that most of theseautomation platforms will give
you as a starting point.
Like open up your email,classify the emails, draft the
reply to those ones that youthink are important, and then
you can go in and and then sendyou a summary on a daily basis.
Pretty common thing you coulddo.
And um, like I think there arethings that people, if they
thought about it, even ifthey're not doing a work-related
(39:30):
one, because there are concernsabout you know uh security and
proprietary, you know,confidential information, things
like that, that would probablymake it harder.
Um, but there are ways you canlearn.
So I think that would be highencouragement is to do that kind
of thing.
So I oh I get so one of thethings you have I think
(39:53):
encouraged me, um, and I haveyet to to really do that, and
I've already mentioned rightthat I have a bias towards Chat
PT, GPT because I'm mostfamiliar with that, and then I
particularly like um Grokbecause you can put on the uh
what do they call it, a persona,I think?
You can do the unhingedcomedian one, and it's like I
(40:13):
just that's more of anentertainment thing than
anything else, just to see whatit comes up with.
But um how do you like whatwhat could you share about what
you know about some of the let'skeep into the LLMs with this, I
guess at this point, like whatlike what do you see from them?
What do you think that they'relike what are their strengths
and weaknesses?
Because it feels like they allhave sort of various things, and
(40:35):
I know it's evolving, so likeby the time this gets out, it's
probably gonna change a littlebit.
But in general, what's yourtake on those?
Tracy Fudge (40:42):
I um so I I don't really use grok.
I have in the past, but not tothe extent.
So my main ones are ChatGPT.
I just pay the 20, and thenClaude.
I I think I did bump it up tothe higher one because Claude um
has some memory and contextissues, and I hate having to
start over in the middle.
Michael Hartmann (41:01):
Uh yeah, yeah.
Tracy Fudge (41:02):
So, which is very irritating.
Um, and then Claude'slimitation was not being able to
hold the context inside offolders or projects.
You know, you can createfolders.
So I have a I have a hugefolder structure, and I was
actually arguing with ChatGPT.
Because it used to be able todrop into the folder and ask a
question and uh about anythingin that folder.
(41:26):
Well, today I found uh that itdidn't it didn't have the
context of a conversation that Ihad just had in that same
project.
Michael Hartmann (41:34):
Yeah, I've found that I've I've had to give
it explicit direction, go tothis other chat that's in the
same folder.
Tracy Fudge (41:41):
Yeah, and that was not the case.
So I don't know what's goingon.
Maybe they realize that theylet too much out with chat GPT
that this is a lot of powerrequirement to make your, you
know, the return of your requesthappen.
Um but I use both.
I actually um I intentionallydid not do Claude Desktop until
(42:04):
a month ago or two months agobecause I didn't want to get too
advanced because I knew thatthat would then uh I would build
MCP servers.
So I've evolved into ClaudeDesktop, MCP servers on my on my
desktop to go into thesesystems uh for me.
But that's not for everybody.
(42:26):
You still have to know what thehell you're doing.
Or what it's doing, you know.
Michael Hartmann (42:31):
Yeah.
Tracy Fudge (42:32):
Build, you know, build the JSON to be, you know,
uploaded into, you know, uh N8N.
I mean, you've still got toknow uh a few things about that.
But that that's what uh and Iuse Claude for that.
So but I still go back to chatGVT for creative.
I've I've started to write moreoften.
Um I love to write and I'vestarted to write more often on
(42:55):
LinkedIn and I've done it.
Michael Hartmann (42:56):
Do you find it do you find it better at
drafting initial stuff with alot of context or you drafting
stuff and now asking it to umgive you feedback on it?
Tracy Fudge (43:07):
I've been training it for a while.
Um, and assuming it's still gotthe same project context.
That was one of my things I wasfrustrated with uh because it
has my brand voice, it has mylast post, it knows sort of
where I left off last week.
Right because I get one loud.
I don't, I I I took last weekoff and I was like, I'm not
going on LinkedIn.
So I took a break.
(43:28):
Um, but it's still so that forthat I use it.
Um uh I use Chat GPT for that.
But Claude, I use for thetechnical stuff.
But Claude's gotten better uhwith the creative and the
writing um as well.
And my you know, there's somuch noise, and I know everyone
(43:49):
says this, there's so much AInoise, and now with all these
layoffs and things, and I just Iwant to post something that
matters.
It's just not more of an echoof what's already and sometimes
that's hard.
Michael Hartmann (44:02):
Yeah.
Tracy Fudge (44:03):
Because you want to call out the noise for what it
is, right or wrong, and then youknow, have your point of view.
And it's it's hard to get apoint of view sometimes when
when it's so noisy out there,you know.
You don't want to, you know,it's hard to not be afraid, you
know, these types of things.
Michael Hartmann (44:19):
So yeah.
It's interesting because likemy biggest, my latest challenge,
even though I've got somefolder structures inside
ChatGPT, because I'm also at apaid level, I think that's a you
have to have at least the basiclevel folders.
Yeah.
Um is that even within that,I've got like I'll I I I
(44:39):
actually did something with mywife too, for some she does in
her work, which there's ongoingdoing a lot of the same things,
and so having context is reallygood.
What I found is when I it goeson for a long time, it um chat
PT GPT tends to lock up, right?
If I if it's if the chat threadgets too long, and so that's my
(45:00):
latest one is like how do Iretain what is like I did a lot
of work training it to voice andstuff like that.
Um but the performance on newasks is still there, and that
that's my latest challenge.
But um it's what's a screenedthis week.
Tracy Fudge (45:17):
I don't know what it is.
Michael Hartmann (45:18):
Well, I've been noticing it for a few
weeks, so I don't think it'sanything specific to this week
um for me, but um and and Ithink it's just just that,
right?
The the threads are just solong now that uh it may just be
browser, right?
The browser rendering it takestoo long and it's just locking
up.
So uh but again, like this islike part of the learning
(45:39):
process that we go through, andit it would help me knowing
like, hey, if I'm building somesort of automation for a work
context and um I wanted thecontext to be there and
available over time, right?
I need to think about how tostructure it so that I don't run
into the same thing, anautomation step where I'm like,
hey, asking a whole bunch ofstuff to a say uh uh LLM engine,
(46:01):
and uh because it's like ifit's building it out over time,
is that gonna cause that to slowdown and not run correctly?
So things like that, right?
So I've kind of tied tying thattogether.
Um let's let's let's wrap uphere because you kind of alluded
to this.
Like, how how how can peoplethink about what's the right way
(46:25):
to start to incorporate AI,LLM, you know, automation slash
agentic stuff into their, let'skeep it to Mops, RevOps,
whatever, uh, into their worldwhere it's it's a complement uh
or a um I like the term I heardsomebody use, like it provides a
(46:46):
time dividend, right?
To like into their theirday-to-day um work effort,
right?
How did like what are some waysthey can do think about doing
that?
Tracy Fudge (46:56):
So are you asking from the perspective of the mops
individual or mops?
Michael Hartmann (47:02):
I think I think mops people as
individuals, right, but as partof a team, even if it's a team
of one or two people.
Tracy Fudge (47:10):
Mm-hmm.
Content is a big thing.
Um, and I think content'sreally relevant.
Uh, but you can't just put itinto Chat GPT and say, write an
article on this topic.
I mean, you've got to iterateit, you've got to give it
examples, you have tounderstand.
But I think getting content outfor the team um for um, you
(47:34):
know, email is another one.
I have not ventured into emailfor exactly what you said.
I want to go in and read myemails and things like that.
And I'm scared to, and I I useuh Google Workspace and I'm
scared to turn on Gemini and myemail.
I don't know what it's gonnado.
Right.
Um, but I do think that, youknow, find a few workflows that
(47:56):
are for you and build those out.
Uh for the business, if it's ifit's easy to show a return,
then do it.
And you know you could show itbecause you want to show ROI.
And that's gonna elevate you toleadership as well.
Look, I I shortened your leadrouting.
We could do this.
(48:16):
Look, let's um, and do somepredictive analytics, pull some
data out of sales if you can,you know, the last four
campaigns.
Uh, what else did those peopledo that converted to a sale?
Did they visit uh a certain andbesides the webinar, did they
do these other things?
Michael Hartmann (48:37):
Um looking for pattern, like looking for
patterns tied to exactly uh thedesired outcomes.
Tracy Fudge (48:44):
Most companies aren't capturing those events,
right?
So you've got to figure out,okay, can I pull in data on the
white paper they read, not justthe webinar they signed on?
Can I pull uh and we alreadyknow what emails are opened?
Michael Hartmann (48:57):
Um those you could provide it, you could
provide it not only the data,but also the content, right?
The examples of the content.
This here's this content, it'stied to this data point or this
kind of activity that's in thedata versus this other piece of
content.
And um you could then tell itto look for what are patterns in
(49:18):
the content that uh are tied toum better conversions, more
pipeline, like whatever that isthat you're trying to optimize
for.
Tracy Fudge (49:31):
Possibly look at the last thing they did, like
they looked at an ROI whitepaper or something.
Let's just make that up.
So then you have the AI notethat, right?
But then craft a draft emailaround RLI.
And it's something they justdid, so it's fresh in their
brain.
It may not be the thing thatpushes them over the to be a uh
(49:52):
cli uh a paying client, but it'ssomething that's relevant.
Um, that's something that a lotof people uh could easily do.
Michael Hartmann (50:01):
Yeah.
I mean, I think I think what Iwould do, like if I was in an
organization that had likeproduct marketing and or just
content marketing teams, and umwe had a sales team that had
some sort of call recordingplatform, I would be like, that
sounds like such great contextif you could capture the
transcripts from those salesconversations, then tie it back
(50:26):
to which of those and ended upturning into deals.
Mm-hmm.
And capture the language that'sin there, both client side and
sales side, to feed back intohow are you thinking about the
content you're publishing andthe material you're providing to
the sales teams, right?
Because you because like Ithink there's so much crap out
there on B2B websites inparticular that is like just
(50:50):
gobbledygook nonsense thatdoesn't really tell anybody
anything about what you actuallydo.
Like if you could drive, likeif you could do that, take that
back, feed that back into yourcontent engine, um, which then
hopefully is gonna drive moretraffic and awareness, right?
Like you could build a prettystrong flywheel, um, I think in
terms of that that's just simplybased on what are the actual
(51:11):
terms and uh things that uhcustomers, prospects actually
care about, and make sure thatyou've got that really well
honed on your public-facingwebsite and the content that
goes with it.
Like to me, that's like it'sit's not trivial, I get it, but
like it talks about like uh anROI over the long run, both
(51:32):
short term and long term.
I think that that one to me isa use case that I would love to
see more people trying out.
Tracy Fudge (51:38):
And then you could retarget people to go to that
new dynamically created blogpost that was created, you know.
So these things are dynamicallyum done because there's APIs in
there that are doing it.
Yeah.
So there's the API work, butabsolutely.
Michael Hartmann (51:53):
Yeah, so anyway, well, hey, I it feels
like as so many of these arethat we could go on and on and
on.
Uh, but unfortunately, we dohave to we do need to wrap up.
So, Tracy, first off, thank youfor sharing.
It's been a lot of fun.
I enjoyed the conversation.
I'm so glad we were able tomake this work uh while everyone
else is out there having a funtime at Mopspalooza.
(52:14):
But if uh if if folks want toconnect with you or learn more
about what you're doing or godeeper on this topic with you,
what's the best way for them todo that?
Tracy Fudge (52:24):
Uh find me on LinkedIn.
Um, my website's AI by Thrive,and I'm Tracy with an EY at AI
by Thrive, if they won't emailme.
Michael Hartmann (52:33):
Perfect.
Well, thank you again, Tracy.
Appreciate it.
Thanks always to our listenersand now watchers now that we're
uh going live with videos.
Well, we appreciate yoursupport.
And as always, if you haveideas for topics or guests or
want to be a guest, you canreach out to Mike, Naomi, or me.
We'd be happy to talk to youabout it.
Until next time.
Bye, everybody.
Tracy Fudge (52:54):
Thank you, bye bye.
Popular Podcasts
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
Are You A Charlotte?
In 1997, actress Kristin Davis’ life was forever changed when she took on the role of Charlotte York in Sex and the City. As we watched Carrie, Samantha, Miranda and Charlotte navigate relationships in NYC, the show helped push once unacceptable conversation topics out of the shadows and altered the narrative around women and sex. We all saw ourselves in them as they searched for fulfillment in life, sex and friendships. Now, Kristin Davis wants to connect with you, the fans, and share untold stories and all the behind the scenes. Together, with Kristin and special guests, what will begin with Sex and the City will evolve into talks about themes that are still so relevant today. "Are you a Charlotte?" is much more than just rewatching this beloved show, it brings the past and the present together as we talk with heart, humor and of course some optimism.
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.