430- Stop Buying AI Projects Backwards w/Michael Baillargeon

Phil Howard: Welcome everyone back to You've Been Heard. so Mike, here's the

thing. I After talking with IT leaders, we realized Nothing gets done in any

company without it touching it. But what we also realized is that whether people

want to admit it or not, nothing gets done without technology touching it. And

most people that are in a leadership role in technology right now, at least in

the mid-market space and enterprise space, didn't go to school for it, or they

might have gotten an MBA, they might have done, something related to it. But no

one grew up saying, I'm going to be a CTO or a CIO. We had bag phones back then.

We still had phones sitting on the wall with cords attached to it. the internet

was brand new. People were still very skeptical of it. And then here we are

today with AI, which is what we're supposed to talk about today. But even till

today, all the work that we've done as technology leaders, we get a seat at the

executive round table. We're not always heard that well. So the mission and

point of this, podcast is that, you know, it's great to have a seat at the

executive round table. How much nicer would it be to be heard? And a lot of C

level executives don't necessarily listen to their IT leadership. They have to

go to places like Gartner, and they have to go to places like insight, and they

have to pay them one hundred and twenty thousand dollars a year. And we like to

say, well, we'll give you Gartner level insight without the six figure price

tag, and we'll give you access to all kinds of people that will sanity check

decisions and everything and make you feel comfortable about what your CTO and

CIO already know. But this isn't about the, it versus the business or anything

like this. It's really about bridging the gap and making sure that we translate

and I guess, get to a point where everyone understands the words that are coming

out of the CTO and CIO's mouth. they may not know what hypervisor is or what to

do with that or anything, but they kind of understand at least enough to say, I

don't need to check with Gartner, who's going to assign an entry level college

student to your account. Does this sound familiar to you at all?

Michael Baillargeon: Yeah. what's interesting is you hit on something. So, I was

an English major in college. I wanted to bring Shakespeare to the masses. And, I

ended up moving from New England out to Phoenix, Arizona. It had the thick

Boston accent at the time. It's kind of dwindled down a little bit, even though

people still say I have it. And, I ended up at a call center and I was a call

center agent. and I was an agent for, I don't know, maybe a year. And I realized

it's a hard job. Two. it's a challenging job. So, an opening came up in the IT

department to be like the junior phone system administrator. Now, I could spell

PBX. That was about all I could do. but I interviewed pretty well. And I tell a

good story. Right. That English teacher. You tell stories, remember what we tell

all our salespeople and what you tell all your IT professionals is, hey, numbers

tell, but stories sell. That's how to really get your point across when you're

at that leadership table. Give them the story. Just don't give them the numbers.

So, I ended up being the phone administrator on this million dollar. AT&T wasn't

even Lucent then. It wasn't even a buyer. It was AT&T, PBX and,

Phil Howard: thirty star codes,

Michael Baillargeon: and Unix boxes. And, I learned a lot.

Phil Howard: What's really wild about this story? Yeah. Is that I was an English

major, and my first technology job started off like this. I was in a call

center. and I remember the IT guys in the bullpen. We know, they were like, oh

yeah, we audited your call today. You shouldn't say. I'll never forget this

coaching conversation I got from my boss. You shouldn't say, Oh my bad. I was in

college and like, that's not professional. You shouldn't say my bad. but I got

moved to quest. Premier. That was the best. Yeah, it was a LAN party because

there was no security back then or anything like that. It was a daily LAN party.

So we only got five calls a day because it was like you had to have one hundred

cell phones or more account. Yeah. So we all had PCs networked together. We had

an OC twelve which was like unheard of. It was like unheard of. Yeah. Where is

this bandwidth coming from? I hadn't even had DSL yet at home. okay, so you

ended up in telecom. I ended up in telecom. And of course, the natural

transition is to all things technology from telecom.

Michael Baillargeon: Yeah. So, that company was kind of a BPO and then we lost

our two or three big customers. And then I went to the on prem PBX world. I

worked at Intertel and ShoreTel and Billy Bob Tel right. So very natural

progression. And then maybe in twenty seventeen, twenty eighteen, I had this aha

moment that the cloud is where. You mean no more hardware, no more power

supplies, no more patches, no, like well done in the cloud. I'm like, tell me

more. And I've just been cloud first for the last, almost ten years now. And

it's fantastic. But, I remember the big kind of part of that was, back in twenty

eighteen, I call it the UC earthquake. It's when Cisco bought Broadsoft. Think

of it. Right. Cisco. The biggest hardware manufacturer on the planet. And

Broadsoft the biggest soft switch on the planet. That signaled the decline of

the on prem PBX that day. You can pretty much put a stake in the ground because

it validated what you do every day and talk about cloud. It just validated that

and come full circle twenty eighteen. And the lockdown that was the biggest beta

of work at home ever, right? and that seemed to work out okay. We coming off

that this and now people are like, Mitel, they're discontinuing everything. If I

was discontinuing everything Cisco. They want to, but it's a little longer haul.

But I mean there's this graveyard out there of on prem pbxs that are no longer

valid. It's crazy how many clients I talk to who still say I have a Panasonic, I

have a Toshiba, I have a ShoreTel have an Intel and.

Phil Howard: Toshiba is great. Toshiba was the best key system. Yeah, so how do

we transition this all to AI? I have Greg on the call because every morning I

talk with him, he's like, oh, that's so like fifteen minutes ago. here was the

question. Right. If your CEO came to you tomorrow and said, show me an AI win in

ninety days. Where do you start?

Michael Baillargeon: To a few places. First of all, there's no skew for AI.

There's AI is not on a box on a shelf somewhere. And without an AI strategy, AI

as a two letter word. That's why MIT said ninety five percent of all projects

fail. and I think when CEOs, CTOs, VP of it, VP of sales read that, they might

say, why should we even try? Right. Why should we even try? If m I t say ninety

five percent, how can we possibly b b the five percent? Well, this and we'll get

to this, I think during a conversation about that readiness. but there's two

areas I think you can get a quick win. one is within your customer service,

customer care, customer experience, whatever you call your call center team,

your contact center team now. I know we're going to hear, but Mike, I'm not

Delta Airlines. We don't have a call center. Let's define a call center like

this. Phil. It's a group of hard working people that either gain new customers

or they retain customers, and it all adds up to revenue. Okay. Gaining retaining

revenue. That's what a call center does. Now they vary by size, scale, and some

complexity. But at the end of the day, that's my job. I want to make this

customer happy, or I want to make this customer buy something in that customer

experience world. A quick win would look like this. Wouldn't it be great? You

call into the call center and it says this call may be recorded or transcribed

for quality purposes. Phil, did you? Only three percent of those calls get

recorded, transcribed, listened to by a supervisor and then feedback given only

three percent.

Phil Howard: It's illegal. Asterisks check box that they have to check is

basically what people are doing.

Michael Baillargeon: Yeah, one hundred percent.

Phil Howard: Like we want to record them. Somebody said at some point, the CEO,

I want to record everything in case we get sued or in case someone needs to

return something or lies and we have to pay out money. That's why they

ultimately, I'm assuming this is the reason why most people want to record

calls. Yeah. If you're somebody that's probably different, they're doing a lot

more. Q and a they're they're working on Csat scores.

Michael Baillargeon: You want to talk about that lawsuit. But, there's a very

technical term for it, and I might need Greg to help explain this. The term is c

y a. that's why they record those calls, right? To cover themselves. Now, what

if we could take that c y a and turn it into a profit center? What if I could

take one hundred percent of those calls, have AI listen to those calls and score

all my agents. based upon a rubric as a corporation, as a company, I make sure

they have validate their account number? Did they validate their date of birth?

did you upsell them all that? in that way I can make my people better because

Phil, you know as well as I do remote work is part of just about every company

on the planet these days. so, in the old days, I had a supervisor who would walk

past me and kind of listen, is my cussing out that customer? No, he sounds like

he's doing pretty good and he walks on. Right. But now I can't visit everybody's

house to do that. So how do I keep that quality level so high when everyone's

working remotely? we call that AI quality assurance or quality management or

quality metrics score one hundred percent of the calls. And then as a human in

the loop, I can still go into each of those. So if your best guy, you get to

ninety and ninety five. And then I see Mike's in the thirties. Why? Why is.

Phil Howard: Mike, can we take this a step further? Is there a way to do this

with. I mean, why wouldn't you do this with your salespeople? Why wouldn't you

do this with all of your inbound outbound calls? It would be great if we could

do it with our Zoom calls. So every Zoom meeting that my people run, like, could

we score that? That'd be cool.

Michael Baillargeon: That would be very cool. And then Phil, here's the other

thing. If you are an organization that has recalls, so you're in manufacturing,

whether it be physical devices or food manufacturing, give you a perfect

example. When I was at that BPO thirty plus years ago, we had Nestle was one of

our big customers and it took us about three weeks of anecdotal knowledge to see

that a certain dog food had an issue.

Michael Baillargeon: Now I can get it done in about three minutes. With AI, it

can surface that you've had ten calls in the last ten minutes about sick dog. So

it's just not surfacing about if I'm a poor agent and Phil's an amazing agent,

but more importantly, that voice of the customer, voice of the patient, voice of

the member, whether they actually talking about. Another perfect example. I

worked with this company. They make, cowboy boots, right? And they only made

square toe ones. Now, I'm not a cowboy boot person, Living up here in New

England. Not a lot of cowboy boots up here. But anyways, they only did the

pointy toe boots. they put on this AI q a q m, and then they learned that thirty

percent of the calls started with I'm on your website and I'm looking for square

toe boots, but I don't see them. Well, guess what? They started making square

toe boots. And now square toe boots are their number one seller. They would not

have known that if AI didn't surface that information. That's the power of AI.

That's a quick win, right? Of let's listen to the data we already have. We

already have all these call recordings. Let's listen to that data and figure

something out. That to me is your quickest win.

Phil Howard: What about the people that just don't necessarily have a big

contact center like, well, logistics, if we use logistics for an example, we

have a ton of inbound calls, people booking loads. we got a lot of outbound

calls doing temp checks and like checking where the loads at, quality assurance

from that standpoint. So there's one example there. health care. We've got

patience and I'm assuming scheduling is kind of a big one. this is all kind of

like telecom examples. Do we have any other examples that are just non-telecom

maybe?

Michael Baillargeon: Yeah, I think there's one question I've been asking IT

leaders quite a bit is, do you think your, coworkers may be talking to a public

GPT pick your poison, right?

Michael Baillargeon: Cod grok chat.

Phil Howard: We know they are.

Michael Baillargeon: We know who they are, the usual suspects, right. And maybe

putting up PII or Phi information, for example. we worked with a financial

advisor company, right? and what they would do is they would take my portfolio

with my name and everything. Copy everything and put it up to ChatGPT. how

should we improve this portfolio now?

Phil Howard: Wow. Genius.

Michael Baillargeon: Yeah. My portfolio went crazy. So what people don't always

understand is if you go to a public GPT you're writing that on the internet. So

here's the other thing. We work with a couple of suppliers. So then the

customer's like, well I'll just buy my own private version of ChatGPT. I'm like,

okay, you can do that. How about if we could do this for you? How about if we

had what we call an LLM aggregator, someone who already bought a subscription to

the top five, seven, ten LLMs out there. And now when I type in fifty two, and

if I'm the CFO and I go to ChatGPT and said, hey, we're the Smith company,

here's our financials. Where are we at risk? What happens is that, LLM

aggregator will redact any of the PII or company information. We'll put up

general numbers up there. And then when the string comes back it'll repopulate

with Smith Company. So now. Yeah. Go ahead.

Phil Howard: So this is good. It's like most people don't know what an

aggregator is, but try to imagine grok ChatGPT, Claude list a few more, Greg.

All in one dashboard for an IT manager to see all of his active users, what

they're actually chatting, what they're doing, and then give that access to all

those elements maybe to a private drive or what do we want to call it? Single

source of knowledge in dip from that and know that the LMS you're using are

secure, and I know like Debs AI, for example, they're hosting, the LMS on their

private, cloud or whatever. and in the terms and conditions, we're not sharing

your information versus an end user, just, downloading it, and then not only

that, then the cost can be spread, the cost per users, maybe thirty dollars,

forty dollars versus thirty dollars for each individual thing. And then you can

share the tokens across all users because not all users. Now here's the issue

with that model. and I'm going to let Greg ask the issues because I was like,

why don't we do the aggregator? we're paying for this, we're paying for that.

We're paying for all these other ones. He's like, well, I like the interface of

Claude, and I can use the Claude desktop version, which I can't really

necessarily do with an aggregator because I've got to use their interface. So I

mean, I can see for maybe like DevOps guys, if they're going to need maybe each

individual one, but for the general user inside a company, even building agents,

you can use, aggregators for. But Greg, this is where you ask the hard

questions.

Greg Le Dall: Yeah, thank you Phil, it's interesting because the way that we

work together, we really witness everything. And me, I really move from the, as

Mike was saying, the public chat GPT to, now really using AI as a tool around

our use cases. And that's where it becomes very interesting when you can start

to see, LLMs, not just as interface. When you ask questions, back and forth, but

it's really a tool that you can integrate on all your different workflows.

That's what we've been trying to do, you know, with a genetic AI and everything.

Now the main difference and this is where, that would be interesting to have

your point of view, Mike, is, we are a small teams, so it means that we can

really control, you know, what we're sending, we know the data that we share,

the one that we don't. Now, if you expand that on the size of an organization,

this is where I really want the, you know, how can enterprise AI adapt to how

fast it move, how fast it grows? Because we know that usually, it takes a lot of

time, to move to the decision maker that approved maybe the use of this model.

And by the time we use this model, as Phil was saying, you say, now this one is

already, like, there's already two models, you know, that came after and just

blew it up. So. How do you manage that?

Michael Baillargeon: Yeah. So, in the past and Phil mentioned we have a DevOps

team in the past, you would tell your DevOps team, go and build this application

for us. You have three months right now. Hey, go build an LLM of billions of

units of data. It's going to take a long time, right? To create your own LLM,

import all your information. I think using an LLM aggregator does two things. It

insulates you from the average user like me, who, I don't know, whatever I put

up there is now in the public domain. I have no idea.

Phil Howard: Like for the user that's using AI, that's better than a search

engine. They're using it to like create reports or use it for XL? Or they're

using it to build PowerPoint presentations, etc.. I'm assuming we're talking

about that user.

Michael Baillargeon: Yeah. Yeah, that exactly. That user or someone who's like,

hey, help me with this portfolio or help me with this accounting sheet. I don't

understand what's going on or even help me with a cold call script. Right. Those

users. and Greg, correct me if I'm wrong. Certain models are better at certain

verticals than others. like I'm a bit of a fan of gamma to create a PowerPoint

presentation. I think it does a really good job for that. versus a couple of the

other tools is that, is that fair to say, Greg, that some models are better

either vertical or depending on what you want to do?

Greg Le Dall: Yeah, yeah, definitely.

Phil Howard: Share our tech stack, share our tech stack. What do we do without,

without yeah.

Greg Le Dall: Usually like now, for example, me, I'm a big user of, of GPT five

point five. so, if, anyone has been following a bit of the drama that, was

around entropic because entropic, I don't know, it's a really interesting use

case. I could go on for forever on this, but basically they were like the top

leader. Like you remember Phil, I would say, no, no ChatGPT. Forget about

ChatGPT. You know, I cancel my subscription six months ago and now I'm all about

cloud. Why? Because cloud, is just like the ultimate. Now what happened is that

entropic, they started to restrict the, subscription model. So for example, we

used to have the two hundred dollar plan, which is quite a lot of, token

available. And when you combine that with your agent, it's fantastic because now

you can get your agent to go on the web, do some research and build some stuff.

So it's basically, like a one tool, one fit it all. Then after entropic decide,

they decided to restrict the use of the subscription to their ecosystem. So that

means that, for example, me, I built everything around cloud with the

subscription. And now we have a system in place. And if I compare, the price I

would pay with the API, that will go from two hundred dollars a month to, I

don't know, that could be two thousand even more, I think. So that's a good

example of, building something, relying on a provider and then that provider

change a strategy and then that's it. You're producing everything.

Phil Howard: We're producing mass amounts of content based on what we had

recorded off the podcast, I mean, infographics. And like, you would have to tell

me how many steps it was taken to go where. But I remember literally overnight

done like back to the drawing board. Now, if that happened to a manufacturing

company or that happened to someone that had invest, it would be we're a

podcast, right? Like, yeah, it was painful. It was problematic. It was like,

this is why we can't go all in on AI. This is why it's going to be point

solutions, Because we have that argument all the time with our friends. like AI

is going to put DevOps teams out of business. It's gone, it's taking over. And

then overnight, you're kind of like, oh, back to the guy as the DevOps guys that

said, see, I told you, hahaha. I do want to give the audience something of like

value here, so number one, where are you at on your actual AI journey? where do

you actually start? where are all the different points that people are at on

their AI journey? If you need to share a slide or something, go ahead and do

that. But, if you were to, for all the people that you talk with on a daily

basis, CTO CIOs that say, hey, we need help with our AI strategy, I really don't

have time for it. I know that I've got a lot of pressure to be under AI. I'm

completely overwhelmed with how fast things change all the time. Every single

anyone that does not admit that they're not overwhelmed by AI. Let's help

people. Where do we get started?

Michael Baillargeon: Yeah. Okay. So in order to have a successful AI project,

you need a successful AI practice. A successful AI practice starts with a

successful AI strategy team. These should be business leaders across multiple,

divisions departments. This team comes up not with what the AI going to do.

Start the other end. They're going to start a business outcomes. I need to know

my strong performers. I need to know when Salesforce or ServiceNow creates a X

level ticket that kicks off X, Y, and Z practice. I need a business outcome of,

if my revenue is dropping X percent across a day, why is that? What are some of

the key triggers? we're on the Zoom platform. So Zoom has a product, AI insight.

I think, where as a leader of the company, why was my service level lower last

Friday? why am I doing this? So it surfaces all this information. We have to

start there. Then we work backwards to find the right AI tools and models to

make that work.

Phil Howard: So how do we know as IT leaders how to surface up whatever outcomes

is that.

Michael Baillargeon: That's a team decision, right? the IT leader is being told,

hey, we need AI in our infrastructure. What does that look like? Right. the

other problem you're going to have is good structured data leads to good

structured AI. Bad unstructured data leads to a bad AI experience. A perfect use

case. So I'm working with this large client, eight hundred contact center agents

on a very modern platform. Yet that modern platform didn't really give some of

the AI experiences that they really wanted. So they brought us in to kind of

talk, the art of the possible of what can be done. And then my very first call

with them, I'm like, okay, what is your single source of truth? And they're

like, oh, we only have like two or three. I'm like, okay, what are they ended up

filled. They had fifteen sources of truth. I said, okay.

Phil Howard: Which were in a very vague, non-descriptive make believe fashion.

What were these sources of truth like? Yeah.

Michael Baillargeon: So you had, Salesforce, you had a CRM, you had an ERP

system, you had two or three sales platform ordering platforms. You had, server

underneath Mary's desks that have been running forever. And every time they go

by.

Phil Howard: This is what they were saying. Their sources of truth were, well,

we got Salesforce.

Michael Baillargeon: To the truth, right? every stakeholder said, well, we use

this and we use that. And the it was like, I had no idea where supporting all

this stuff.

Michael Baillargeon: Right. So in order to get that good AI experience, you need

a collective unify. I want to say single unified source of truth. That's where

knowledge management companies really come in and to help kind of aggregate all

that data and put it all in one spot. So either I can use it as a internal chat.

GPT

Phil Howard: Talk to me for a second. Some people might not know what that is.

this knowledge management type of content. In other words, data cleanup.

Michael Baillargeon: Data cleanup. Yep. Yeah. We used to call it a knowledge

base, but now, marketing gets involved. Phil. So we have to up our terms. Now

it's a knowledge management platform. So what they do is they clean up all your

data that's out there. There's a term called. Redundant, outdated. I don't

remember what the T was.

Michael Baillargeon: I can't remember what it was. So what these platforms do is

look at all the sources of truth, right? From simple stuff like on OneDrive and

In SharePoint, Salesforce, your ERP on the server underneath Mary's desk. It

looks for that rot. What's redundant? What's outdated? Right. The worst thing

you want to do is a agent kind of goes into an FAQ and says, oh yeah, our return

policy is ninety days where that's an outdated policy, right? You got to get rid

of all that. So these knowledge management systems come in and kind of clean

that up. So it can be used. Yes. Again, I'm a bit of a call center guy. It can

be used in the call center in this way. A call comes in. I'm a remote agent.

Question comes in and the question I wasn't trained on. I look to the right.

There's the dog. I look to the left. There's the cat. How am I going to answer

this? I could put the person on hold for ten minutes. I could go into teams

chat. Right? In the old days, when I was in the office, I had the flag I could

raise. Now with AI, AI will automatically surface the right FAQ or the right

document for the agent based upon what it heard the customer say. And you really

need that single database for that because it reaches out for that latest one.

Now once you have.

Phil Howard: The way, the T stands for trivial.

Michael Baillargeon: Trivial. Thank you, thank you. Trivial. So you have to get

that data in order. And now talking non telecom, that data is also really

important for those workflows. it's nothing more than if this happens then do

that. Now it reasons what those then should be. Greg you dropped agentic on here

right. So the biggest difference between conversational and simple generative.

That was if then right. It was if they say this, then do that. If they say this,

then do that. It was like one hundred IFS and one hundred thens. Now with

Agentic, it reasons it. There's one if if this comes in, here's all the thens

that are possible, right. So it should do it a little bit faster. But from an

internal point of view, internal it that data, right. The readiness of that data

will help with all those internal workflows. If this thing kick this off.

Phil Howard: Now comes first, what are we trying to achieve then? Okay, well, do

we have the data that we can? That's a everything but redundant, obsolete and

trivial third thing.

Michael Baillargeon: You have to have the team, with the right stakeholders in

place. And then from an IT perspective, you have to have people ready for this,

right? You have to have a bit of a voice prompting skill set, right? like I

said, conversational with all writing a hundreds of if then statements Agentic

is you have to have the right prompt so there is some training required. Now,

what's nice is when you work with Phil and I and we have a bunch of suppliers,

they do all the training and whatnot. But if your team's not taking, basic voice

prompting classes, you're going to be behind.

Greg Le Dall: Yeah. It's really interesting what you're saying, because, when

Phil, when I was always saying like, yeah, prompt engineering, prompt

engineering, and then there's a term, that came around like a few months ago.

And I think that's really the right term is context engineering. And you really

learn by doing, when you build an agent, because it's really like training an

employee. And I experienced it. I just realized that my data was a mess. I was

always upset against the AI agent. Stop hallucinating. I didn't say to do it

like that. And going into the flow. So AI is always going to try to do its

thing, and figure out by itself, how to come up with something. So that's when

you realize, okay, now I need to have the proper explanation of how it works.

You know, you do this and this and this and that. Repeat after me. Good. You got

it. Okay, let's keep going. Let's keep going. That's really the context. And

Mike, I think that's what you're referring to. You know, clean data. Like

there's no room for, guessing.

Michael Baillargeon: Yeah, thirty years ago, we used to call it garbage in.

Garbage out. Right? So you do, COBOL, Fortran programming. Number four. Quattro

Phil everybody's favorite governance, compliance and security. Right.

Phil Howard: So don't kill it all. We should probably just start with that.

Michael Baillargeon: Yeah, yeah, yeah. The answer is no. What's the question? So

how do you get to the. Yes. Right. I mean, that sales one hundred and one. I

know we're not really sales people on this call, but, everyone sells in their

own way. But how do you get to that? Yes. So it's incredibly important to

understand if you're in healthcare, what the HIPAA rules are. If you're

government cmmc what those rules are and then where to put. So the term you hear

all the time, right, Phil, is guardrails. Where are those guardrails? most of

the vendors who we work with have those guardrails in place where if we go out

to our LLM, it's going to be to your point earlier, it's our private. LLM it's

not going to be out on the exposed, internet website, whatever it might be. So I

am not a cybersecurity expert. Phil, you have experts on your team who are, I

tend to think that infosec is one of the first conversations you have to have

whenever you have a new project. Now, it doesn't have to include the VP of sales

or the CRO. It can be a separate team. But you got to clear infosec and

governance compliance, right, and regulations upfront before the project even

gets halfway done. And then you realize you spent all this time and energy and

money only to hit a roadblock you just can't get over.

Phil Howard: So if he's on that team to begin with, we gotta have the

stakeholder team. He's just got to be on the stakeholder team like, yeah, one

hundred percent.

Michael Baillargeon: Yeah, you call him doctor? No. Or her doctor? No. Because.

How do you get that to be, doctor? Yes. You have to get them on board with

everything.

Phil Howard: do we have a fifth piece?

Michael Baillargeon: Well, I mean money, I guess money comes into it, right? we

create these business outcomes. Are we going to generate an ROI? I mean, that's

what m I t talked about. it wasn't that they failed. they just didn't generate

the ROI they expected. so, your next Google is to Google the Gartner hype cycle

for artificial intelligence. I tell you what, whoever puts some of these names

together, kudos to you.

Phil Howard: The hype cycle because they are The hype. It's like, how can we

create something before someone calls us it?

Michael Baillargeon: Yeah. So what happens is they get this like sine wave,

right? And in the middle, it's the peak of inflated expectations. Kind of like

my bank account, right. And then what happens is people have all this idea, oh,

AI is going to save us money. It's going to reduce headcount. It's going to give

us a competitive advantage. It's going to be easy. And all of a sudden they drop

way down to the trough of disillusionment. I mean it sounds like a like the

Candyland game right. That's what it sounds like.

Phil Howard: And we're kind of in that trough right now.

Michael Baillargeon: In that trough. Because what happens is CEO says, go get me

some AI. Right. And it's hyped up and then we can't figure it out. And we're

like, we're part of the ninety five percent. But if you put a strategy around

it, I think these five points that Phil was talking about are critical, right?

We want to get the stakeholders involved. We want to make sure we start with the

business outcome. Listen, sales one hundred one. You talk to a salesperson,

they're going to start with a product and they're going to find a solution. The

product is a hammer into a hammer. Everything looks like a nail, right? So what

you want to do is don't start with the vendor. right? Start with the internal

task force strategy team, Tiger team, whatever you want to call it. Have your

business objectives there. Look at your data. Can you aggregate the data? Can

you make it safe? Can you make it one? Is there rot in your data infosec? I got

the data. what guardrails can we put on that data. And then the last one is if

we go through this project and we hit all our markers, it's safe. Secure. It's,

going to generate the business outcome. What's the ROI on that? Microsoft put

out something crazy like people who added AI saw a three hundred percent ROI.

Okay, show me how they got there. Talk about that path because you need that end

to end strategy. Even before you talk to a vendor who can do something for you.

Once you have that in place, then you're ready to look at product set and then

at that point, Phil, as the big question is build versus buy. As I mentioned,

you buy the LLM aggregator as you're making your own LLM to kind of bridge that

gap and make everybody safe. And that's just one idea. That one strategy that we

put together for our clients with those five points.

Phil Howard: Let me make sure I get this straight again. Stakeholders outcome

data, guardrails, ROI sounds good.

Michael Baillargeon: Money comes last. Like you figure out the ROI at the end.

If you do everything right, the ROI should happen.

Phil Howard: Well, I think some of the biggest things are is. the outcome, which

would be. I'm trying to think of other ways to say that it's like, what

challenge do we want to solve with AI?

Michael Baillargeon: Maybe once we once we put AI in, this will happen right.

Phil Howard: Yeah. It's also how do you word that challenge like. So if I think

of some of the challenges that we have, it might be, growth of the podcast, but

that could be what that could be social media, it could be growth of the podcast

on YouTube or whatever. We're primarily a technology leadership focused,

LinkedIn only audio podcast, right? So it's kind of like, is AI the solution to

that? Or is it just by LinkedIn advertising and advertise the podcast and how

does AI fit into that? And then obviously I can do copywriting and all kinds of

other things. So it's not a simple question to answer.

Michael Baillargeon: what about AI scoring of the podcast or the reviews of the

podcast? See what topics really kind of, generated more attention versus you and

me talking about life in New England and guy in New York is like, yeah, I'm

done. Click. Right. so that might be, one area of.

Phil Howard: Let's go, let's use me as the example. I would love some more

examples. So that's a great example. Greg, how have we done that? Because we

have used AI to obviously, look up all of the trending topics. Let's fire away.

let's try to stump Mike.

Greg Le Dall: the research, just confirmed what Mike was saying because, the

first thing was, yeah, I like this one. Most AI roadmaps are built backwards

organizations by tools first. Then they search for use cases.

Michael Baillargeon: You buy a hammer, you look for a nail, Right. Start at the

outcome. Start at want to happen and work backwards.

Phil Howard: And a lot of it guys are going to agree with that. we don't use it

as the it's not the hammer again. Okay. I guess summary theme of this entire

thing. You're an IT leader. Be an IT leader. You first have to have control and

the ear of your stakeholders. Otherwise they're running rogue and wild on you.

So first have your stakeholders AI, round table steering committee, whatever you

want to call it inside your organization, maybe make it unique. That would be

awesome. and maybe do your initial, lunch and learn burrito, specialty day,

whatever it is. and brainstorm what your number one challenges are, outcome

solutions that you want AI to help solve. So now you've got your stakeholders,

you've got your solutions or your challenges that need to be solved. And then we

can help you take care of the rest. We can help you clean up the data. We can

help you put in the security guardrails, and we can help you find and implement

the right products from a vendor neutral perspective without the six figure

Gartner magical price tag. from a vendor neutral standpoint. Sound about right,

Mike?

Michael Baillargeon: Sounds great. it.

Phil Howard: It has been a pleasure having you on the show. Any final words of

wisdom or anything?

Michael Baillargeon: Do not let AI drag you down. It's a journey. It's a

marathon and not a sprint.

Phil Howard: I have another question and I forgot to ask it. Eighteen months

from now? What is your prediction and anything, technology. I mean, actually

could just be anything.

Michael Baillargeon: I think the number of self-driving cars will increase. I

don't know about you, but when I was sixteen, the first thing I want to do jump

in the car and drive it. Now, all these self-driving cars. I do think we'll see

EVs kind of making a bigger comeback, especially with gas pricing. And I do see

more autonomous vehicles on the road.

Phil Howard: That's a great prediction. So okay, Mikey B You've Been Heard.

Michael Baillargeon: Thank you.