October 27, 2023 • 21 mins
Get ready to revolutionize your approach to hiring! We're bringing you cutting-edge insights from two insightful CEO's - Mateo Cavasoto of Emi Labs and Jeremy Schiff of RecruitBot - who are reshaping the world of recruiting and sourcing.
Mateo will walk you through Emi Labs' pioneering methods to automate communication in the hiring process for frontline workers. Understanding and integrating into a hiring managers' daily routines can make technological adoption smoother and more efficient.
As we pivot our focus to sourcing, Jeremy from RecruitBot reveals how their passive outreach tool, much like Netflix's recommendation system, leverages machine learning and data science to engage with hard-to-fill roles.
We also delve into the importance of customer success for both Emi Labs and RecruitBot. So buckle up for an episode loaded with invaluable tips and insights. Tune in now!
Special mini series recorded with Oleeo at HR Tech 2023 with hosts Ryan Leary, Brian Fink, and Shally Steckerl.
Listen & Subscribe on your favorite platform
Apple | Spotify | Google | Amazon
Visit RecruitingDaily
Twitter @RecruitingDaily
Join the Secret Sourcing Group
Learn more about #HRTX Events
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:07):
Ladies and gentlemen, boys and girls, party people,
welcome back to HR Tech.
This is Brian Fink.
I'm joined with Shaly Stekrel,the godfather of sourcing.
We are rocking the mic here atOlio's podcasting booth live on
the floor in Las Vegas.
We are joined by not one, buttwo guests from two different
organizations.
We've got Matteo from Emmy Labs, right yeah.
Speaker 2 (00:29):
Are we good?
Speaker 1 (00:29):
Awesome sauce.
And we've got my man, jeremy,who has already been on the
podcast one time, to talk abouthis sensational sourcing tool
for all things digital andenvironmental, and I love what
is her name at Recruit Bot.
Speaker 4 (00:50):
Oh are you referring to Arby.
Speaker 1 (00:51):
Yes, okay, all right, I was, so we're going to talk
about who Arby is in a minute,so we'll get into that in just a
second.
That was what I was strugglingfor for words, shaly.
What's going on, my man?
What's the vibe on the floor?
It's looking pretty good.
Yeah, we're getting, we'regetting busy, all right.
So, speaking of getting busy,what I want to do is I want to
get busy, busy, busy, busy.
Be as quick as can be.
What I want to ask, I want toask a use case for the problem
(01:15):
that each of you are trying tosolve in recruiting and sourcing
today who wants to go first?
Okay, I'll go first.
Speaker 2 (01:22):
My vote is but damn Okay, I go.
Thanks for having me first ofall Happy to.
So typical use case of ME so westreamline the hiring process
for front-line workers and atypical process that we find is
that in the recruitment processof hiring of front-line workers,
hiring managers are in thefield.
They're either at a work at aworkhouse, at a facility or at a
(01:45):
retail store, so getting theiravailability it's really hard.
So we streamlined all the kindof experience using chat.
But we found that streamliningthe process of the hiring
managers it's critical.
So typical use case it's theimplement ME on top of the HES
and they automate communicationwith the candidates but also
internally with hiring managers.
(02:06):
So they usually implement ME toreduce the time to hire but
specifically to save time fromrecruiters and hiring managers,
impacting revenue, not onlyproductivity and cost All right,
so I got it.
Speaker 1 (02:17):
I know I said I was going to do one thing.
I'm going to kind of whoop,whoop, whoop and whatever.
All right, how do you trainmanagers to use this technology?
Because at the end of the day,they're not sitting in their
offices surfing LinkedIn to seewho's got what status update.
How do you?
Speaker 2 (02:31):
train them, first of all, talking to them,
understanding their day today.
So we work with Walmart,kinecant and on, and we went,
talked to them and we understoodthat they are not in front of a
computer.
So we asked them which tools doyou use?
And we found that, for example,they all had phones, text in
some populations, like WhatsApp.
So we built our tool on top ofthe things that they were
(02:52):
already using and we designedthis for them.
So if you design from thatpoint of view, then the change
management it's always tough.
It's a bit easier than tryingto push down their throats a new
interface, a new system,something that is not designed
for them, like a calendar linkor something like that.
So that's the starting pointfor us.
Speaker 3 (03:13):
All right, so this is scheduling specifically no, we
do end to end.
Speaker 2 (03:17):
From accounting it says hi be a any job board and
they are redirected to anywithin the application in the
ATS and then we streamline toretention.
But that's the typical use casethat we are encountering the
hiring manager back and forth,all right.
Speaker 1 (03:30):
I want to jump on that word streamline and I'm
going to turn the conversationover to Jeremy to talk about his
use case and what.
Okay, so real quick.
I've already interviewed Jeremybefore.
I'm fascinated by everythinghe's done with machine learning.
I'm actually sitting across thetable from an actual data
scientist.
So when I talk to people whohave built recruiting tools,
they're usually coming from asales side.
They're not coming from thetechnology angle.
(03:51):
So tell us about RecruitBot andthe streamline that it provides
.
Speaker 4 (03:54):
Yeah, always great to be here.
I love talking to RecruitingDaily, so really appreciate the
opportunity.
Again I'm great to be chattingto you guys.
So RecruitBot again is apassive outreach tool.
So our job is to help engagewith very hard to fill roles,
specifically in white colorrecruiting.
But that can range from theoriginal problem that I had,
(04:16):
which was when I was runningmachine learning and data
science at OpenTable.
I could not hire enough machinelearning engineers.
They're just, they're purplesquirrels, they're impossible to
hire and the processes thatyou're usually using.
Speaker 3 (04:29):
Sorry about that.
That was me.
Speaker 4 (04:33):
The processes you're usually.
Oh yeah, it makes senseProcesses.
You're usually using theprocesses that you're usually
using are pretty manual and itoften is requiring really smart
sources to run all over theinternet and find information
from six or seven differentsites, to go and cross-reference
(04:54):
it and say here's the set ofpeople that I want to go and
reach out to.
And so Recruitbot is ultimatelyfocused on how do I do that all
in one spot and how do I makethat incredibly easy?
So, rather than going to sevendifferent sites to go and find
your candidates and reach out tothem, you just log into
Recruitbot, run a search, thenuse real machine learning to go
and find people who are similarto the people you like.
(05:15):
We like to say the same waythat Netflix would recommend
movies similar to movies youlike.
You can do that, except we'regoing to recommend candidates
that are similar to thecandidates you already like,
just giving feedback on a one tofive star scale.
And because we have all of thecontact information, emails and
phone numbers, we can automateemail campaigns to make it
really easy to sort of talk tothem.
Or you could even load thatinto your favorite CRM and
(05:36):
automate texting or anythingelse from there.
So it makes it really easy tosolve the problem suit to nuts
of engaging with really hard tofill talent.
Speaker 1 (05:45):
Jeremy, there's one question that I don't know.
We've had three conversations.
This is our third conversation.
I think we're at three now,yeah, oh, all right.
Speaker 3 (05:53):
Hey, we're going to prom.
I love it.
Speaker 1 (05:55):
He's keeping count.
So one of the things that I Letme keep coming back, One of the
things I don't know is thatdoes Recruitmont use natural
language search to fund thosecandidates?
If I say hey, I'm looking fordata scientists who went to MIT
who live in Boston.
Speaker 4 (06:14):
Yep.
Speaker 1 (06:15):
Okay, Maybe I should have made it a little bit harder
than MIT in Boston right, yeah.
Right.
Speaker 4 (06:21):
So in some ways yes and in some ways no.
So we aren't yet literally Ican go and type in any generic
text string and have it work.
We are playing withtechnologies like that
internally, but having that workreally robustly for users is
really important, and so we'llsee sort of how that evolves.
But when I refer to the machinelearning that's going and
(06:42):
predicting which candidates arerelevant to you, that is already
and has been for many yearsusing machine learning and
natural language processing.
That's how we're actually goingand prioritizing the resume.
So what's happening here is thesystem's going and saying I'm
trying to generate statisticalpatterns of what's in common
with the candidates that aregood and what's in common with
(07:03):
the candidates that aren't good.
Right, and so this might beobvious, things like oh, these
job titles are this amount ofyears of experience.
But it might be much morenuanced.
It might be that they haveleadership words, they have
ownership words or they are verynumbers driven if they're a
sales role.
These are sort of more vagueconcepts that a natural language
processing system can go andunderstand, and go and
(07:24):
understand the patterns at aconceptual level when it's going
and working out how torecommend it to you.
So we've been integrating thatinto our system for many, many
years now.
Speaker 1 (07:33):
Okay, and one of the things that I think the two of
you have in common, even thoughyou're coming at different
problems, is the level ofcustomer service and customer
success that you're all nodding,that you guys are vested in.
What and why is success to thecustomer important?
I know that it's like justsigning on the dot of the line,
but like, what tangible resultsare we trying to produce for our
(07:54):
customers?
Who wants to handle that onefirst?
Speaker 4 (07:56):
I mean I'm happy to start.
I mean tangible results arereally easy.
They come to us with a veryspecific problem.
They're all these roles.
They're really hard to fill.
We don't know how to do them.
We're not getting in the volumeof inbound.
We're not getting the rightpeople in through LinkedIn or
whatever other mechanisms we'reusing.
What do we do right?
Like they're Basically in panicmode.
They're like we need to hirethese sorts of people.
This, this is essential for thebusiness.
Speaker 3 (08:18):
What they just can't find what they're looking for.
Speaker 1 (08:20):
Yeah so it was a reference to you too, with the
sphere you can playing at thesphere.
Speaker 3 (08:26):
Yeah, they're playing right now.
Yeah, they're playing, yeah,they're gonna be playing tonight
.
Speaker 1 (08:29):
I think it's like $8,000 a ticket.
Shali, can you front?
Yeah?
No, you said yep.
I heard a yep.
Speaker 3 (08:36):
Yep no.
Speaker 1 (08:38):
Right.
Speaker 3 (08:38):
Yes.
Speaker 1 (08:39):
So, mateo, what about ?
What about?
Speaker 2 (08:41):
CX, cs, like I think it's critical.
I divided, like in two or threePoints of contact.
The first one is when youimplement a platform like CS and
being closer to the customersupport.
It's critical because you needto understand the specific pains
.
So technology is just like atool to solve problems.
So you need to understand whichare the specific pain
(09:02):
structures and what they need.
And that's a two-wayCommunication where you teach
also maybe the customer whichare the best practices, what our
customers are doing, and youlisten, you try to solve their
problems.
Then you got the changemanagement part, which we
discuss a bit when talking abouthurry managers, which is
technology.
Again, it's just one point ofwhat, one starting point.
(09:23):
But then you need to make usersuse an at all, like me, which
we're talking about hundreds ofrecruiters.
And then the final piececonnect you to the ROI and what
we are talking about, which islike Showing the ROI the money
and, from a CS point of view,continuing that relationship and
like Helping your championsinternally to follow up that
(09:43):
Financial return, like over andover throughout the entire life
cycle of the customer.
So for us, like we did manymistakes in the past and we took
that and that was a bigdifference beyond the product.
Speaker 1 (09:54):
All right about big differences.
I'm gonna let Charlie get aquestion in here edge-wise,
because I've been taking up the,I've been sucking all the
oxygen out of this.
What you got, uh.
Okay, great, we were talkingabout the Tell them who you're
talking to.
You pointed to you.
Pointed to Jeremy.
Speaker 3 (10:11):
I pointed Jeremy okay , alright and we were talking
about the language.
We sort of later what you weretalking about before here, that
the natural language Recognitionand natural language querying.
So I just wanted to kind ofdive into that a little bit and
Find out the the answer to thequestion that I had asked you
(10:32):
before, which is you know, youobviously you're not using,
let's call it, off-the-shelf AItype of, you know, large
language models.
It's not quality I, because itreally isn't.
You said you've got somethingdifferent that is, it's a large
language model on top of yeah,so we have to be in.
Speaker 4 (10:51):
We haven't launched it yet, so I have to be a little
bit cagey about how it's goingto work.
Speaker 3 (10:54):
Yeah, but concept conceptually.
Speaker 4 (10:56):
Yeah, I mean effectively it's, it's what
we're just talking about earlier, right like the.
The way that people want totalk about these things are they
just want to say I want to findpeople that understand this
technology and this technologymay be a really recently
burgeoning technology that wedon't have.
A lot of people on LinkedInthat have said like hey, this is
(11:18):
a skill of mine, right, andthat's it's.
Speaker 3 (11:20):
That's exactly the problem, right there is that
there's all these keywords thatpeople are used to searching
with, and the language changesso quickly that you really would
have a very large Booleanstatement with all these
esoteric terms in it to stillonly capture a small portion of
the population anyways.
So that's why I think thebenefit of these large language
(11:43):
models is be able to expand thatontology, right.
Speaker 4 (11:47):
Totally agree.
I mean, they were even talkingabout it earlier today, about
how much they're like.
They're like big, likecompanies are really focused on
making sure that they're stayingahead, right.
And so if you're staying ahead,you're going and pulling your
engineering team and productteam and design team and you're
saying, hey, what is what's thetechnology?
We need to know.
Speaker 3 (12:03):
We need to know, because that's what we need to
hire, for you want to hire.
Speaker 1 (12:06):
I go and then type that keyword into LinkedIn and
whatever else and that they'relike what do you mean?
Speaker 4 (12:10):
this technology Like I, like I love seeing.
Hey, I've been you, I'm a, I'ma generative AI expert.
I've been using chat GPT forfour years and you're like cat,
gpt hasn't been around for fouryears.
Speaker 3 (12:19):
That's yeah.
What are you talking?
About and and so, yeah, it'sexactly the keywords.
Yeah, so so the question that Iwanted to get to was the data
are you using?
Are you looking throughhomogenous data?
Is that's kind of where I not.
In homogenous data, things tendto get really messy.
Speaker 4 (12:39):
Yeah, so there are many different steps that are
necessary along the way right.
So a great example is we haveour database of 600 million
people.
Speaker 3 (12:46):
But those are people, so that's homogenous.
Speaker 4 (12:48):
Exactly, but that's coming from a number of
heterogeneous sources.
And then we part of ourtechnology stack is actually
going and sort of synthesizingthat down into a way that, hey,
data source A refers to thinksabout data in this way, data
source B does it in this way.
Speaker 3 (13:07):
And it's organized.
Speaker 4 (13:08):
And we have to organize and there's a lot of
mapping.
That's, frankly, just humans inthe loop going and cleaning up
sort of incredible amounts ofdata, and then there's having to
match this and then there's thebeing able to do it at scale,
right.
I think we're dealing with over200, or two billion data points
now to generate our 600 millioncandidates every six weeks, and
(13:28):
we're pulling in more sourcesliterally every time.
Speaker 3 (13:30):
We're building in our data set.
Yeah, and that's where largelanguage models have Are going
to flourish.
Yeah, that's what I'm saying.
Yeah, exactly.
Speaker 4 (13:36):
So if you can go and sort of homogenize the data like
the way you're framing it, thatmakes the task for the large
language model a lot easier togo and sort of do things on top.
Speaker 3 (13:46):
And predictable.
Speaker 4 (13:46):
That's exactly right.
And again, we all have these.
Like there's all these problemsright where it'll invent.
Speaker 1 (13:51):
Like it'll invent sort of Hallucinations yeah
hallucinations.
Speaker 4 (13:54):
A lot of people aren't familiar with the term so
I try to avoid it.
But yeah, exactly, we got toeducate everyone about it.
It'll hallucinate all sorts ofdata that's not there, and so
the simpler the problem andit'll sort of, it'll be easier
for it to compensate for it.
Speaker 3 (14:08):
The more congruent.
That's exactly right.
Speaker 4 (14:11):
And so if you're doing this right, there's going
to be a lot of balance between,like there's sort of this naive
solution which is I just go intochat GPT and I type in whatever
I want and I get my Boolean'sring.
That's the misconception andlike A, that'll work okay, but
to your point before, when wewere talking before, it's going
to be really general it's goingto be super generic and you're
not going to get sort of.
The whole point is that you'renarrowing in on what you want.
(14:33):
I often talk about the sourcingdonut where, like, you don't
want to find the center of thedonut, where everyone's looking
for the MIT software engineerthat we're in In Boston.
Yeah, in Boston that works atGoogle because, like everyone's
already reached out to thosepeople, you want to find the
actual donut.
You want to find the people whoare great for your company but
are only okay for sort ofeveryone else's company, and so
(14:56):
the more you like.
What we've basically found iswe're sort of have a very large
advantage because we've spentyears building out these
incredible data sets andpipelines that allow us to
homogenize the data, which thenallow us to lay other things
like other types of naturallanguage processing technology,
on top to sort of provide a lotmore power.
(15:17):
And if you just threw it at thehomogenous data sets, you
wouldn't get anything out.
Like every now and then, I seesomeone that's like we have two
billion, we have two billioncandidate profiles, and I'm like
, yep, you did not deduplicateyour data.
That is what I am hearing.
When you have two billionprofiles, it is not that you
magically have inventedcandidates that like don't exist
in the world.
Speaker 1 (15:37):
You have solved the labor crisis.
Speaker 3 (15:39):
That's right, yeah, you have two billion, but 1.6
billion of those are the same.
That is correct.
So, carrying it over to Mateo,what patterns are you
identifying in your side of?
Speaker 2 (15:52):
the world, so I'll be cautious, because Jeremy here
is an expert.
So what do I say?
No, but since we work withfront line workers, we
comeropyvs so far Many of theseworkers is from them.
Workers are offline, so there'sno data online.
Yeah great, so there's noLinkedIn data.
Wait a minute, they're not onFacebook.
They are on Facebook a lot, butyou can spread out under that
(16:14):
name and not under like.
You don't have detail abouttheir work history, what they
want to do, so there's like theyare sort of it's Facebook, like
every everybody's on Facebook.
Yeah.
Speaker 1 (16:24):
Unless, I know, I know everybody over 30.
Speaker 2 (16:26):
Yeah, but so for us, like that data said, like we
part of what we're doing, it'snot getting to the source in
space, so we're not looking toget new kind of a right.
Yeah, but we're looking to toway to explore things on how
data enables us to provide valueto our customers.
So, for example, like this,this workers coming this data
(16:50):
from, from from each one of ourcustomers, allows us to
understand what's the perfectcandidate profile for this
company, for this kind of worker.
So, for I'm we're talking aboutour employers.
They have like hundreds orthousands of workers, which
means that we process likemillions of candidates.
We got a counter in our booththat it's in real time that
shows how many from andcandidates will process so far,
(17:11):
and it keeps ticking and we areat 7 million candidates right,
which making your point.
It's not that we can talkgenerally speaking about data
points, but each one of ourcustomers.
They got a big chunk of those.
So we are playing around toplay somehow To see how we can
provide unique value to them andwhat are the?
(17:33):
yes, so you're finding patternsamongst that, exactly for
example and under there are moregeneric things that we are
doing which are more basic, whatyou are saying, which is, for
example, enabling a betterConversational interface with
the candidates.
So we get to understand withyou got, if you got seven
thousand, seven millionCandidates that went through the
similar conversation, you can,in case and understand how that
(17:56):
Be more preemptive, exactly,yeah, yeah it's one of the
biggest challenges I think withwith people in our space and
technology and HR it just ingeneral.
Speaker 3 (18:05):
Talent technology is ideation.
You know, the technology hasexisted For a while but people
just don't.
They haven't been able tofigure out how to apply it
because they don't really haveany ideas on what to apply.
Speaker 2 (18:21):
The problem set to the problem is when you start
with that, like what we I agree,complete with you like we
thought, first of all we need toresolve, solve a real problem,
which is the efficiency of thehiring.
Right in the process, we getall this data right, we make
these candidates which areinvisible, visible, yep.
And on top of that, we keepexploring new ways of telling
(18:41):
value.
But you can start the wayaround, like getting data and
then thinking, okay, what's thevalue that we'll deliver?
Speaker 3 (18:47):
because at least you can build a company like that
first value, you get data andthen you explore new ways and
it's to stack value on top ofthat and identifying that
essentially figuring out whatthe problem is that you want to
solve is not something that TApeople are really good at,
because they're looking at theproblem head-on, hiring people,
(19:09):
but they don't really.
Speaker 1 (19:10):
That's why I ask the questions about what wise the,
Not the five wires.
Speaker 3 (19:16):
Yeah yeah, exactly.
So you know the problem thatpeople have with LinkedIn
nowadays, becoming dependent onit and having nothing else but
LinkedIn as their only source.
That's not a new problem.
Before LinkedIn it was careerbuilder, before that it was
monster.
Before that it was net temps orthe fact, or online career
center or right, so that there'salways been this dependency on
(19:36):
that one technology, because itwas the ones, or solution,
because that was the onesolution everybody knew.
But the.
The origin of the problem isn'twhere do we go to find
candidates, it's when are thecandidates?
Speaker 4 (19:51):
And, specifically, how do I actually engage with
them right like?
Speaker 3 (19:53):
after you find those exactly like finding like people
you want.
Speaker 4 (19:56):
You want a thousand software engineers.
Here you go like it's supereasy to do.
You want.
You want a thousand softwareengineers that'll talk to you
very, very different problems, acompletely different problem.
Speaker 2 (20:06):
And the real problem is you need the software
engineers to run your company,to build product or, in your
software, your retail associateto sell on your retail.
So you need to go, I think,that deep and then go to the
problem like what's the realproblem?
Okay, having people, then whydon't you have the people?
Okay, because my Harry Majorstell me that I don't have the
people.
Speaker 1 (20:23):
And then you get Because they're not on the All
right, all right, we're notgonna beat up on LinkedIn too
much tonight.
All right, so real quick, justto recap we have been joined by
Jeremy at recruit bot.
We've been joined by Mateo atEmmy labs.
I've been joined by thegodfather of sourcing, mr Shally
stick.
Well, we are coming to you livefrom the olio booth at HR tech
(20:45):
in Las Vegas.
We wish you were here.
We're sorry that you're not.
Thanks for joining us and we'llsee you on the next episode.
Ladies and gentlemen, boys and girls, party people,
welcome back to HR Tech.
This is Brian Fink.
I'm joined with Shaly Stekrel,the godfather of sourcing.
We are rocking the mic here atOlio's podcasting booth live on
the floor in Las Vegas.
We are joined by not one, buttwo guests from two different
organizations.
We've got Matteo from Emmy Labs, right yeah.
Speaker 2 (00:29):
Are we good?
Speaker 1 (00:29):
Awesome sauce.
And we've got my man, jeremy,who has already been on the
podcast one time, to talk abouthis sensational sourcing tool
for all things digital andenvironmental, and I love what
is her name at Recruit Bot.
Speaker 4 (00:50):
Oh are you referring to Arby.
Speaker 1 (00:51):
Yes, okay, all right, I was, so we're going to talk
about who Arby is in a minute,so we'll get into that in just a
second.
That was what I was strugglingfor for words, shaly.
What's going on, my man?
What's the vibe on the floor?
It's looking pretty good.
Yeah, we're getting, we'regetting busy, all right.
So, speaking of getting busy,what I want to do is I want to
get busy, busy, busy, busy.
Be as quick as can be.
What I want to ask, I want toask a use case for the problem
(01:15):
that each of you are trying tosolve in recruiting and sourcing
today who wants to go first?
Okay, I'll go first.
Speaker 2 (01:22):
My vote is but damn Okay, I go.
Thanks for having me first ofall Happy to.
So typical use case of ME so westreamline the hiring process
for front-line workers and atypical process that we find is
that in the recruitment processof hiring of front-line workers,
hiring managers are in thefield.
They're either at a work at aworkhouse, at a facility or at a
(01:45):
retail store, so getting theiravailability it's really hard.
So we streamlined all the kindof experience using chat.
But we found that streamliningthe process of the hiring
managers it's critical.
So typical use case it's theimplement ME on top of the HES
and they automate communicationwith the candidates but also
internally with hiring managers.
(02:06):
So they usually implement ME toreduce the time to hire but
specifically to save time fromrecruiters and hiring managers,
impacting revenue, not onlyproductivity and cost All right,
so I got it.
Speaker 1 (02:17):
I know I said I was going to do one thing.
I'm going to kind of whoop,whoop, whoop and whatever.
All right, how do you trainmanagers to use this technology?
Because at the end of the day,they're not sitting in their
offices surfing LinkedIn to seewho's got what status update.
How do you?
Speaker 2 (02:31):
train them, first of all, talking to them,
understanding their day today.
So we work with Walmart,kinecant and on, and we went,
talked to them and we understoodthat they are not in front of a
computer.
So we asked them which tools doyou use?
And we found that, for example,they all had phones, text in
some populations, like WhatsApp.
So we built our tool on top ofthe things that they were
(02:52):
already using and we designedthis for them.
So if you design from thatpoint of view, then the change
management it's always tough.
It's a bit easier than tryingto push down their throats a new
interface, a new system,something that is not designed
for them, like a calendar linkor something like that.
So that's the starting pointfor us.
Speaker 3 (03:13):
All right, so this is scheduling specifically no, we
do end to end.
Speaker 2 (03:17):
From accounting it says hi be a any job board and
they are redirected to anywithin the application in the
ATS and then we streamline toretention.
But that's the typical use casethat we are encountering the
hiring manager back and forth,all right.
Speaker 1 (03:30):
I want to jump on that word streamline and I'm
going to turn the conversationover to Jeremy to talk about his
use case and what.
Okay, so real quick.
I've already interviewed Jeremybefore.
I'm fascinated by everythinghe's done with machine learning.
I'm actually sitting across thetable from an actual data
scientist.
So when I talk to people whohave built recruiting tools,
they're usually coming from asales side.
They're not coming from thetechnology angle.
(03:51):
So tell us about RecruitBot andthe streamline that it provides
.
Speaker 4 (03:54):
Yeah, always great to be here.
I love talking to RecruitingDaily, so really appreciate the
opportunity.
Again I'm great to be chattingto you guys.
So RecruitBot again is apassive outreach tool.
So our job is to help engagewith very hard to fill roles,
specifically in white colorrecruiting.
But that can range from theoriginal problem that I had,
(04:16):
which was when I was runningmachine learning and data
science at OpenTable.
I could not hire enough machinelearning engineers.
They're just, they're purplesquirrels, they're impossible to
hire and the processes thatyou're usually using.
Speaker 3 (04:29):
Sorry about that.
That was me.
Speaker 4 (04:33):
The processes you're usually.
Oh yeah, it makes senseProcesses.
You're usually using theprocesses that you're usually
using are pretty manual and itoften is requiring really smart
sources to run all over theinternet and find information
from six or seven differentsites, to go and cross-reference
(04:54):
it and say here's the set ofpeople that I want to go and
reach out to.
And so Recruitbot is ultimatelyfocused on how do I do that all
in one spot and how do I makethat incredibly easy?
So, rather than going to sevendifferent sites to go and find
your candidates and reach out tothem, you just log into
Recruitbot, run a search, thenuse real machine learning to go
and find people who are similarto the people you like.
(05:15):
We like to say the same waythat Netflix would recommend
movies similar to movies youlike.
You can do that, except we'regoing to recommend candidates
that are similar to thecandidates you already like,
just giving feedback on a one tofive star scale.
And because we have all of thecontact information, emails and
phone numbers, we can automateemail campaigns to make it
really easy to sort of talk tothem.
Or you could even load thatinto your favorite CRM and
(05:36):
automate texting or anythingelse from there.
So it makes it really easy tosolve the problem suit to nuts
of engaging with really hard tofill talent.
Speaker 1 (05:45):
Jeremy, there's one question that I don't know.
We've had three conversations.
This is our third conversation.
I think we're at three now,yeah, oh, all right.
Speaker 3 (05:53):
Hey, we're going to prom.
I love it.
Speaker 1 (05:55):
He's keeping count.
So one of the things that I Letme keep coming back, One of the
things I don't know is thatdoes Recruitmont use natural
language search to fund thosecandidates?
If I say hey, I'm looking fordata scientists who went to MIT
who live in Boston.
Speaker 4 (06:14):
Yep.
Speaker 1 (06:15):
Okay, Maybe I should have made it a little bit harder
than MIT in Boston right, yeah.
Right.
Speaker 4 (06:21):
So in some ways yes and in some ways no.
So we aren't yet literally Ican go and type in any generic
text string and have it work.
We are playing withtechnologies like that
internally, but having that workreally robustly for users is
really important, and so we'llsee sort of how that evolves.
But when I refer to the machinelearning that's going and
(06:42):
predicting which candidates arerelevant to you, that is already
and has been for many yearsusing machine learning and
natural language processing.
That's how we're actually goingand prioritizing the resume.
So what's happening here is thesystem's going and saying I'm
trying to generate statisticalpatterns of what's in common
with the candidates that aregood and what's in common with
(07:03):
the candidates that aren't good.
Right, and so this might beobvious, things like oh, these
job titles are this amount ofyears of experience.
But it might be much morenuanced.
It might be that they haveleadership words, they have
ownership words or they are verynumbers driven if they're a
sales role.
These are sort of more vagueconcepts that a natural language
processing system can go andunderstand, and go and
(07:24):
understand the patterns at aconceptual level when it's going
and working out how torecommend it to you.
So we've been integrating thatinto our system for many, many
years now.
Speaker 1 (07:33):
Okay, and one of the things that I think the two of
you have in common, even thoughyou're coming at different
problems, is the level ofcustomer service and customer
success that you're all nodding,that you guys are vested in.
What and why is success to thecustomer important?
I know that it's like justsigning on the dot of the line,
but like, what tangible resultsare we trying to produce for our
(07:54):
customers?
Who wants to handle that onefirst?
Speaker 4 (07:56):
I mean I'm happy to start.
I mean tangible results arereally easy.
They come to us with a veryspecific problem.
They're all these roles.
They're really hard to fill.
We don't know how to do them.
We're not getting in the volumeof inbound.
We're not getting the rightpeople in through LinkedIn or
whatever other mechanisms we'reusing.
What do we do right?
Like they're Basically in panicmode.
They're like we need to hirethese sorts of people.
This, this is essential for thebusiness.
Speaker 3 (08:18):
What they just can't find what they're looking for.
Speaker 1 (08:20):
Yeah so it was a reference to you too, with the
sphere you can playing at thesphere.
Speaker 3 (08:26):
Yeah, they're playing right now.
Yeah, they're playing, yeah,they're gonna be playing tonight
.
Speaker 1 (08:29):
I think it's like $8,000 a ticket.
Shali, can you front?
Yeah?
No, you said yep.
I heard a yep.
Speaker 3 (08:36):
Yep no.
Speaker 1 (08:38):
Right.
Speaker 3 (08:38):
Yes.
Speaker 1 (08:39):
So, mateo, what about ?
What about?
Speaker 2 (08:41):
CX, cs, like I think it's critical.
I divided, like in two or threePoints of contact.
The first one is when youimplement a platform like CS and
being closer to the customersupport.
It's critical because you needto understand the specific pains
.
So technology is just like atool to solve problems.
So you need to understand whichare the specific pain
(09:02):
structures and what they need.
And that's a two-wayCommunication where you teach
also maybe the customer whichare the best practices, what our
customers are doing, and youlisten, you try to solve their
problems.
Then you got the changemanagement part, which we
discuss a bit when talking abouthurry managers, which is
technology.
Again, it's just one point ofwhat, one starting point.
(09:23):
But then you need to make usersuse an at all, like me, which
we're talking about hundreds ofrecruiters.
And then the final piececonnect you to the ROI and what
we are talking about, which islike Showing the ROI the money
and, from a CS point of view,continuing that relationship and
like Helping your championsinternally to follow up that
(09:43):
Financial return, like over andover throughout the entire life
cycle of the customer.
So for us, like we did manymistakes in the past and we took
that and that was a bigdifference beyond the product.
Speaker 1 (09:54):
All right about big differences.
I'm gonna let Charlie get aquestion in here edge-wise,
because I've been taking up the,I've been sucking all the
oxygen out of this.
What you got, uh.
Okay, great, we were talkingabout the Tell them who you're
talking to.
You pointed to you.
Pointed to Jeremy.
Speaker 3 (10:11):
I pointed Jeremy okay , alright and we were talking
about the language.
We sort of later what you weretalking about before here, that
the natural language Recognitionand natural language querying.
So I just wanted to kind ofdive into that a little bit and
Find out the the answer to thequestion that I had asked you
(10:32):
before, which is you know, youobviously you're not using,
let's call it, off-the-shelf AItype of, you know, large
language models.
It's not quality I, because itreally isn't.
You said you've got somethingdifferent that is, it's a large
language model on top of yeah,so we have to be in.
Speaker 4 (10:51):
We haven't launched it yet, so I have to be a little
bit cagey about how it's goingto work.
Speaker 3 (10:54):
Yeah, but concept conceptually.
Speaker 4 (10:56):
Yeah, I mean effectively it's, it's what
we're just talking about earlier, right like the.
The way that people want totalk about these things are they
just want to say I want to findpeople that understand this
technology and this technologymay be a really recently
burgeoning technology that wedon't have.
A lot of people on LinkedInthat have said like hey, this is
(11:18):
a skill of mine, right, andthat's it's.
Speaker 3 (11:20):
That's exactly the problem, right there is that
there's all these keywords thatpeople are used to searching
with, and the language changesso quickly that you really would
have a very large Booleanstatement with all these
esoteric terms in it to stillonly capture a small portion of
the population anyways.
So that's why I think thebenefit of these large language
(11:43):
models is be able to expand thatontology, right.
Speaker 4 (11:47):
Totally agree.
I mean, they were even talkingabout it earlier today, about
how much they're like.
They're like big, likecompanies are really focused on
making sure that they're stayingahead, right.
And so if you're staying ahead,you're going and pulling your
engineering team and productteam and design team and you're
saying, hey, what is what's thetechnology?
We need to know.
Speaker 3 (12:03):
We need to know, because that's what we need to
hire, for you want to hire.
Speaker 1 (12:06):
I go and then type that keyword into LinkedIn and
whatever else and that they'relike what do you mean?
Speaker 4 (12:10):
this technology Like I, like I love seeing.
Hey, I've been you, I'm a, I'ma generative AI expert.
I've been using chat GPT forfour years and you're like cat,
gpt hasn't been around for fouryears.
Speaker 3 (12:19):
That's yeah.
What are you talking?
About and and so, yeah, it'sexactly the keywords.
Yeah, so so the question that Iwanted to get to was the data
are you using?
Are you looking throughhomogenous data?
Is that's kind of where I not.
In homogenous data, things tendto get really messy.
Speaker 4 (12:39):
Yeah, so there are many different steps that are
necessary along the way right.
So a great example is we haveour database of 600 million
people.
Speaker 3 (12:46):
But those are people, so that's homogenous.
Speaker 4 (12:48):
Exactly, but that's coming from a number of
heterogeneous sources.
And then we part of ourtechnology stack is actually
going and sort of synthesizingthat down into a way that, hey,
data source A refers to thinksabout data in this way, data
source B does it in this way.
Speaker 3 (13:07):
And it's organized.
Speaker 4 (13:08):
And we have to organize and there's a lot of
mapping.
That's, frankly, just humans inthe loop going and cleaning up
sort of incredible amounts ofdata, and then there's having to
match this and then there's thebeing able to do it at scale,
right.
I think we're dealing with over200, or two billion data points
now to generate our 600 millioncandidates every six weeks, and
(13:28):
we're pulling in more sourcesliterally every time.
Speaker 3 (13:30):
We're building in our data set.
Yeah, and that's where largelanguage models have Are going
to flourish.
Yeah, that's what I'm saying.
Yeah, exactly.
Speaker 4 (13:36):
So if you can go and sort of homogenize the data like
the way you're framing it, thatmakes the task for the large
language model a lot easier togo and sort of do things on top.
Speaker 3 (13:46):
And predictable.
Speaker 4 (13:46):
That's exactly right.
And again, we all have these.
Like there's all these problemsright where it'll invent.
Speaker 1 (13:51):
Like it'll invent sort of Hallucinations yeah
hallucinations.
Speaker 4 (13:54):
A lot of people aren't familiar with the term so
I try to avoid it.
But yeah, exactly, we got toeducate everyone about it.
It'll hallucinate all sorts ofdata that's not there, and so
the simpler the problem andit'll sort of, it'll be easier
for it to compensate for it.
Speaker 3 (14:08):
The more congruent.
That's exactly right.
Speaker 4 (14:11):
And so if you're doing this right, there's going
to be a lot of balance between,like there's sort of this naive
solution which is I just go intochat GPT and I type in whatever
I want and I get my Boolean'sring.
That's the misconception andlike A, that'll work okay, but
to your point before, when wewere talking before, it's going
to be really general it's goingto be super generic and you're
not going to get sort of.
The whole point is that you'renarrowing in on what you want.
(14:33):
I often talk about the sourcingdonut where, like, you don't
want to find the center of thedonut, where everyone's looking
for the MIT software engineerthat we're in In Boston.
Yeah, in Boston that works atGoogle because, like everyone's
already reached out to thosepeople, you want to find the
actual donut.
You want to find the people whoare great for your company but
are only okay for sort ofeveryone else's company, and so
(14:56):
the more you like.
What we've basically found iswe're sort of have a very large
advantage because we've spentyears building out these
incredible data sets andpipelines that allow us to
homogenize the data, which thenallow us to lay other things
like other types of naturallanguage processing technology,
on top to sort of provide a lotmore power.
(15:17):
And if you just threw it at thehomogenous data sets, you
wouldn't get anything out.
Like every now and then, I seesomeone that's like we have two
billion, we have two billioncandidate profiles, and I'm like
, yep, you did not deduplicateyour data.
That is what I am hearing.
When you have two billionprofiles, it is not that you
magically have inventedcandidates that like don't exist
in the world.
Speaker 1 (15:37):
You have solved the labor crisis.
Speaker 3 (15:39):
That's right, yeah, you have two billion, but 1.6
billion of those are the same.
That is correct.
So, carrying it over to Mateo,what patterns are you
identifying in your side of?
Speaker 2 (15:52):
the world, so I'll be cautious, because Jeremy here
is an expert.
So what do I say?
No, but since we work withfront line workers, we
comeropyvs so far Many of theseworkers is from them.
Workers are offline, so there'sno data online.
Yeah great, so there's noLinkedIn data.
Wait a minute, they're not onFacebook.
They are on Facebook a lot, butyou can spread out under that
(16:14):
name and not under like.
You don't have detail abouttheir work history, what they
want to do, so there's like theyare sort of it's Facebook, like
every everybody's on Facebook.
Yeah.
Speaker 1 (16:24):
Unless, I know, I know everybody over 30.
Speaker 2 (16:26):
Yeah, but so for us, like that data said, like we
part of what we're doing, it'snot getting to the source in
space, so we're not looking toget new kind of a right.
Yeah, but we're looking to toway to explore things on how
data enables us to provide valueto our customers.
So, for example, like this,this workers coming this data
(16:50):
from, from from each one of ourcustomers, allows us to
understand what's the perfectcandidate profile for this
company, for this kind of worker.
So, for I'm we're talking aboutour employers.
They have like hundreds orthousands of workers, which
means that we process likemillions of candidates.
We got a counter in our booththat it's in real time that
shows how many from andcandidates will process so far,
(17:11):
and it keeps ticking and we areat 7 million candidates right,
which making your point.
It's not that we can talkgenerally speaking about data
points, but each one of ourcustomers.
They got a big chunk of those.
So we are playing around toplay somehow To see how we can
provide unique value to them andwhat are the?
(17:33):
yes, so you're finding patternsamongst that, exactly for
example and under there are moregeneric things that we are
doing which are more basic, whatyou are saying, which is, for
example, enabling a betterConversational interface with
the candidates.
So we get to understand withyou got, if you got seven
thousand, seven millionCandidates that went through the
similar conversation, you can,in case and understand how that
(17:56):
Be more preemptive, exactly,yeah, yeah it's one of the
biggest challenges I think withwith people in our space and
technology and HR it just ingeneral.
Speaker 3 (18:05):
Talent technology is ideation.
You know, the technology hasexisted For a while but people
just don't.
They haven't been able tofigure out how to apply it
because they don't really haveany ideas on what to apply.
Speaker 2 (18:21):
The problem set to the problem is when you start
with that, like what we I agree,complete with you like we
thought, first of all we need toresolve, solve a real problem,
which is the efficiency of thehiring.
Right in the process, we getall this data right, we make
these candidates which areinvisible, visible, yep.
And on top of that, we keepexploring new ways of telling
(18:41):
value.
But you can start the wayaround, like getting data and
then thinking, okay, what's thevalue that we'll deliver?
Speaker 3 (18:47):
because at least you can build a company like that
first value, you get data andthen you explore new ways and
it's to stack value on top ofthat and identifying that
essentially figuring out whatthe problem is that you want to
solve is not something that TApeople are really good at,
because they're looking at theproblem head-on, hiring people,
(19:09):
but they don't really.
Speaker 1 (19:10):
That's why I ask the questions about what wise the,
Not the five wires.
Speaker 3 (19:16):
Yeah yeah, exactly.
So you know the problem thatpeople have with LinkedIn
nowadays, becoming dependent onit and having nothing else but
LinkedIn as their only source.
That's not a new problem.
Before LinkedIn it was careerbuilder, before that it was
monster.
Before that it was net temps orthe fact, or online career
center or right, so that there'salways been this dependency on
(19:36):
that one technology, because itwas the ones, or solution,
because that was the onesolution everybody knew.
But the.
The origin of the problem isn'twhere do we go to find
candidates, it's when are thecandidates?
Speaker 4 (19:51):
And, specifically, how do I actually engage with
them right like?
Speaker 3 (19:53):
after you find those exactly like finding like people
you want.
Speaker 4 (19:56):
You want a thousand software engineers.
Here you go like it's supereasy to do.
You want.
You want a thousand softwareengineers that'll talk to you
very, very different problems, acompletely different problem.
Speaker 2 (20:06):
And the real problem is you need the software
engineers to run your company,to build product or, in your
software, your retail associateto sell on your retail.
So you need to go, I think,that deep and then go to the
problem like what's the realproblem?
Okay, having people, then whydon't you have the people?
Okay, because my Harry Majorstell me that I don't have the
people.
Speaker 1 (20:23):
And then you get Because they're not on the All
right, all right, we're notgonna beat up on LinkedIn too
much tonight.
All right, so real quick, justto recap we have been joined by
Jeremy at recruit bot.
We've been joined by Mateo atEmmy labs.
I've been joined by thegodfather of sourcing, mr Shally
stick.
Well, we are coming to you livefrom the olio booth at HR tech
(20:45):
in Las Vegas.
We wish you were here.
We're sorry that you're not.
Thanks for joining us and we'llsee you on the next episode.
Popular Podcasts
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.
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
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!