AI Adoption in Power Plants and Beyond – Trends, Challenges, and ROI Insights

ROI modeling is much deeper than that, and that's stuff you have to actually— we

have workbooks to help walk you through the steps and, you know, what

are we doing today? You know, what's the blended hourly rate

for the person this is going to affect? And so we really kind of try

and make this as simple as possible for our users. But if I was to

provide you one kind of anecdotal story about a company's

journey, I have to produce executive reports around ROI and

Copilot. We have about 160 licensed users out of a company of

400. Our original monthly ROI

from our investment in Copilot was running somewhere in the neighborhood

of $3,000 a month when we first started using it, which

barely covered the cost of Copilot. 2 years later, we're now

reaching close to $80,000 a month. Wow.

So we're seeing close to $1 million of assisted value

before we even start to go in and look at the individual use cases

and start to evaluate those.

Hello and welcome back to Data Driven, the podcast where we explore the emergent

industry that is artificial intelligence, data science, and of course, all of

it is underpinned by data engineering. And once again,

I am joined by my favoritest data engineer in the world, 3 times in a

row actually, Uh, as we release these, uh, Andy Leonard. How's it going,

Andy? It's going well, Frank. How are you? I'm doing all

right. I'm doing all right. It's been, um, been a

thunderstormy kind of day, and, um, which is good because we put down a bunch

of sod. We ripped up, did a lot of landscaping here at, uh, the chateau.

And, um, so we're getting ready for fall

and kind of cleaning up the, um, the front property.

But it's a good time of year for that. It really is. It's not

too hot. Most of the summer's behind us, but

I'm very excited to talk to today's guest.

Today's guest is Jim Spignardo. He's Director of Cloud Strategy and

AI Enablement at ProArc, a global IT

services and consulting firm. Jim's been in the IT

field for 25 years and holds a number of Microsoft

certifications. As we said in the virtual green

room, it's always good to have somebody who's back in the, you know, who's in

the Microsoft ecosystem. And remember, kids, I didn't leave the Microsoft

ecosystem because I hated it. It's just because I got pulled into

other directions. Although my feelings on Windows 11 are well

known, and in the interest of positivity, I will just stop

talking about Windows 11 right there. How's it going, Jim? Good. How are you

guys doing? I'm doing all right. It's good to see you. And I see

that, um, are you up in, up, uh, in upstate New York still?

I am. I'm, I'm right outside the Syracuse area. I've been— okay, cool.

I lived here, yeah, for about, about the last almost 30 years now.

Okay, cool. I'm from downstate, again, because you guys would call it—

whereabouts? I actually grew up downstate. Oh, okay. So I

spent my formative years in Staten Island and Queens. Okay, I'm not

all the way downstate. I grew up in the Hudson Valley. Newburgh, New York.

It's really nice up there. For those who are not aware, the

southernmost point of New York State is Staten Island. Yes,

it is almost New Jersey. If you look at it on a map, you really

wonder like, why is it not in New Jersey? I did spend the other half

of my formative years in New Jersey, so perhaps that was

foreshadowing, who knows.

What is it that you do? I see you've been a network engineer.

Your company is short for Pro Architecture.

How do you— what is it you do now? Sure. Yeah. So

for probably the last almost 2 years now,

I have been tasked with helping our own

organization internally adopt and

enable AI technologies for the betterment of our organization.

But at the same time, I am responsible for developing solutions

and products to also assist our customers on that journey.

primarily focused around M365 Copilot, but not entirely.

Our organization has a very large digital engineering team,

probably actually the largest portion of our organization. And so,

you know, we've been doing AI before AI was a thing and, you know, we're

really into machine learning and various technologies.

So we also work pretty heavily in with Azure AI

Foundry as well. And, you know, and also the big 3 players that are out

there. Yeah. What

exactly is Azure AI Foundry? Because that's something that I've never really—

I've done a lot of private AI type stuff, sovereign AI

stuff. Yeah. So if you think of Copilot as kind of

being the no-code to low-code solution, when you think about

Copilot Studio, as far as the development of

agentic solutions, Azure AI Foundry is really kind of the

pro-code option there. And so you're building

solutions, typically web app services on

top of Azure, but then you're integrating Azure AI Foundry,

which has a whole host of AI services

from things that Microsoft does themselves, whether it's

text OCR recognition, voice-to-speech,

but also the opening up the ability to attach

thousands of different AI models. So really, whatever solution you

want to build, Microsoft gives you the opportunity to

choose whatever model works best for you, whether it's

something from DeepSeek, if that's the direction you want to go,

or Mistral or ChatGPT or Anthropic or

LLaMA or, you know, or even some of these more niche open

source models. They're pretty much available to Azure

AI Foundry to build solutions on top of. That's

very interesting. So being able to cherry-pick

like the best of this model and put it alongside the best of this

model. So like Frank, I had heard of Azure AI

Foundry, but I've not yet tried to do anything with it. Now

I'm intrigued, Jim. You've got me thinking that it's,

you know, it does what it says. It actually is a foundry.

Yeah, it really is. And it integrates really well with

Microsoft's data platforms as well.

So data lake and now with—

losing my mind— Fabric, right? Fabric really

brings all of the various pieces together. So with Fabric, you're

getting the business intelligence component with Power BI,

you're getting the Azure AI Foundry components, you're also

getting the ability to stand up data

platforms like data lake or data house, data There's

50 different names to come up with now. Sure. That's all SQL

as well. I know you said you're a SQL guy. All of that's under

one giant roof of tooling that you can apply.

When it comes to building enterprise AI solutions,

Fabric is intended to be that enterprise data platform where

you can ingest all that data using a medallion architecture

and then attach the tooling on the back end, whether it's Power BI

reporting or Copilot or some other AI solution.

So I'm digging that. I too am a consultant and

love consulting. I joke with people who

ask me about what is consulting like, that it's

actually an illusion that you're working for yourself,

but it's a, but it's a very powerful illusion.

So curious,

I'll throw a question out about this. Can you give us a case

study or real or generic? Sure.

About doing, applying what you just said, Fabric plus AI Foundry.

I'd love to learn more. Yeah, absolutely. Yeah, same. I'm very

curious about Fabric. I was a Microsoft employee when

Synapse was the thing and then

Yeah, so I'm very curious. Yeah, so

one solution that we're actually bringing to market within the

next few months or so, we have a ton of power plant customers,

power generation customers throughout the United States. We do managed services for them,

so we both manage their IT infrastructure from a

support level. We also do some level of management on

their operational technology side, so the technology that actually runs the plant.

And we also provide some security services using tools like

Defender for IoT. So that connection brought us

to one of our customers who had a potential

use case where they would like to see us implement a data

platform with some intel— AI intelligence behind it.

And what we're doing right now is taking the

data that comes off of the sensors within a power

plant, whether it's a pump or valve or there's tons of different

pieces of equipment. I'm not the most knowledgeable on all the inner workings

of them, but what we're doing is we're taking all that data, ingesting

into our data platform, and then using AI to

provide business intelligence back to the customer to

show them anomalies within those

pieces of equipment to see if they're operating within the

normal bounds. So if their RPMs are where they need

to be based on the age of that device and the

model of that device based on what the manufacturer spec says,

and then we can start to detect anomalies so that the plant

operators can make adjustments or they can predict when these

devices start to fail and actually make some

interesting decisions. Even small

configuration changes within these devices has the potential of

saving a small plant tens of thousands of dollars a

month in how much energy they're able to produce and how efficiently they can

produce energy. So I get it. The downtime alone

probably pays for what they invest in your company and

implementing AI that way. Just a little background. Back in the

'90s, when the years began with a 1, I was a

manufacturing integrator and ran a small electrical

contracting firm. And mostly I refurbished

electrical control panels. I speak PLC, programmable logic

controller, and manufacturing execution systems,

MESs, and HMIs, human-machine interfaces. So I'm

tracking along. I even rebuilt a panel once for Duke Energy. I'm in

Virginia, and they're here and across the border in

North Carolina. So really identifying with,

with what you're putting down today. And I'm sure you're

familiar with all of that. Before we had what we call IoT today, we

did have sensors out there collecting data, sending them across

19.2K nets, mostly

proprietary stuff that when we integrated with PCs, and

probably still, it's still that way. Yeah, for the most part. Yeah,

yeah, that's probably more common than anything. And we're very

lucky because we're kind of at the forefront of a modernization

revolution within these industries that for the most part have

spent the better part of a decade or more

using the same thing over and over. You go into some of these plants and

the interface looks like straight out of Windows 95. It provides

them with intelligence, putting data on a screen

and giving them feedback. But a lot of times it's

reactive. They're seeing what's happening right now. They're not

able to look at the trends and the the

analysis and do predictive analytics. And also

now also layer in artificial intelligence where we

can feed it the documentation from the manufacturer and make

some predictions or even assumptions about

the operation of those devices. I love that. And

I'll just throw one more thing in and then I'll shut up and let Frank

talk. The reason I—

and you know this, I mentioned the 19.2K networks. And

you mentioned that, yeah, there's still a lot of that tech in use. I'm not

surprised by that. The— and I

imagine some of our listeners are saying, what,

you know, we're running 1 gigabyte Ethernet here and

understand the different use cases and why it is that way, the way

that it is. And this is going to lead to my question. So the

reason you want 19.2 kilobaud

networks still operating your power plants. The reason

they run that slow in the data transfer is because

they are deterministic. They're inherently

deterministic. And whereas Ethernet is

designed mostly to be able to respond to errors and

error corrections and stuff like that, it's designed to work

in an inherently non-deterministic environment. But it

doesn't mean non-deterministic in the sense of

AI being non-deterministic. And that's my

question. If you're interacting with specifically

utility companies and you're— you reinforced again,

you're looking at manufacturer's documentation, you're trying to predict

end-of-life predictive maintenance and stuff like that, which

I can see AI doing beautifully. Yep. What's— how does that

juxtaposition work For a company that

is called a utility because, you know, it's a utility.

How does that work? So we're very intentional about trying not to

cross that line between the data and the

operation of the devices, right? We don't want our

stuff touching— it's really read-only. So

even when we pull data, it's coming off the historian. It's pulling out of

the historian, which is just essentially the data collector, and then we're

reading into it and kind of looking at the trends and analysis. Even when we

talk about our security practices, when we're looking at the

alerts and monitoring, we're just, you know, again, a lot of these devices have

their own protocols, right? People think everything runs on TCP/IP. Nope,

not in a power plant. And so Defender for IoT needs

to be able to understand those protocols. And so

we pull all that into kind of a central repository where it gets

analyzed and we can you know, see if a device is

performing or doing things it's not supposed to, or if there's a

new device that showed up on the network that, you know, a manufacturer showed up

and didn't tell the plant operator or manager and just drop it on the

network. But all, all of it's very hands-off.

We work with a lot of the manufacturers to kind of help

them and to understand the impact that they're

having, because a lot of times these vendors aren't talking to each

other. And so, and even the plant operators

don't know what they have. And actually being able to

see it on a network map and understand how everything's talking together

really gives them a lot of great insight. It sounds a little

like AI-powered manufacturing execution.

Yeah, yep, absolutely. And we're actually working with some of our manufacturing

clients too. That's probably where we're going to go next. But since we have such

deep expertise in the power generation, industry. That's where we're going to start

and then probably shift over to offering these, uh, this

product to, uh, to manufacturers as well. Very cool.

Yeah, I mean, I think it's interesting you point that out because, you know, one,

you know, the '90s were maybe 25, 26 years ago at this

point, right? Not that long ago in enterprise

IT particularly. I've noticed the pattern is that the more

things are reliant on real-time and, you know,

infrastructure-critical stuff. They run older tech, not because

they're lazy, it's because it's tested and proven. Yeah. Right?

Why rip out something that's worked for 20,

30 years for the sake of getting something new and shiny?

And I think that, that introduces some interesting things. 'Cause Andy and I were talking

about this the other day where, you know, Andy gets a lot of guff

for, working on SSIS, right? I

remember, what was that tweet somebody said? Radio still loves you.

I think they were quoting Radio Gaga or something like that. Because you

were talking, you were doing something new with SSIS. And it's kind of like,

not everybody. Yeah. And in your talks, Andy, that you've given,

you've said like, how many people have stuff on the cloud? And a lot of

hands go up. How many people have critical things on the cloud? A lot less

hands went up. Right? And I think that is, it's not a knock on

the technology or the technologist. It's just the risk

of replacing something that works

to someone who is the HIPPO, the highest paid person's opinion, is just

too high, right? So it's kind of like, if it ain't broke, don't fix it.

And I think that even if there are, there has to be a compelling

use case. So you're nodding, so you tend to agree. So I'll leave you with

one more thought. My teenager referred to the '90s as the late

1900s, which made me feel really old. Um,

but I mean, he's not wrong, but I had it— I had it took the

breath out of me. I'm like, oh my God, the framing sounds a little off.

Yeah. Um, yeah, but, um, no, but like, what,

what would— what do you see as the compelling use cases for these, these

industries to upgrade? And I would imagine they don't upgrade

kind of like whole rip and replace. It's kind of more of a gradual

layering. And again, on the IT side of the business, we

have a lot of opportunity to show them kind of what modern looks like

because we're not running turbines and nuclear power plants.

So that's a pretty easy way to convince them to

lean heavily onto cloud services and cloud solutions. As it relates

to the operational side of their businesses, you know, we show them, okay,

you're working with Siemens, you're working with GE, whatever. They have their systems.

The software is what it is. It runs on whatever it runs on, but that

doesn't mean you can't start to bring in cloud services that sit on the edge.

For example, Defender for IoT will take

up the signals from all these devices, but then it's pushing it into

Azure where all the actual data analysis is happening.

Because it doesn't need to be real-time and it's not critical, and if it

goes down, it's not the end of the world. You slowly look

at how do you introduce some level of modernization, to those

elements of the business where we have the ability to make an impact.

And then, you know, as time goes along, and if they have a

vendor now that's a bit more modernized, and the system is,

you know, requiring new infrastructure, we look for opportunities there to

kind of see where we can modernize without rocking the boat

too much. Has there been a lot of pressure

on utilities In regards to the AI data center

race, because I've heard, if you're watching the video,

that's actually Bloomberg playing in the background. I'm a Bloomberg junkie. I

was there today. They were talking about, you know, that

data center AI usage, they tend to blur the two, but

data center usage is about 4% of today's electricity in the

US and is projected to, in the next 4 or 5 years, hit

something like 25% or 20%. Yeah.

If I'm a power utility, if I'm a utility company, I'm probably

both excited and alarmed by that statistic.

Equally. Equally. What are they trying

to squeeze more efficiency out of things or are they

trying to build new capacity? Because I know building new capacity is—

Yeah, it's kind of all the above, right? So yeah, I was just reading

some articles just recently. Meta and Microsoft have

committed to recapturing a lot of the

heat that their data centers are producing that they can store

and then provide back to the

communities that they're actually setting up in, right? They're also looking at

all kinds of ways to be more efficient in the way

that they utilize electricity, right? And it's inherently going to happen right now

because the development is going so fast. we haven't—

the ability to squeeze as much efficiency as we'd

like hasn't caught up yet. But at the end of the day,

these companies are incentivized to be more efficient, right? Because

they're gonna keep the prices where they're at. Right. They're just gonna reduce their costs.

So I expect you'll probably start to see some really innovative

ideas because the other thing too, as you're seeing, is there's a lot of

public scrutiny over these, over data centers. And

so if these companies can't prove that they're

being good stewards of the water and the power and the, you know,

energy in those communities, they're not going to get built.

And so I think the ones that can do that and ones that can prove

are going to get their projects approved, and then the other ones are going to

have to adapt to, you know, in order to be able to compete.

And I can see, Jim, how that plays right into the

services and the offerings that you're providing.

Sure. The very collecting the data is the very first

step, right? What's it really doing? That we always play this game of

estimated versus actual. You're reading the actuals, you're

able to feed that back into it. It's 100%

closed-loop engineering, which is how we've done

all of the engineering that's been successful the past couple centuries.

Yeah, no, absolutely. Yep.

What motivates the highest-paid person's opinion to take on the risk of

upgrading systems, or they do it in non-critical things like they'll have

Power BI reading data from existing real-time.

I always put air quotes around real-time because once when I

was but a humble younger Microsoft employee, I

mentioned, talked about real-time monitoring to a bunch of physicists at the

Silicon Valley Research. Microsoft used to have a

research campus there, and boy, did that go over poorly

because then it devolved a bunch of PhDs in a room

and physicists, and they said, well, what is time? Ultimately, that— talk about going

down a rabbit hole. Highly philosophical, yeah. This was like a— Sounds like a conversation

with my son, yeah. Bottomless, Bottomless

rabbit hole of like, what is time? So I always put real-time on that because

for some people, real-time is a, you know, a weekly report,

daily report, you know, or because I

also think that you do hit a vertical wall of if you needed, the cost

goes astronomically up if you need

sub-second kind of responses. And most people don't really need

that. No, for most people. Same thing with disaster recovery, right? Everybody

wants to be able to recover to the last minute or last second. Right. You

know, we're doing, we're doing RPO and RTO reviews, you know, real-time

point of— yeah, restore point objectives and recovery

time objectives. Nobody needs that, right? 1 hour, they can, you can

maintain, you know, your business and lose 1 hour of data. When it comes

to the conversations we have with our customers, as far

as trying to convince them that why they should modernize, or

why they should improve the efficiency, it always comes back to

proving out the value. And if we can't make that argument, we

can't show them why it's going to save them money in the long term,

or in some way give them some sort of competitive

advantage, then they're probably not going to listen to us. So we do a

lot of ROI modeling. We do a lot of total cost

of ownership when we're going to do a project. You know, how much is it

going to cost you? When is it going to pay off? You know, what's the

price difference between upgrading that on-prem data center and

renewing all those license contracts versus going to the cloud?

Right? And we're very transparent about those costs. In some cases, it

doesn't make sense. And so in that case, we say you

either leave your workloads on-prem or you do a combination of

the two. You can take some workloads to the cloud, but you leave some on-prem.

You know, we're not, we're not here to Just push cloud for the sake of

cloud. Right. You know, and it's a thing because I,

you know, haven't worked in that field in my, in a previous career

30 years ago. The whole idea of when I jumped the fence

from being a hobbyist coder and, you know,

manufacturing integrator and went to work at a— I kind of

got to where I am through manufacturing. I went to

work for a large plant that was doing similar things to what you were doing.

They were manufacturing. Instead of, you know, a utility.

But at the same time, one of the things that struck me, and maybe

some of our listeners are hearing this and being struck the

same way, when I jumped that fence, they had a

project management office. And I was like, we need an

office for that? I mean, this is part of the job. You can't

do this work in the manufacturing vertical.

without just considering a host of things that they kind of

partition off into project management and

enterprise software development, enterprise database, you know,

stuff. So pretty interesting that you're bringing all of that up.

Yeah, and to be honest, I mean, I think when you are in a manufacturing

space, everybody thinks like a project manager and an engineer,

but for a lot of our customers, they don't have that discipline and

you know, they've tried to do these things in the past, whether it's AI,

whether it's a data transformation, whether it's an

upgrade of their environment, and they've failed because they don't have the discipline.

Right. And they really need an organization like mine that

can plan it out, can plot it out, can define the risks,

can establish the timelines, the objectives, the goals, the

deliverables, because they just don't have that type

of structure and discipline within their organization. So,

well, yeah, lifecycle management is huge. And you're right

that plant engineers may think about that, but

people below them are, are so focused on, you

know, on granted important stuff, right? Keeping the plant

running and making sure that, you know, exactly, you know,

and respond— Literally, literally, literally.

And respond accordingly. Yeah. You're right. That is, that's a

different approach. That was actually a different observation than what the one I was

trying to make, but in the opposite direction, but equally

important. So bringing project management skills

into this arena, specifically in their own,

not a silo really, but, you know, in and of themselves,

very important. Yeah. And bringing this back to kind of the AI

conversation, because this is what I'm mostly, you know, doing

today. That's where most of these projects fail, is they don't have

a sense of why, why are we doing this, right? Why did we— did

someone just tell us we were supposed to, or was there an actual business,

a business problem we're trying to solve? And before, and

before you can get to the tooling, you need to have the conversations around what

outcomes are we expecting and what are we trying to solve. And

understanding that, you know, especially with AI, it's, it's a big

transformational moment in an organization's history

that affects change management and affects

almost practically everything people do and the way they're going to be doing

things in the future. Very true.

One of the things, one of my customers for a time when I was in

sales at Microsoft was a utility

company in Pennsylvania. They, we were trying

desperately to sell them Azure. Right? Because that's what we were

comped on. But the guy there said, well, because

it's a power utility, our incentives are not

the same as a regular enterprise to basically

rent servers. It's cheaper for them to buy stuff. Is that true?

It can be. I mean, it really kind of depends on the, you know, if

you look at utilities, they're getting their energy at a discount

rate. Right. Right. But also too,

sometimes it comes down to the type of people they're hiring

and the skill sets they have. And so sometimes the resistance

is, is not an economic one. It's a

skilling and a kind of a,

a anxiety situation. So we've definitely run into organizations where

folks have felt threatened by it. Yeah. And we have to, we have to kind

of try and get above them and say, This is not here to

replace this individual. It will change the way they work and they will

have to reskill, but it's not meant to eliminate

them. But again, depending on the individual

or the culture of a specific IT group

can sometimes dictate whether or not they're willing to

invest in modern technologies. We're back

to the human factor being a major Player. Always there.

Yeah, for now, at least till the AI takes over.

That's so, so true.

The, um, so, uh, Proloquo specifically, you'd mentioned some of the,

uh, some of the software you're developing. I know you can't talk about what you

mentioned that you're hoping to release in the next couple of months. What do you

have out there now that, you know, is helping, and maybe a use case

on how it's helping? Sure, yeah, we just launched a—

we're a cloud solution provider for Microsoft, which, Frank, you know, probably

know what that is, but we are able to resell licensing and services

from Microsoft. We're a Tier 1, which means we go direct.

And so one of the biggest things that other CSPs

have complained about and griped about constantly was

the ability to see all this licensing and be able to

give their customers a portal into their

environment and see how their licensing is being utilized, how they're

consuming resources in Azure, and then giving them insights into

how you can make adjustments. So we built an entire platform

that is AI native. We call it Vector.

And Vector, essentially, we sell both to CSPs,

but we'll also— we also use it ourselves for our own customers. But it gives

you a platform to go in and actually view what

you're spending with Microsoft and how it's being utilized

and potentially where you could be saving some, some money. So, you know, you can

look across your Azure VM instances and say, you know what, you

could probably change the sizing of these. This is some of the

telemetry we're pulling from Microsoft, but it's under one

single roof and we're going to continue to add additional capabilities to that

over time. And that's been very well received because Again, like I

said, it's one of the biggest complaints from customers is we just don't feel

like we have enough visibility into where our money's going.

And so this makes it very clear. You can simply pull

up the chatbot and ask it questions. You know, what did we spend

last month on Azure? What did we spend this month?

What's why? What's the difference? Well, it seems like, you know, you had,

you had to go up a SKU level with your Azure Managed Instance. Right?

And, you know, track that back to some decision that somebody made

or some performance issue you had. So again, this is the, you

know, that's our most current platform that we're moving forward with.

And it was about a year's worth of development and feedback

working with our customers to see exactly how they wanted this data

to be displayed and how to manipulate the data.

But over time, we're going to continue to add additional functions and function

capabilities to it. Yeah, the visibility

into AI spend, I think, is going to be a hot topic because

we're beyond that point of let's just throw money at AI and

the ROI will magically appear. Yeah. Clearly that didn't happen.

And I don't think that's throwing money magically at something and

ROI appearing. I'm not aware of quite when that

happened, but everyone seems to do it.

But now I think the question is, you know, everyone is

talking about an AI bubble, AI burst, right? And all that. But I think

the question, the smart people are going to be asking questions, where's the money

going? Right. Right? Before the rug gets pulled, because

you want to make sure you don't pull the rug over the, you

know, where the money's being made, right? You want to pull the rug

where the money's going out the door with no value. What's your

take on that? This is something we spend a lot of time on, and

when we do our engagements with our customers, I think

for the first thing is they don't even know how to begin to calculate ROI.

Right. And how to establish use cases to determine

the value that it's bringing to the organization. So we spend a lot of time

on educating them on, well, first of all, pick the right use cases

if this is your initial foray into AI. You want to pick those

high-value, low-effort use cases, baseline them

today, find the metric that matters, and then

determine how you're moving the needle. And then actually implement

to that. So if you see that something's saving you X amount of

dollars, but it's only 40% being adopted,

lean into that. Give some, give your users some more training or figure out where

maybe the process needs being, needs more rework. Because you

didn't spend enough time thinking about the business process before you actually brought

technology into the equation. We have another managed

service that we'll be introducing in the next few months that's all

around AI. It's really good. AI managed services, and it's

going to be looking at those particular signals, and we're going to

help organizations define those use cases and their

overall spend on AI, no matter where it is. We're going to pull out all

of it, whether it's all Microsoft or if it's enterprise

ChatGPT or enterprise Claude, get those into the

system and then start documenting your use cases and the

value they're bringing. So you'll have a widget right on your

dashboard that shows you this is what you're spending, this is what

your return on investment is. And, you know, for every

use case you add or every other investment you make, how does

that adjust your ROI? And again, I

think customers are absolutely begging for this because I think to your

point, Frank, right now it's, here's some

tools, we have some extra money here, go

use it, you know, tell us how it's working. But there's no

real formalization of how are you measuring it and how are

you defining the value that it's bringing back to the

organization. And what gets measured gets

managed, as they say, but also, but also drives

behavior, right? You saw very quickly Amazon had

a— I think it was Amazon or AWS, whichever— had

a token leaderboard, right? How many

tokens you could burn. Yeah. And then after a few

surprise bills, again, for Amazon, I don't

know if they disclosed the number, but if they had trouble,

if they had issue with with paying, I can imagine. Yeah, well, they're measuring the

wrong thing, right? And I wrote— Exactly, yeah. I wrote a couple articles about this.

Meta was doing the same thing. They were basing your performance review

based on your AI usage adoption, right?

Usage does not equate to outcomes or value.

And that's where it's crazy to think that an organization like

Meta or AWS is not— doesn't know how to

solve for that or how to speak to that. But that's really where

we focus all our energies. Yes, adoption matters. Yes, we want

your users to be using the tools. We want to know how they're using

them. We don't want to know exactly what workflow

or work process is being transformed and

what is your vision for the future. So we also do road mapping. So

today you're, for instance, capturing meeting notes. A good example is our

help desk group. Our original use case was capture meeting notes with

Teams, get those notes into your tickets, you're going to improve the quality of your

tickets, we'll give you templates to use, you know, it's better feedback to our

customers. But we said in the next 12 months we

want to be able to integrate our ticketing system to Teams so that we can

actually push the meeting notes into them automatically, right? So

reducing the friction, reducing effort, and even driving even

more value. So now we're not relying on the users to

remember, oh, I didn't turn on transcription and I forgot to copy the

notes and, you know, whatever the case may be. So we

help those— our customers through that entire journey. I love that.

It feels like an automated knowledge base

application, and I can see so many uses for that. As

I can also see for that widget you described, uses outside of not

just utilities but also outside of manufacturing. Heck, I'd love

it for my software development enterprise.

Yeah, I really think that there's the real opportunity here is

knowledge capture or capturing an organization's knowledge into a knowledge

graph or whatever hip term, you know, there is

because— Yeah. If you, if somebody retires,

somebody leaves, somebody leaves the organization or someone's new to the organization,

there's a significant onboarding, right? I mean,

it's not unusual for like if you're working at a big tech company, you don't

really feel like you got your sea legs for about 12 to 18 months. Yeah.

And it's kind of like when you think about, I'm not

naive, I don't think you can get it down to a day, right? But if

you can get that down to a year or 6 months, right? That's

an enormous— because there's just so much tribal knowledge locked away in

largely documents, I think. And, you know, not

necessarily in relational data stores or anything like that, right?

That I think that they— AI, what excites me about

AI, and right now RAG is the poster child for that. I don't think

RAG, RAG will get you a long way, but I think RAG is going

to evolve into something better in the future.

But yeah, I think that, and I think Microsoft actually has a pretty good

advantage here because think of all the stuff that's locked away in Word documents, Excel.

Yeah. Sitting in SharePoint servers, right? Right.

Yep. That it just seems like they have the home field advantage.

Yeah, I wrote an article on that too regarding, you know, that a

lot of companies are looking at Anthropic or ChatGPT and saying,

well, we can connect our— those systems to my M365

environment. But when you do that, that's literally just

RAG, right? You're just retrieving information. The advantage

Microsoft has is they now are implementing something called WorkIQ.

which is providing a contextual layer over top of all

that data. Mm-hmm. So it's, it's applying the understanding of

what your business is, what your departments are, who works in

every role, what they work on, how they work. So

even just recently, and I use all these products on a regular basis, but I've

noticed significantly different outcomes when I'm

trying to get a response from either ChatGPT or

Claude related to data within our M365 environment,

it doesn't have that context. Right. Now I ask Copilot and it has much

deeper context, and that context only builds and gets better

over time. Kind of like hiring an assistant. Day one, they don't really

understand the business. You know, within a year, you can just look at them

and give them a glance and they know exactly what you're— what you mean and

what you want done. Going back to the other topic about tribal knowledge,

A lot of these industries have a workforce that's aging out.

Yeah. And there's a lot of technical skill that's leaving the door,

manufacturing especially, very specialized skill sets.

And so wherever you can apply AI to cover

that gap, great. We, I have another customer I work with,

manufacturer in the Buffalo area, and they're looking to roll

out an agent to their plant floor staff.

that on their overnight shifts that will help them

troubleshoot problems on a line so they don't have to call

up an engineer in the middle of the night and say, I'm getting an error

code on this device, or I'm getting an output and I'm

getting streaky lines on this printer

or whatever. How do I fix that? So what they're doing is feeding all of

this data into this agent source data,

and allowing the operators to be able to ask questions

and actually walk them through how to troubleshoot before they have

to pick up the phone and interrupt somebody, or they have to hire

someone to stay on that shift to walk around and help those individuals.

I call it the 3 Ds, right? During the age of

robotics, which we're still in, but, you know, the concept of bringing

robotics to the industry was to remove the dirty,

the dull, and the dangerous, right? Those are the things that you want robots to

do. In the day— age of AI, and the— for the information worker,

it's really about removing the dull, the draining, and the distracting.

So if you can remove those 3 elements, you can really allow people to focus

in on the core parts of their job. So to your point,

Frank, maybe that new hire you onboard, we've

reduced what they're actually responsible for because now they

have a team of agents that do the other stuff. They don't

have to be an expert in administrative work, right? The

administrative stuff gets done for them. They can be experts in the thing

that they're really good at and let the, like, the technology take care of

filling out timesheets and generating pre-sales proposals

and what have you. No, I mean, that's a great way to put it, right?

Like, you know, you— every system, every company has

a different way of doing basic stuff, submitting expense reports,

logging time, right? It would just be nice if you you could just talk to

a chat saying, hey, like, I'm going to take next week off or whatever,

like, or I have a doctor's appointment, right? Like, it'd be nice to do that,

block your calendar out. Like, these are all things that I think are technically possible

today. We're getting pretty close. Yeah, we're getting really close. It's

just the interconnect between these systems varies pretty

wide, wildly or widely. I don't know. I think

both apply. Both. Yeah, both actually apply. You mentioned

Copilot, and I have to say Copilot, Look, the first

version of Internet Explorer was awful. Version 2,

less awful. Version 3 was acceptable. I think Copilot

followed a similar trajectory, although now we don't have neat version numbers.

But I recently, I tried Copilot when it first came out,

right, to analyze some, to do some data cleanup for me in

Excel. And it failed miserably. It was awful. So

much that I didn't touch it for like 3 months. 3 years. Yeah. So the

other day I installed Claude for Excel and it

did what I wanted it to do. And I was like, you know, just for

grins, just out of the interest of science, right? I

was like, let me see what Copilot would do with this. And I

was suitably impressed, right? Like, I'm not a Microsoft MVP

anymore. I don't have to watch what I say or whatever, but I was suitably

impressed with it actually did what I

expected and a little more. It seemed to have a little more. I think you

hit on this earlier, context, right? It pulled in context. Now, some of the

context it did pull from a worksheet tab that

Claude generated, but I thought that was interesting. It didn't have— it

thought to do that itself. I literally asked it the same question, can you find

patterns in this data? And I had the same range selected. And

I was impressed with how cogent the response was,

how quick it was, and it seems like it's matured

pretty well. Yeah. Yeah, if anyone hasn't looked at it in 6

months, I really would suggest they do because here's the other thing

that most people don't realize. Copilot is driven by

Claude and ChatGPT models. And you

can select— you're not trapped into any specific models. If you know

ChatGPT does one job better than Claude, pick that model.

And that extends to chat, it extends to the applications, it extends to

co-work. which really is a very, very competent

solution now for Microsoft. It does cost you extra. It's

credit-based. But we were in the

beta period, Microsoft likes to call it frontier, and we used it for

about 3 months. And within a 2-day period, Microsoft dropped

on us, we're going to usage-based modeling. Yeah. And

our users almost lost their minds because they had

become so accustomed to using it. It was

already embedded in so many of the things that they do, and they had

found so much use in it that they were coming to me nonstop. Are

we going to get rid of it? Are we all going to be allowed to

use it? How much credit— how many credits are we going to get? You know,

we had to quickly get a snap meeting going with our AI Governance

Council, figure out what we could budget, figure out what

policies we would put in place to let people use it. And then, you know,

as the enablement person, It was my responsibility to tell them,

here's where you should use it, and here's where you shouldn't use it. Here's,

here's the right tool for this job. I actually built an agent that

will help our users determine what's the best tool for the

job and actually give them a kind of an implementation strategy

and actually help them even try to calculate how many credits

this particular task would take before they just jump and use Copilot

Cowork. to do those tasks. But yeah, if, if you've been,

if you've been overlooking or, or looking past Microsoft for a while

now, I would, I would suggest you come back because

it's not the mic— it's not the Copilot of, I would even say,

6 months ago. And, you know, it's the, the ability of it to

create incredible content, to reason, to

pick up on those, that contextual information across your organization.

To be able to find things you don't know where they are. Our users stop

searching. They just go to Copilot. I'm looking for this document. Boom, there it is.

Right? Which used to be a huge pain point. Which is a

huge time wasting. And I think that when the dust

settles of this, the ROI is going to be finding that knowledge,

even if it's your own stuff, right? Yep. And not, and, or your

team, you know, your team stuff, right? That is a huge pain point. And I

do wonder, there is an ROI there. There is a there in them

thar hills, as they say. But, but how

do you measure that, right? If I say, like, you know, oh, it would take

me like an hour to find this one PowerPoint presentation, right?

Like, I think a lot of that goes unnoticed, whereas now I

can find it in 5 minutes. That's the stuff that's hard to track. We do

track what we call assisted value, which is, okay, things

that, that you're using this on a frequent basis

and the intensity of your usage. ROI modeling is

much deeper than that. That's stuff you have to actually— we have workbooks to

help walk you through the steps and what are we doing

today, what's the blended hourly rate for the person this is

going to affect. We really try and make this as simple as

possible for our users. But if I was to provide you one

anecdotal story about a company's journey, I have to produce

executive reports around ROI and Copilot. We have about

160 licensed users out of a company of 400.

Our original monthly ROI from our investment in

Copilot was running somewhere in the neighborhood of

$3,000 a month when we first started using it, which

barely covered the cost of Copilot. 2 years later, we're now

reaching close to $80,000 a month. Wow.

So we're seeing close to $1 million of assisted value

before we even start to go in and look at the individual use cases

and start to evaluate those, which is what we're beginning to do because

we're our customer zero for this new managed service. And we

want to be able to document all of our use cases, all of

the users that are affected by it, and our expected

return on investment and be able to plug that into the system. So

That's phenomenal. And I could see the products you mentioned growing out of

that internal effort. And I'm just curious, did you see that

happening when you were developing it internally? Did you, did you cross a

threshold where you went, our customers could use those?

Yeah, yeah, absolutely. And, and, um, it's not without a

lot of work, right? So that's kind of the thing that a lot of clients

are a little misled. Anything that, that's going to bring enough,

bring transformational change to an organization is going to take

effort. But absolutely, we, we are very firm believers in

consuming our own services. So we consume all of our own security

operation services. We are a client of ourselves. We're a client

of ourselves on the support front. We're going to be a client of ourselves in

the managed AI space. And because

we've kind of figured these things out and we can then go to our customers

and say, and they tell us, and a lot of customers said, it does, but

does it really work? Well, yeah, in fact it does

because I can show you that it works. So right out of the gate before

we have one single signed customer, we can show how

we're using it to leverage and be able to demonstrate to our executive

teams. That's cool. It

becomes a— you dogfood it and you show it because I think

that's really going to be the conversation in the corner office when anyone

comes and tries to sell AI anything. Right? You know,

they're gonna— they're, instead of being blown away or not understanding, they're

gonna sit there like this with their arms folded, you know, oh, another AI

solution, right? Yeah. Like, how do you prove it? I think that's a— I think

knowing what you need to measure and figuring out how you do it

is going to save a lot of companies who are selling for sure.

But I mean, to have that conversation, because now the question is you threw this

much money at something, what'd you get? Yep. Again, like, We

circling back to where we were not that long ago talking about that. But no,

I really think that that's going to be the next frontier of questions

that anyone has to answer from the board to the sales.

And from our perspective, we had to have a kind of a mind shift as

far as how we pitch this stuff to our customers.

We've stopped talking about tools and technology and started talking about

outcomes. We're selling outcomes. We're not selling the tools. The tools and

technology are almost irrelevant, right? Right. We can swap in anything we want.

And in some of our managed services, we intentionally don't talk about

what we use on the backend so that we're not getting some

people who are, oh, I hate that company. I don't want you to use that

product. Really what we're delivering is outcomes. And if you

can demonstrate that what you're spending with us is worth it,

then it almost becomes irrelevant of what tools or technology we're

using. Well, it sells itself as well, you know,

and, and having that foundation, which,

you know, a lot of folks in consulting, I think,

confine themselves to consulting. And I'm not saying that's a bad

idea. It's going to depend on the personalities involved and the problems

you're trying to solve. And certainly not all problems lend

themselves to the class of problems and the opportunities that you guys

are creating. But often you'll see

consulting companies do something similar to what you're doing. They'll

start seeing the same problem over and over again. They may

experience something like it, something analogous internally like

you have. And even if they don't do that, just seeing

it with several clients starts to present that opportunity

that, hey, you know what, we could maybe productize

this. And, you know, or even in something as,

I'd say, as nebulous maybe as a methodology, but certainly

you could take it to an application. And then

that's, that I think is wonderful. We have an instance of that going on

internally with Frank. He's done a lot of work

around automation for the podcast. Mm-hmm. And so

this podcast, we're recording it. I'll say the date. It's the It's

actually on the East Coast, 2:06 PM on August 12th.

And I say that so that you can look at your clock, viewer, and

see how long it took for this to get to you. And if you run

that comparison against other podcasts, which have

more than 2 people working on it, you may see

that, you know, the automation work that Frank's put into this now for what, Frank,

8, 9 years? Something like that. Yeah.

And if you look at franksworld.com, I'm able to get out a blog

post amazingly fast. Like, I think— Yeah, I tease my wife,

if I got hit by a bus tomorrow, she's going to see posts from me

on LinkedIn for the next 2 weeks. So, don't freak out. It's

funny you mentioned that because earlier this year, I had a concussion,

January 8th. And if you were just looking at

my output, whether it was on the blog, on the podcast, or

whatever, You wouldn't have noticed. I had about 2 weeks of stuff

kind of queued up. Yep. So you would have noticed after 2 weeks things—

I was still on— I was still on the mend, but like, you could— it

took about 2 weeks. And that was more of— that was not a function— that

was just a function. I only queued up 2 weeks of stuff. If I had

done months' worth of stuff, which sounds ridiculous, but you wouldn't have

noticed at all. Right. So you're right. Like, there, there is—

and I encourage everyone within the sound of my

voice is to try to automate just stuff in your own personal life, whether it's

your own personal blog. Yes. Because the nice thing is that if you do it

on your own time, you can be very experimental. A lot of the stuff I

figured out has been for Frank's World, has been for this

podcast. And I don't have to— the only person I have to convince is me.

I do have enough of my own home lab where the

AI is just electricity in my house. Right. And when I

have everything up and running, I have the Spark up and running and the room

gets a little warm. But in the winter, that's a positive.

Yeah, exactly. Yeah, no, you know, our CEO has been

very— he kind of pounds this drum to the teams. I'm

on our innovation team that because of AI, you know, with

the expectations for the ability to deliver

is really coming down from months to

weeks. And so as an organization, we have to start thinking

that way. We have to accelerate the ability to deliver

in a much more compressed timeframe because the expectation is so much greater. Because if

we don't, there are already other companies that are doing that. So

that is A, automation, B, AI, and in that

order too. So, you know, we have a lot of conversations with customers

and They go, well, couldn't AI do that? And I'm like, yeah, but that's not

an AI problem. That's an automation problem. So let's start with the automation and

then layer in it, right? AI is kind of secondary to

automation. And so it kind of opens their eyes up. Well,

wasn't it the same thing? I'm like, no, not at all. You know,

RPA has been around a lot longer than, than AI has, but AI

definitely brings some unique and interesting capabilities that you can layer

on top of that. Go ahead. I was

going to say, Jim, you probably experienced this. We

have, both Frank and I. There's a whole

host of things that in the past I would have looked at it, had an

idea, and thought I should— I could build this, you know. And then

I've gone, you know, my inner practical geek has done the

math and gone, I could, but I don't really have time.

And no, that's no longer a thing. No. Yeah,

no, it's, it's, it's truly amazing. And I live in a house where

everything's automated, right? My, my desk is automated. You know, my

parents, who are now— my father's in his 80s, my mother's almost, almost

in her 80s now— they come here and like, your house almost has its own

rhythm in life because things just happen, right? The lights come on, the lights

go off, the blinds close, the blinds come on. It's based on the season.

And, you know, it's intentional. I don't want to have to deal with this stuff.

And if I can automate it, I will. And it's kind of nice to know

it's on doing its thing. The vacuum comes out, cleans the kitchen at 10 to

30. The, you know, the— and then it does its weekend job and cleans the

rest of the rooms. And I'm just sitting here working away, right?

It's those 3 Ds you mentioned earlier. Absolutely.

Yep. Yep. That's cool. So I

see we're, we're a little bit past time. Want to be— we could talk for

another hour or 2, but we want to be respectful of your time. Easily.

Um, Andy and I like to talk, which, uh, sometimes is an advantage, but

definitely with the podcast it's an advantage. But, uh,

where can folks find out more about you and ProArc? Sure. Probably

the easiest way to find out about everything I'm up to is on my

website, which I actually just launched, uh, last month.

Um, it's my last name,

spignardo.com. Spignardo.com.

I have a link to my, my book that I recently published in,

in March. All of my LinkedIn articles are on there. A

lot of my podcasts, any projects I'm working on,

any appearances, a wealth of information there. As

far as Proarch is concerned, you can follow us on LinkedIn,

P-R-O-A-R-C-H, and our

website, proarch.com. We have tremendous amounts of

resources, webinars, white papers,

all types of different tools that people can download to kind of

help them in their daily activities. And

it does a great job of explaining all of our services and our products as

well. Very awesome. All

right.