185 | Boost Sales With an AI Agent That Researches, Reports & Preps Sales Calls with Ashleigh Stearn - Leveraging AI

Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Isar Meitis (00:00):
Hello and welcome to another live episode of the

(00:03):
Leveraging AI Podcast, thepodcast that shares practical,
ethical ways to grow yourbusiness Improve efficiency and
advance your career.
This is Isar Meitis, your host,and we have a really exciting
show for you today.
We've said many times since 24,and definitely in the first few
months.
So this year that 2025 is gonnabe remembered as the year of
agents.

(00:24):
And this has turned to becompletely true.
Everybody wants agents,everybody wants to deploy
agents.
Everybody wants agents a part oftheir workforce.
The problem is the vast majorityof business people don't know
what agents are and even lessknow how to actually build them
and what it actually means.
What are the steps you need totake, which tools you can use,
and so on.
So what we're going to do ontoday's show.

(00:44):
Is demystify it and actuallyshow you how you can build your
own agents.
But we're going to dive evendeeper into that.
We're gonna show you a specificuse case on how you can build a
team of agents that can supportyour sales team, that can do
research, that can grabinformation, that can update
your CRM, that can do a lot ofthings that are going to help
your salespeople do their salesjob better.

(01:08):
Now, the really cool thing aboutall of this is it's done without
writing any code and in toolsthat literally anybody can use.
And even better is that ourguest today, Ash Stern, is a
certified expert and partner ofrelevance.ai, which is one of
the platforms that a lot ofpeople use to build these
agents, and she has beendeveloping and deploying agents

(01:29):
for multiple companies.
In multiple industries in thepast year.
So she has been doing this for aliving in the trenches, using
relevance for a while, and she'san amazing expert and a great
teacher that can explain exactlyhow she does it.
So I'm personally really excitedand I know a lot of you are too.

(01:50):
So Ash, welcome to leveragingai.

Ash Stearn (02:36):
Thank you.
Thanks so much for having me.
Very excited.

I am as excited as you.
I think I'm actually moreexcited than you because you
know exactly what you're gonnashow.
And I'm like, I want to see it.
I wanna see, and I'm sure a lotof people are the same way.
And so, great stuff for all ofyou who are joining us right now
on Zoom and or on LinkedIn Live,first of all, welcome.
Thank you for spending this timewith us.
We're very excited to have you,go introduce yourself, say where
you're from, say kinda likewhere you're in your AI journey,

(03:02):
what do you know about agents?
So we know where you are and wecan relate to that as we do our
show.
If you're not with us live, thequestion is why aren't you with
us live?
We do this every Thursday atnoon pm Eastern with amazing
people like Ash.
we're gonna dive with each andevery one of them into a
practical AI use case.
And if you're here a, you cannetwork with amazing people who
are business people who careabout ai.

(03:23):
And also you can ask questions,which you cannot do if you're
doing this while listening afterthe fact.
So that's.
One item.
The second thing that I wannasay before we dive into this
topic is that we.
Have started the registrationfor the next cohort of the AI
Business Transformation course.
It's the course that we havebeen teaching for two years now.
It's crazy that it's been twoyears, but since April of 2023,

(03:45):
I've been teaching this coursemyself.
I.
Online on Zoom with cohorts ofpeople.
And so this cohort starts on May12th.
There has been hundreds ofbusiness people that literally
transform their businesses andor their careers with the
knowledge they acquired in thecourse.
And so the course obviouslyupdates every single time we run
it.
We run it mostly privately fordifferent companies and

(04:07):
organizations.
I'm actually teaching twoprivate courses in parallel to
this public course, but thepublic course starts on May
12th.
And we do the public coursesonly once a quarter.
So if you haven't done anyproper training for yourself,
you owe it to yourself to doproper AI training so you can
really transform everything thatyou're doing across multiple
aspects of the business.
there's gonna be a link in theshow notes, and then we are

(04:29):
gonna drop the link also in thechat right now.
And if you use promo codeleveraging AI 100, so the name
of the podcast, 100, you willget$100 off just for listening
to this podcast.
How cool is that?
So that's it about the course.
And now let's dive into agents,how they're built, how they're
coordinated with themselves, andhow they actually are working.

(04:50):
Ash, it's all yours.

Ash Stearn (04:52):
Awesome.
All right, cool.
So what I'll do is I'm gonnajump straight into the platform
and actually show you, but kindof explain everything as we're
going.
and I think that's the best wayto do it.
I do like presentations andslides, but there's nothing
better than actually putting twoand two together by actually
going into the platform itself.
So I'm just gonna, share myscreen and we can get to that.

(05:14):
Okay.
So what we've got here is we'regonna essentially start off with
the team that we've got Now, ourteam currently is sitting under
our manager, which is managerMitch.
Now when we come down into thisis the agent dashboard,
essentially, this is what you'regonna see as soon as you start
creating your very own agent.
now before we kind of get into,I guess, the details of how it's

(05:37):
all set up, I'll essentiallyshow you, what it looks like.
All right.
So just some context here.
this was set up, and wasconnected to my Calendly link.
So what would happen is thatsomebody would book with me
through Calendly, they'd fillout a form, which included
obviously these details thatwe've got up here.
And then this would actually betriggered and sent directly to

(05:57):
my agent to start the entireresearch process.
So when we click, so this is theinformation, we click down and
we can actually see our manager,which is who we're currently in
at the moment.
I'll pause you just

for one second.
For those of you who are justlistening and not watching this,
two things.
One, there's a YouTube channelthat you can find very easily by
clicking in the YouTube channellink in your show notes.
So if you wanna watch this, andyou can do that.
But if you're driving or walkingyour dog, or doing the dishes or
whatever it is that you'redoing, listening to podcasts or
yoga or I don't know what you'redoing, we'll explain everything
that's on the screen.
So in this particular case,we're looking at a screen in,

(06:30):
Relevance and on top there'sinformation that came in that
triggered the agents to startworking.
So it's the email, first name,last name, meeting type.
Again, all the information thatbasically come Oh, time zone,
the information that comes infrom a meeting that you set up
on Calendly.

Ash Stearn (06:46):
Right.
Right.
Exactly.
So, so with that informationthat we've got, our agent, you
can actually see the process ofour manager agent delegating
this information to oursub-agents.
Now we've got two sub-agents.
Okay.
You can actually see here thatit's delegated to our lead
researcher and then delegated toour CRM update agent.

(07:08):
Okay.
So essentially what our manageris doing is that it's
communicating, it's beingprogrammed in a way to
communicate and delegate tasksto our agents.
Our agents do what they need todo.
So obviously our lead researcheris re doing all sorts of,
research based on our lead.
And then we've got our update aswell into our CRM.
So this is the information thatwe get from all of that

(07:31):
research.
So information includes, theemail, job title, LinkedIn
profile.
We've got company information,such as the website, LinkedIn
page, a company summary.
And then we've also got apre-sales call report.
Now, I really wanted a pre-salescall report written for me so
that I was very well preparedbefore going into the discovery

(07:51):
calls.
Now there's nothing more offputting than going into a sales
call and just not knowing whoyou are talking to and just how
you can really serve them.
So our pre-sales call report,there are two ways that you
could do this.
You could do it within relevanceai, but at this time, when I
did, this particular setup, Iactually had it connected to
Make, now, if you haven't heardof make is an automation

(08:12):
platform.
It's a very popular one.
So it's a visual builder and itallows you to visually build out
workflows.
Essentially make is you areselecting a whole bunch of
different nodes with make, andthey are connected, right?
So you're connecting all ofthese different components in
make that make up an automation.
Okay?
Now, when creating agents, youneed to understand that

(08:34):
automation is a part of that.
and so I set up a littleautomation.
That's added my presale scorereport to a Google document.
Now, the wonderful thing aboutrelevance AI is that they've
now, added in a whole load ofintegrations that you can now
add, to your agents directly.
And one of those is Googledocuments.
so essentially you're not gonnaneed any of these external
integrations with like otherplatforms, like Make and Zapier

(08:57):
'cause, like one that'sannoying.
Two, it's technical as well ifyou're not familiar with those
platforms.
but just to show you how thereport was actually structured,
this is what it looks like now.
This is actually being donewithin Relevance ai.
that will actually give me thelink to, to my Google Doc
without going through make andstuff.
So we've got top expert themes.

(09:18):
We've got a pre-sales, strategy,so rapport, building elements,
professional achievements.
Then we've got pain points andsolution alignment.
Value proposition and thenshared content and priorities.
Now, I like this because I liketo see how active my lead is on
LinkedIn, and if they're activeon LinkedIn, obviously if
they've got recent content, itreally paints a picture of like,

(09:39):
what are they interested in and,in terms of business and what
they're doing.
So I like that.
so essentially this is what ourreport, looks like.
now if we go back into quickquestion.

Yes.
on the pain points side.
Yeah.
how does it know their painpoints?
Does it do research on its own?
Is it based on what they wrotein the form?
Is it a combination of thesetwo?
Like, what is the process to getthat.

Ash Stearn (10:00):
how the pain points are done is that you actually
have within your particularagent who's creating the
presales core report.
In this case, it's my leadresearcher.
They actually have myinformation.
So my information is thencombined with their information
and just the research that it'sdone based on that lead so that
it can perhaps find some painpoints where my services and

(10:22):
what I offer is going to be ofvalue to them.
I have found that sometimespeople will actually post
content on their pain points,like what are they struggling
with in their business and stufflike that.
So that is another indicator ofpain points as well.
but our lead researcher does allof that for us, which is really
cool.

And we're gonna dive into that afterwards Right,
exactly.
To see what it's doing under thehood.

Ash Stearn (10:43):
Yeah, exactly.
So we'll get into more detailabout that.
Right.
So that is essentially, ourteam, a little mini team of
agents.
Now, it's not a big team.
Some of these teams can getquite complex.
but one of the main things weneed to keep in mind here, and
what we'll do is we'll actuallygo into the lead researcher now,
just so I can really show youexactly how it works.

(11:04):
Now, within relevance ai, youneed to try and understand
what's your workaround, how areyou going to actually build
agents, what's yourinterpretation of agents?
how do you wanna approachbuilding agents?
Relevance ai, as you can seehere, is really based on prompt
engineering.
Now, this is one of the thingsthat people need to understand

(11:24):
is that if you're going to buildagents, whether it's with
relevance, ai or anotherplatform, prompt engineering is
a part of that.
Your agents are built on, LLMmodels.
So prompt engineering.
Just to set an expectation, hereis a skill that you'll need to
have when building any kind ofagent.
If you're using LLM models, youneed to know prompt engineering.
so relevance ai, the agents herethat we build, the brains of

(11:47):
them are the core instructions.
So the core instructions areessentially, one big prompt.
And in this, particular case,what we use is a prompting
technique called chain ofthought prompting.
So it is a very common way toprompt LLM models these days.
What it allows us to do is thatit allows us to really help our
agent understand the process itneeds to go through in order to

(12:10):
get the best output.
So as we're looking at our leadresearcher agent, we can go
through this prompt and see howI've actually programmed it to
work.
All right?
So we've got a role, so we'regoing to always give it a role.
we then have an objective, okay?
We then have context as well.
And now what's different with anagent prompt in relevance AI is
that you're gonna have an SOP.

(12:31):
Now usually if you're just usingthis prompt with an LLM model
like chat, GBT or clawed, orwhatever else it might be,
you're not usually gonna use anSOP.
You might use instructions,okay?
And only just instructions.
but in this case, because we'rebuilding an AI agent, an SOP is
good to have in there.
So we have that.
And within this SOP we're reallygetting quite granular on the

(12:54):
different, steps it needs to bedoing, the, process it needs to
go through.
So we've got our tools tagged inthere as well, which you can add
by just doing a backslash and itwill actually add in the tool
for you.
And then we've got just somevery brief overall instructions
of just the overall tasks itneeds to do for us.
Then of course, we list, so Iwanna pause you just

for one second.
Yeah.
So, how did you come up withthis, right?
Because you've been doing thisfor months and you have a lot of
experience.
If I'm just getting started.
And I wanna build one of those,how do I get started from
scratch?
First of all, do they havetemplates scratch, or do I ask
Chachi PT for recommendations?
Like what's the best process ifsomebody just getting started,
how to figure out, because wejust named the topics, right?

(13:37):
We said, okay, there'sobjectives and there's context
and there's SOP and there's ageneral instructions and then,
but what actually goes in them?
And we can obviously read thisto everybody, but that's not the
point if, because people wannabuild something else, what you
built, right?
So how do you get started?
if you don't know what you'redoing,

Ash Stearn (13:51):
okay.
If you dunno what you're doing,then I would actually get
started with what is a processyou're currently doing for your
business that you don't wannado.
so if you have a processcurrently that you don't want to
do and you want to, streamlinewith AI agents, then that is the
first starting point.
That's where I started andthat's why I built this lead
researcher.
I didn't wanna waste timeresearching leads.

(14:13):
I wanted to actually take moretime in preparing for my meeting
and doing other things likebuilding agents.
So what I actually recommend topeople is going through the
current processes that they'redoing and seeing what can
actually be, automated straightaway with AI agents.
and then go from there if you'vegot the process in detail,
right?
Even better, because then you'rereally getting quite granular on

(14:36):
what is the process, what needsto be automated, and how you're
gonna build towards that.
Now with your agents, you've gotbuilding your agent for one
particular process.
In this case, what we're lookingat is we are researching leads,
but within the lead researcheryou've got.
Tools.
Now your tools are, what areyour automations?

(14:57):
And now this is what I mean byunderstanding that.
building agents, you have todevelop multiple skills.
so if we take a look at, let'ssay our pre-sales core report
tool, which is one that isresponsible for actually writing
out report, you'll actually seethat this is an automation and
it's very similar to what you'dsee in another automation
platform like Make or Zapier.

(15:19):
and essentially our tools arethe skills for our agents to
leverage and use to performtheir tasks.
But we have to build these.
So when you think of a leadresearcher, you think that's
their role.
So you think of a role in ahuman sense or a human
department sense.
Let's say for example, what Iwas doing was content production
or Social media management.

(15:40):
Within that social mediamanagement role, I had to.
Write content.
then I'm gonna build an agentthat's gonna write content.
So it's very similar in thiscase where lead research, that
was one entire role.
I didn't wanna do that.
So I'm going to, automate thatwith an agent.
And then what are the mini taskswithin that role?
So we have research, we havegenerating the presales core

(16:01):
report.
and then we also have verifyingthe leads email as well.
those little mini subtasks, youthen have to create tools to do
them.
so

I want to pause.
You wanna pause you for twothings.
one yeah.
Is I want to give some idea forpeople'cause you said something.
That is brilliant and we didn'tdive into that.
And I think it's important tosay, start with by mapping your
existing processes and peopleare like, yes.
Oh my God, that's a great idea.
But like, how do I do that?
And so, the way I work with myclients when we map processes is
I literally ask people tonarrate what they're doing as

(16:32):
they're doing it.
So open your screen.
Open the tools you're using.
Hit record on whatever recordplatform that you want to use.
You can use, the usual suspects.
but you can use Zoom.
Like you can have a meeting inZoom on your own and just run
that, and just record the screenand explain what you're doing as
you're doing it.
Do it three times.
If there's multiple people doingit, ask each and every one of

(16:54):
them to do it three times.
Upload all of that into yourfavorite LLM Could be Gemini
Clause, Chachi pt.
I don't care.
And say, Hey, I want you tocreate an SOP of this process.
Here are transcriptions ofrecordings of Six different
occasions doing it.
'cause then.
People do it a littledifferently, different times,
and it will create an amazingSOP.
Then you wanna ask it, to askyou questions to figure out if

(17:14):
there's any gaps.
It's gonna ask you questions,you're gonna answer them, and
you have your SOP.
That could be a very solidstarting point for your agent.
So that was, I wanted to add,what I want to ask about these
tools is I assume like a sectionhere that is creating these
tools, right?
And so can you explain what doesthat mean to create a tool?
show us a little bit more detailon that?

(17:36):
Absolutely.

Ash Stearn (17:37):
Yeah.
So when we come into ourdashboard, this is relevance ai.
This is your profile.
This is how it's all gonna lookwhen you're in here.
You're gonna come to your toolsection, and now you're gonna
see a whole bunch of tools, thatI've actually created.
all of these tools are wherethey, all live.
now you can build them fromscratch, but you can also use

(17:57):
templates.
relevance ai, when they firststarted, they actually only had
six tool templates.
now they've got an entirelibrary of both agent templates
and tool templates.
these are great to start using,especially as you are learning
and getting started.
And in fact, how I learned,quick or quicker than, trying to

(18:17):
rely on documentation.
There's documentation back thenand content.
To learn how to do this was, notvery good'cause it wasn't a very
well known platform at the time.
But I had six tool templates andall I did was I clicked on one,
let's say for example, performGoogle search.
So I clicked on it.
I then have to, clone the toolto be able to see what's going

(18:37):
on in the back end of it so Ican understand how they actually
did it.
And then I would actually createa tool completely from zero from
scratch.
And I just copied, so I justcopied these steps.
in our tools, we've got ourinputs, and then we've got these
different steps that we alsohave to add in there as well.
Okay.

(18:57):
Now this one's a very, a smalltool, but it's, so it's
basically

like a smart flow chart, right?
You start with an input and thenwhat if, or what steps you need
to take.
And that becomes a tool thatthen the cool thing is, I guess
they're all reusable because anyof the agent can use any of the
tools that you created, right?
So if you create them generic,then you can use them across
other tools as well.
So in this particular case, aGoogle research, while you wanna
do research across probablymultiple different kinds of
agents.

Ash Stearn (19:20):
Exactly.
and obviously a Google search,tool can be used across.
All sorts of domains as well, soit's not, tied to one domain
either.
so they can be used quiteregularly by various different
agents as well.
And what, something is reallyimportant here to know is that
how do our agents actually usethese tools, right?
So we know that they get giveninformation, okay, but they

(19:43):
actually dynamically give or usethat information that has been
given to it, and it willactually fill out the inputs.
By themselves.
We don't have to put in anyinformation or anything like
that into this area.
Our agent actually does all ofthis for us in order to use the
tools, effectively.
so they dynamically add theinformation to our tools for us.

(20:07):
and it doesn't matter what typeof input it is, it could be a
text input.
So in this case it's gonna be aGoogle search query, but then
we've got various otherdifferent types of inputs.
So numbers, we've got files.
We have an options menu.
So yes, no, and our agent willactually fill all of these
different types of inputs out byitself.

(20:28):
what I would generally do withthis was that, okay, so I cloned
it.
I then go back into mydashboard.
I'd go back to my tools here.
And then I would create a newtool.
And what this helped for me whenI was learning how to build
agents was that it allowed me tounderstand what needed to go
where.
So I would go back into the toolthat I've cloned, I'd

(20:49):
understand, okay, well this isthe title, right?
And I'd simply just copy andpaste When I needed to add a
step, which was Google Step, I'dcome into here, I'd add a step
and I'd click Google Search.
Okay.
now there are a lot of littlenuances to it, including
variables.
So it does get quite technical.
And this is another thing I wantto set an expectation for as
well.

(21:10):
It's technical in nature.
it is something you're gonnahave to learn and the learning
curve can be quite steep.
so in saying that, cannon-technical people, build AI
agents?
Absolutely.
You are looking at one.
I was writing blogs before I wasdoing this, so yes.
But don't underestimate thelearning curve, because if you
consider yourself anon-technical person now, well

(21:30):
by the end of all of this you'llbe considered a technical
expert.
So, just wanna set anexpectation there.
'cause it's not, somethingyou're going to probably learn
in two weeks.
depending on your learningstyle.

I wanna add two things to what you said because
I think it's really important,two aspects of what you said
that I think that are critical.
One, it's perfectly finestarting by copying.
Others, and others could betemplates from the platform.
Absolutely.
Or people who share how they dotheir thing.
And a lot of people do that.
That's what we're doing rightnow.
Right.
Ash is literally showing you theagent, like you can watch the
video copy word to word what youwrote, and you have an agent, a

(22:04):
starting point.
So it's perfectly fine to startby copying others because it
allows you to understand thenuances and the processes and
the little things that if youjust try it on your own, will
take you a lot longer.
Yeah.
And two is the ROI is stillthere.
Yes.
We're like, oh my God, now Igotta invest a month to learn
this.
Yeah, but this is gonna save youa month, every month.
Then it's worth investing thefirst month in learning the

(22:27):
platform.
On the bigger, broader sense,from a company perspective, from
a team perspective, you need oneperson who needs to know how to
do this.
You don't need everybody on yourteam.
You don't need everybody in thecompany to know how to do this.
And there's two ways to do that.
One, you can hire Ash, right?
you don't have to have thatperson in house.
You could be an external personthat knows how to build that,
that can build it for you.

(22:48):
two, you can get one of yourteam to be certified or train on
how to use, either relevance orother platforms to build agents.
And then that person can be thego-to person in the company, to
build a platform.
That obviously means they needtime and resources and other
stuff, but I'm putting thataside for a second.
the learning curve is worth it.
Because the yes variety ofagents you can build can help

(23:08):
across almost everything in yourcompany.

Ash Stearn (23:11):
Yeah, and I will just add something here as well,
with the speed of how everythingis advancing at the moment.
These are only gonna becomequicker, more reliable.
They're gonna get better.
and they're eventually gonna becheaper, hopefully in the future
as well.
so it is something that isconstantly improving.
And I've seen relevance AI wherewhen they first started to where

(23:32):
they are now, and just theamazing leaps that have come in
the, incredible updates thatthey've done recently.
They're making it, the platformis specifically targeting
industry experts.
they're not targeting technicalexperts.
It is a platform that they'retrying to build into something
that anyone, any industry expertcan come in, say they want an
agent and have the agent, built.

(23:52):
they've actually got a newfeature that was just, released
not too long ago where you canactually create, you can invent
a tool.
You can also invent an agentjust by putting in a prompt.

and you will write all the details and all the
steps and all the other stuff.

Ash Stearn (24:09):
it's gonna write the core instructions.
Now the core instructions andthe prompting that goes into
that is actually a very creativeside to agents when you're
building them with ringrelevance, ai.
And I always say to people andrecommend, experiment with it.
just because my way has workedfor me doesn't mean that another
way of writing the coreinstructions is gonna work.
or it's not gonna work.
it, there's no right or wronghere.

(24:30):
And I think that just goes toshow that the platform is
flexible and the agents aren'tintelligent.
They will get it.
As long as the instructions areclear and they're clearly
articulated, they will get it.

let's do two things because I'm very curious.
I'll start with a question andthen that will probably guide my
next step.
When the agent is actuallyrunning, can you see what it's
doing?
Like, can you follow the stepsas it's doing the steps?

Ash Stearn (24:56):
what we can see happening once, a task has been
triggered, we can see what'sbeing performed in the
background.
So these are the tools that thisagent is using.
so it's used, obviouslyverifying the leads email first
because in this case I wasgetting both, leads coming in
that were filling out with abusiness email as well as a
personal email.
Now, that was really core in howwe extracted the domain to

(25:18):
understand the company that theyworked for.
So I had to verify and then itwent through then and research
the lead if the email was abusiness email.
And then when we actually clickon the tool that's being, used
in this case.
This one here, we can actuallysee the inputs that have gone
in.
So our agent has done that forus.
We can actually see then theoutputs that have been generated

(25:41):
from this tool.
We can't see what's going.
So again, for those

of you who can see, it got an email address.
It checked if it's a businessdomain, and then it went and did
the research on the company byextracting the domain name.
So he came back and said, thisis a, financial services
company, the company URL onLinkedIn, A summary about the
company, and stuff like that.
Probably top people in thecompany and things like that,
right?
So, decision maker yes or noabout this particular person.

(26:05):
So it pulls all that informationbecause it is a company domain
and then it knows how to findinformation about it, and that's
very cool.

Ash Stearn (26:13):
Yeah.
Yeah.
And this was another thing thatI wanna mention as well, that
sometimes you don't think of thechallenges or the obstacles that
are gonna come up until youactually start testing your
agent.
I had to think of a workaround.
I have to now understand thatwhat if someone uses their
personal email?
Obviously if they're extractingthe domain of Gmail or outlook,
it's not going to do the properresearch because that's not the

(26:34):
company that the person worksfor usually.
so, the tool that I've done inthat case is gonna be structured
slightly more different to whathow this one is structured.
'cause there needs to be aworkaround for how it's going to
figure out.
Where the lead, who the leadworks for.
and that's something I didn'treally put two and two together
until I actually started testingthe agent out, and then had to

(26:55):
accompany for that.
So yeah, it's just interestinghow it all falls into place and
you gotta change things.

so the first step, it gets an email, it researches
the email, it gets this kind ofinformation.
What does it do next?
I just want to go step by stepso people understand what the
agent actually does on its own.

Ash Stearn (27:10):
Sure.
What we can do, we'll start offwith the verify email first.
So let's open this one up andthen we'll get into the research
tool.
So this is all coming from thelead researcher, by the way.
these are all the tools that arehave been added to this lead
researcher agent.
So the first step it's alwaysgoing to do within that SOP and
the core instructions I've laidout, it's gonna use our verify
lead email tool.

(27:30):
This one is just a very simpletool, is an LLM step in here
that we've added.
We just need to clarify, is thisa personal email, or is this a
business email?
Okay.
So in this case, we've addedjust a simple LLM step and give
it instructions.
So it's not a very big prompt,it's just a pretty medium sized
prompt instructions of how itneeds to figure that out.

(27:51):
Some examples as well.
Examples are always very handyto have in your prompts.
And then just some, notes don'tgive me pretext or context and
only, output the correctclassification.
And that's all we need to do forthat one.
Once it's done that it's gonnaoutput the, Whether it's
business, the Yeah, whetherpersonal, whether it's business
or personal.

Yeah.

Ash Stearn (28:10):
Right.
Exactly.
And based on that answer, ouragent is then going to use the
next tool.
So in this case it is a businessemail.
So it's now going to use theresearch.
So this is very quite clear.
Research lead if the email'sverified as business.
Okay, so this is the title ofthat tool and then the
description of that tool.
Now this one is a very big tool.

(28:31):
you can see on the left handside it's lengthy'cause it's
doing so much.
Now this rolls into how complexdo our tools need to be.
Now I could have quite easilyhave split this tool into two.
So we researched the lead first,then we researched the company.
But in this case, I found iteasy enough to just put
everything into one tool.
That's what we did.

(28:53):
so when I was mentioning beforeas well, just having our company
information, this is the companyinformation that I've added that
is relevant for our research.
So we can see here now the fullname has gone in, the leads
email has gone in and our agenthas added, has put all of that
in for us.
And then we're gonna go throughall of these different steps.
Some of them is research, someof them is manipulating the

(29:16):
data.
some of them are LLM steps.
And these particular ones nowget into our LLM steps.
So this is the LLM steps here.
Really collecting all of thatresearch, just doing stuff with
it so that we can then generateour outputs at the end.
Now our outputs at the end,there's a lot of them.

(29:37):
Okay, so there's all of thesedifferent outputs.
Now, if we actually, so again,just

to give examples to people.
We have lead, job title, leademail, company name, company
revenue, company size, all thesedifferent things.
These are the different outputsthat this agent is going to
output.
After doing all the research,filtering all the filters, and
packaging the information in theway that will be useful for the
continuation of the process.

Ash Stearn (29:59):
Yeah, exactly right.
So, like I said, this tool couldhave easily been split into two
different tools.
It didn't have to be all packedinto one.
but in this case, I've just putit all into one tool.
just because that was just myway of thinking.
Now my way of thinking is notgonna be the same as somebody
else's.
there's not one right or wrongway to create your tools.
If it's giving you the outputyou want and you need, then

(30:22):
nothing else matters.
You just gotta get it workingand that's it.
and it doesn't matter how itlooks either.
so in this case, as we can see,like we've just run our tool,
it's done all of the stuff thatit's needed to do and this is
what the outputs are.
So we've got our company,LinkedIn, URL, the industry
number of employees, the leadsummary LinkedIn, URL, and then

(30:42):
all this other, data companyrelated, sorry, lead instrument.
yeah, exactly.
And so next our agent is gonnatake all of this information,
which is a lot of information,and it's then going to generate
our pre-sales or report.
and then what this one does, itgoes through a whole different

(31:02):
process of, again, just doing, alittle bit of content research
on their LinkedIn, to see ifthey've got any recent content
that they've posted.
And again, it's going to thengenerate our pre-sales core
report, which is what we saw inthat Google document sheet.
Now, just to reference again,just to show it again, this is
what that looks like for us.

Yeah.

Ash Stearn (31:24):
Right.
So this is now all of thatresearch all into one thing.
We've done a whole bunch ofdifferent stuff with it, and
then we've generated thisreport.
Now after it's done, whathappens then, after our lead
researcher has actually done allof this, it then sends all of
the lead information to our CRMupdate agent.

(31:45):
Now if we view the conversationhere, we can actually see what
information has been passed toit.
Okay?
So in this case, it is, thisparticular information that we
want uploaded into our CRM.
It's used only one tool, andthat is just, running some
different steps that areconnected to my CRM that I use.
And it's, capsule.
Is it capsule?

(32:05):
It's capsule I think.
and then this is the informationthat is uploaded.
Now.
I've seen people obviouslyupload it into, more well-known
CRMs like HubSpot.
And within HubSpot I actuallyset up for a client, to have
this pre-sales core reportactually then uploaded into, I
think it was the notes section,of that, new lead that has just
been, input into their CRM.

(32:27):
so there's various differentways that you can use this as
well.
This is just my particular setupthat I had for me.
And then that's it.
And then our manager at the endof all of that will just give us
a little update.
the information that's beenupdated into your CRM, just for
our reference and here's thepresales core report.
but what I was doing is I wasactually already seeing that
within the CRM, so I wasn'tusually coming into the
dashboard at all.

(32:48):
I straight after a call, I wouldthen go into my CRM, just make
sure that the new lead has beenuploaded into there, and then go
into my Google Docs and I'll seethat document that was in my
Google Docs already.
and then that's when I'd readit.
but for my discovery call.

That's fantastic.
I want to touch on a few pointsto expand people's horizon
beyond, yeah.
this, and this by itself isincredible.
So quick recap and then I'll addmy 2 cents.
So the recap is, relevance isone of several tools that are
for non-developers to be able tocreate agents.
And the way you do this is youcreate instructions for
different, a, well, first ofall, you map your processes, you

(33:24):
understand what needs to happen,you define the different agents
that needs to work, and youdefine which tasks it needs to
do.
And for each task, you basicallyneed to generate a tool to allow
them to do it.
Then you orchestrate it justlike a regular team in this
particular as a manager, andthen there's a researcher, and
then there's the update, theCRM, kind of thing.
But you can add stuff beyondthat.
So if you think about like thebroader process in a company,
the next step would be to, Do alead score and decide what

(33:47):
should be their next step in theCRM itself.
So are they ready to buy?
Do they need to be nurtured?
Do, are they ready for a callwith a sales agent?
Like what should be the nextstep?
And you can analyze that with anagent and update the CRM
accordingly.
You can even generate.
The content that they need toget as the next step.
So let's say you wanna send theman email, instead of just
sending them a generic emailfrom your CRM that's just

(34:10):
automated, you now know all thisstuff about them.
So you can send a very personalemail that's gonna be uploaded
to the CRM and send from the CRMjust like all your other emails.
So it kept being tracked andeverything else that you get
from the CRM, but it'scustomized to this particular
person based on now you want totake this to the next level.
You can connect this to yournote taker.
So I use Fathom, but there's agazillion other note takers.

(34:31):
So you can grab the actioninformation from the note taker
on the actual meetings you hadwith them and other people from
their company.
And that could roll intodeciding what's the next step
that could go into the nextpiece of content that it
generates and so on.
And over time you add more andmore of these steps.

GMT20250424-154653_Rec (34:48):
Exactly.

And more and more these agents.
So.
One of the problems, the thingpeople have is they look at the
big picture and like, oh my God,how do I even get started?
This is insane.
But once you break this down,like, okay, and I'll go back to
what you said because I thinkit's awesome.
What is this thing I hate to dothe most or waste the most
amount of time for me or for myteam?
And start with that.
And then, okay, what's the nextannoying thing that I don't

(35:09):
necessarily want to do?
and I think it becomesinteresting because that makes
you decide where you're actuallyproviding significant value.
And where you're not and whereyou're actually providing
significant value, you couldstill be a part of the process.
You could have this send you anemail and say, this is what I'm
thinking of sending them as anemail.
Please make your suggestions andcorrections and send it back to

(35:30):
me.
And then you email back theagent and the agent's gonna say,
oh, awesome.
And then it's gonna keep ongoing from there.
And then you've given your inputon what you talked about with
this guy when you're havingcoffee the other day and you
talked about his, him and hisdaughter going on the strip.
And you can add that stuff thatthe system just doesn't know, or
won't think that it's importantto add.
So this is where you startadding the human inputs and just

(35:51):
taking away the tedious workthat is just tedious.
Like doing research is not,you're not providing any value,
right?
Like these tools will do betterthan you every single time if
you created the agent, properlyand it's gonna do it in 10
minutes instead of you investingthree hours.

Ash Stearn (36:06):
Right.

And so, exactly.
You start, need to startthinking on how to break down
the processes that are happeningin your company.
I shared earlier how you can dothat and then start doing small
steps.
Build the first agent with thefirst three tools, tweak it, and
tweak it again, and tweak itagain.
And you're like, okay, this isawesome.
This actually does this part ofthe work.
Then go to step two, step three,step four, and add more agents

(36:26):
and more stuff.
and this is how you take awaymore tedious work from more
people and making them happieremployees and making them
actually work on stuff thatmatters versus doing research or
updating the CRM.

Ash Stearn (36:38):
Yeah.
Yeah.
I think you really banked it,like you nailed it on the,
right.
Spot on with all of that.
I think what I love about, doingthis kind of stuff is that you
can just gradually add to yourteam.
you don't have to stop at justwhat I've shown you.
Oh my goodness.
The stuff I've built since thathas been huge and I'm just
constantly adding to it.
thinking of differentdepartments that I wanna have,

(36:59):
and how is that all gonna bestructured?
And then of course, what youcreate, and you, right now it
doesn't have a purpose or ameaning to it, but you might
find a meaning and a place forit in the future.
it can so easily, happen likethat, And I think it's, quite
creative.
As much as it is technical, itis quite creative as well.
in that, the core instructions,I know that I keep going on
about it, but it really is whenyou think of like how it's

(37:21):
supposed to work and function,it's so, language based, so be
creative with that language andhow you're giving it and how
you're instructing it.
And there are so many differentways and approaches to it, I
love that it's not rigid it'snot the workflow.
The LLM model is the center ofit and the brains of it.
And that's what I love.
yeah.

Question about LLMs.
There's a question fromCatherine that's saying, are
LLMs any good at walking youthrough how to build these
agents and set up all the steps,or is it too new?
non-tech, I built an app on theweekend, which at GPT, but
basically the question is, canyou go to Chachi, PT or Gemini
or Claude or whatever and say,Hey, I'm working on this agent.

(38:00):
Help me build up the coreinstructions as an example,

Ash Stearn (38:03):
right?
So what I would do if you wannado that, and I've done that with
other tools, not so much withrelevance say I, but other tools
I've done, I'm like, please helpme out here.
Like, what am I supposed to do?
you've gotta provide as muchcontext as you can.
So if the tool, for example,relevance, they have a lot of
documentation.
You take their documentation,you copy paste it, however that
looks in a document or directlywithin chat GPT, for example,

(38:24):
and give it as much context asyou possibly can give it.
and then ask the question, allright, I wanna build an agent
for this particular process, tryto be as detailed as you can in
that process.
Just like what you mentionedbefore about going through the
process and the details andmapping it all out.
Give that then to chat GBT andit will be able to help you,

(38:44):
think of how you can approachbuilding it.
It might not understand thenuances to it, the little
things.
those are things that you willjust have to learn as you go.
But in terms of structure, interms of plan, yeah, it's gonna
have it all, just context.
Just provide as much as you can.

Couldn't have said it better.
This was perfect.
Context is everything to thesetools, and the more you give
them, the more, the better andmore relevant the answers are
going to be.
the two things that I will addto what you said one thing is
today, you don't necessarilyhave to copy and paste stuff.
You can give them the URLs tothe documentation.
They know how to read web pagesactually really well.
the other thing that you can do,which I started playing with,
not necessarily with amazingsuccess, but with some success,

(39:20):
and I think that's the future ofany tool user manual, is I
started using Gemini.
So Gemini in their, experimentalversion.
So if you go to.
AI studio.google.com orsomething like that.
basically their experimentalenvironment.
There's a live option and youcan share your screen with it,
and then you can see what's onthe screen.
And then you can literally justtalk with your voice, say, Hey,

(39:42):
I'm developing this agent.
Here are the instructions.
This is what it's doing.
I need your help in writingbetter instructions for it to
avoid the thing that it's doing.
And then it's actually seize thescreen to get it with you.
And you're having a conversationwith your own voice, just like
there's a person there.
Now, I think what's gonna happenover time, I think that's gonna
be.
The user manual.
There is not gonna be a usermanual.
There's gonna be, okay, walk methrough what I'm trying to do.
And it will explain to you andhighlight things on the screen

(40:04):
live as you're doing the thing,because you're gonna have the
ultimate expert there with youat any given time.
Right now, Gemini is limitedwith what it knows about it.
But again, you can tell theGemini, okay, let me open the
user manual here and readthrough it, and then let's try
to figure this out together.
what I found is when I run itfor about 10, 15 minutes, it
just crashes.
So when I'm finally almost aboutto solve the problem, but I did

(40:25):
have a few successes when it wasshorter sessions and it's
magical because it's just, yeah,there and it's an expert and it
can help you and give youguidance and so on.
It's actually looking at thescreen together with you, so
that's another thing that youcan do.
Ash, this was fantastic, like, Ithink as an introduction for
people to the world of agentsand how they work and how
they're built and theintricacies of, what's required
to build them.
This was an awesome session.

(40:45):
If people wanna follow you, workwith you, learn from you, hire
you.
What are the best ways to dothat?

Ash Stearn (40:52):
look, just LinkedIn.
I'm very active on LinkedIn.
I post quite regularly.
I have, also just, opened thedoors to my new community for
women in AI as well, where Iactually teach beginners and
non-technical, people how tobuild AI agents with relevance
ai.
I only stick to relevance AIbecause there are so many tools
out there.
I'm like, no, let's just keep itless as stress free as possible.

(41:13):
so yeah, I just really focus onrelevance AI for that.
But, but yeah, LinkedIn.
Yeah.

Awesome.
Perfect.
anybody who's joining us onLinkedIn or on Zoom, I really
appreciate you.
I know you have other stuff thatyou can do on Thursdays, early
afternoon in the US or whereverthey are in the world.
There's usually people from allover, joining these sessions.
So thank you so much forspending this hour with us.
If you're just listening in yourcar or in your headphones or
whatever, I appreciate that aswell.
we just crossed 250,000downloads since this podcast was

(41:41):
launched, which is an incrediblenumber to me.
but one of the things that wewanna do is we wanna learn what
you want more and less of thispodcast.
So in the show notes, there'sgonna be a link for you to.
take a survey.
It's gonna take you a minute totake the survey and it will give
us feedback on what you like anddon't like and will allow us to
focus and do the podcast evenbetter for each and every one of

(42:02):
you.
So I appreciate you being alistener of the show or joining
us live.
I really appreciate you Ash, forjoining us and sharing your
brilliance.
and thanks everybody and have anawesome rest of your day.

Ash Stearn (42:11):
Awesome everyone.
Thank you for having me.
Bye.

Absolutely.
Bye-bye.

All Episodes

185 | Boost Sales With an AI Agent That Researches, Reports & Preps Sales Calls with Ashleigh Stearn

Episode Transcript

Popular Podcasts

Stuff You Should Know

Dateline NBC

On Purpose with Jay Shetty

.css-15opob5{left:0;position:absolute;top:0.8rem;} All Episodes

.css-14f5ked{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:2;overflow:hidden;}185 | Boost Sales With an AI Agent That Researches, Reports & Preps Sales Calls with Ashleigh Stearn