From pilots to performance: Accelerating AI value in energy and utilities

Send us a text

 How can energy and utilities move from AI pilots to measurable performance? CGI’s Peter Warren and Frédéric Miskawi explore how AI-led software acceleration, smarter data use, and the right algorithms drive rapid ROI, resilience, and business value. Tune in to learn how to turn experimentation into sustainable AI success. 

Visit our Energy Transition Talks page

Episode Transcript

Peter Warren (00:04):
Hello everyone and welcome to another edition uh
on our podcast series EnergyTransition Talks.
Uh today we have a veryinteresting one where we're
going to be diving into how doyou get AI to ROI?
Uh we're we've a lot of peoplehave been playing with AI, uh,
but now it's uh time to movebeyond that and actually get
results that actually improvethe bottom line, improve

(00:25):
efficiency.
Uh to that point, I have agreat guest.
Fred, uh, do you want tointroduce yourself?

Frederic Miskawi (00:30):
Hi, everyone.
Uh Fred Miskawi.
I'm a part of our global AIenablement team at CGI, and I
lead our AI innovation expertservices, which uh luckily has
me work across nine SBUs acrossgeographies, and I've been
involved one way or another withartificial intelligence since
the 1990s, and I love what I doevery day.

Peter Warren (00:52):
And SBUs is uh our business unit, so that includes
uh countries like uh Canada,North uh, and uh the United
States.
Uh it includes all of Europe,uh, Australia, and so on.
So we uh have quite a bit of amixture in that nine different
groups in the UK as well.
Um so let's kick it off hereand talk about you know AI ops,

(01:13):
data maturity, where do thingsstart?
You know, a lot of things havebeen moving forward.
Uh people are looking forresults.
How do they get something inthe next 90 days that actually
brings a benefit to them?

Frederic Miskawi (01:25):
I mean, the uh the fastest way that I've seen,
and this is what we do on aday-in-day-out basis, both with
our internal teams as well aswith client teams, is AI led
software acceleration.
Number one value for thesetypes of uh tools, this
technology.
We can deploy licenses fairlyquickly, but even with those

(01:45):
licenses, what we're learning isthat it's just not enough.
So we bring in a lot ofexpertise and layering above
that.
But you can get start gettingvalue very quickly.
It's gonna start in that 3-5%uh productivity improvement.
And it's gonna start a littleslow, but you can get and gain
that expertise very quickly.
You just need to have the rightpartner, the right guidance.

(02:09):
Uh, a lot of online learning aswell kind of can help.
Uh, and the biggest thing is towork with employees who are
leveraging these tools to learnhow to use them and to uh cater
to them and let them know thatit's okay to experiment and
learn.
And through these these uhapproaches, what we've been able
to do is to get peopleaccelerated fairly quickly.

(02:29):
And you're gonna get value veryquickly because you're
accelerating the process ofvalue delivery.
And software is what's poweringmost of our businesses today.

Peter Warren (02:38):
I think there's a belief out there that I can't
get started until my data ispristine or uh you know I'm
gonna have this is gonna be amajor overhaul.
And there's certainly a bit ofchange management, as you just
sort of alluded to.
You know, how do peopleapproach those things?

Frederic Miskawi (02:52):
That used to be true.
In artificial intelligence,you've got different types of
algorithms.
You've got uh structuredapproaches, unstructured
approaches.
And with these algorithms, whatwe what we were getting used to
over decades is that you neededvery clean data sets.
Uh key value pairs, you neededto be able to have large of the

(03:13):
large sets of these data, cleandata, to be able to find the
patterns, to enshrine thepatterns, to tweak the
parameters to get where you wantit to.
Um, what we're finding out withthese new algorithms is you're
now getting into a realm wherethe dependency on highly uh high
quality data sets is beingreduced.
And the more this technologyevolves, the less that

(03:36):
dependency exists.
To the point where what we'rehearing from our partners, uh
hyperscalers in other edge labsthat uh that live on the edge of
this technology is we'regetting into a new era where
data quality will not matteranymore.
So we're already seeing in thelabs uh approaches that

(03:58):
automatically clean the data,collect the data, get the data
ready for what's needed.
What we're seeing today inproduction is the ability to add
layering above the solutionsthat we deploy.
That layering enables us toapply certain heuristics to the
data that comes in.
So even if your data is notclean, we could say, by the way,
if this document is older thanX, or if this is version 5.3,

(04:23):
and then you've got a newversion, maybe let's take a look
at that.
Uh so there's very simpleheuristics like that that you
can apply in a um in these typesof solutions.
But because of this layering,and now with agentic approaches,
you're getting even lessdependency on high-quality data
sets.
Um, and what we're gonna see isover time, these solutions,

(04:45):
these agents are collecting moreand more data.
They're collecting it with thelevel of quality that they need
for the next generation offine-tuning or training.
Um, and you see this flywheeleffect that's happening today.
So, no, you don't need to havehigh quality data sets to get
started.
It certainly helps for certaintypes of algorithms, and it will

(05:06):
always be there for veryspecific types of algorithms.
But what we're seeing is areduction in the dependency on
those on those data sets.

Peter Warren (05:15):
Maybe let's talk about algorithms.
You've mentioned it a bit.
You've got a bit of a famoussaying there.
I think I I enjoyed it the lasttime we chatted.
I'll let you I'll give set youup to start off with that.
But um, you know, it you know,people are looking.
Do I build, do I buy, do Ipartner, how do I do this, how
do I put in governance, and uhuh maybe kick off the thought
about the algorithms.

Frederic Miskawi (05:35):
Yeah, so I think uh about life in terms of
patterns, in terms of data, interms of algorithms.
And for me, it's the um thebest algorithm for the job.
And we saw that very early onwith this technology in the last
two years where we were askingthese models to calculate 5,333

(05:56):
times 55.
And next thing you know, you'vegot thousands of multiplication
multiplications happeningbehind the scenes to get you an
answer, which may or may not beright.
And then the labs quicklyrealized that, well, hold on,
maybe we can just kick off avery simple little algorithm
that's procedural so that wecould get an answer in the way
that we know and and love.

(06:17):
So that's what we're seeing thebest algorithm for the job.
And our job in this business isto figure out what is that best
algorithm for the need that wehave, for the value that we need
to deliver.

Peter Warren (06:29):
So building upon the concept of the best
algorithm for the job, I mean,uh, one of the big concerns in
our industry is assetmaintenance.
Um, it's heavily impacted bythe weather storms, uh, both uh
and fires recently.
There's a lot of things goingforward.
How do you see AI sort ofcoming in and helping the
operations improve?

Frederic Miskawi (06:49):
Yeah, and that gets connected to that concept
of the enterprise neuromesh ordigital triplet where you get
that near time, near real-timeview.
We're seeing an evolution inthat space.
You're getting new solutions,weather maps, weather data that
is being fed into some of thesystems that we're working on.
Uh, partners like Microsoft,for example, are introducing

(07:12):
that kind of capability.
And with a new layering on topof that and new maturity in how
to absorb that data, you canstart working in that next
generation of predictivealgorithms, leveraging the data
to be able to navigate the dataand understand where which areas
of the network might be readunder certain weather
conditions.

(07:32):
That level of visibility,transparency comes together with
that growth of algorithms in amulti-agent type of ecosystem.
And we're seeing that evolve.
It's not a revolutionnecessarily, it's an evolution.
But that evolution is moving onan exponential curve.
So as you're evolving,improving your digital

(07:54):
solutions, going through digitaltransformations, using the
technology to accelerate themigration of legacy systems,
your long tail of digitaltechnical debt, um, you're
building this new capabilitythat enables you to absorb these
new data sets, absorb this newinsight, these new patterns.

(08:16):
And now you've got companieslike us that come in and build
this layering to give you thattransparency, visibility, and
understanding.
And from there you could startfeeding that into your planning
cycles.
And your planning cycles startaccelerating a little bit.
And now you're you're you'reempowered with a new generation

(08:38):
of solutions and and patternrecognition engines that enable
you to fine-tune what'shappening across the enterprise
and making sure that for limitedassets, you're deploying them
in the in the best placespossible.
We're seeing that evolve.
It's an evolution, not a not arevolution.
Uh, but I think it's animportant evolution of the

(09:00):
technology.
So when we hear about AIbubbles, for example, I I laugh
because I see the value everyday, I understand it, uh, I see
it evolve very quickly.
And it's um really at the endof the day, it's our human
ability to absorb it and to putit in practice.
And that's what we're seeing.

(09:21):
And and a lot of that data, alot of this this empowering of
the planning process, forexample, uh, you're not gonna
necessarily see that in revenue.
You're not gonna see thatnecessarily in margins, at least
not yet.
But it's there, it's happening.
It's accelerating your nearreal-time understanding of
what's happening in theenterprise.
And that that evolution, eventhough it's moving at an

(09:43):
exponential rate, I think isincredibly important.
And you've got to understandit.
You've got to understand thethe the digital push, the wind
that's that's causing thesethings to evolve, so that you
can start planning for the nextgeneration of solutions, of
digital transformations, of oflegacy realignment.
Um and and these are thepatterns that we see every day.

Peter Warren (10:07):
Yeah, it's interesting.
It's not a big bang, it's aslow uh evolution, as you said,
or a steady evolution, if notslow.
Thank you very much.
We'll talk about the nextpoint.
So don't use a large languagemodel just to do calculations,
in other words, uh use somethingthat's uh dedicated for it.
Um this really before we maybewrap up this uh part, and we'll

(10:28):
catch up in part two where we'lltalk about large language
models versus small languagemodels and the impact of
hardware, uh both large andsmall.
How do you see about the KPIsand budgets?
Where are people going, youknow, in the first part of this
coming year, uh the end of thisyear even?
Uh where do you think thingsare going and you know what
should people be putting inplace as a leader?

Frederic Miskawi (10:49):
So the the KPIs, we tend to work with
clients in a very fine-grained,fine-tuned way.
So depending on what particularbusiness outcomes they're
looking to get, we're gonnafine-tune the nature of the type
of KPIs that are being used.
So if we talk about um softwareacceleration, for example, um,
a lot of what we're beingrequested is to look at

(11:09):
developer productivity, the uhquality of what comes out, the
trust factor that comes withwhat gets produced, and
understanding how to leveragethis new technology in a way
that can cut your time in halfor by two-thirds.
And that requires a certain setof data points that you got to
collect.
So we work with hyperscalers,we work, we built our own uh

(11:32):
data collection engines anddashboards to be able to get a
feel for uh what are the trends.
Um, personally, when I look atthat, I don't look at individual
productivity levels.
I think it's a you're you'renot getting the value for the
money when you do that.
What we tend to look at aremaybe at the from a granularity
level, we look at the teamlevel.

(11:53):
We look at the value that'sdelivered by the team, value
delivered over time, the qualitythat comes with that.
So we have a set of KPIs thatcome with it.
So depending on the particularbusiness goal and the nature of
the solution that you'redeploying, there will be a
different set of KPIs.
When you're looking at uhchatbots or knowledge engines,
where you've got uh the need tounlock the power and the

(12:17):
knowledge of the enterprise, ofthe industry within and give it
to the hands of your employees.
When you take that path, you'vegot to look at certain things
like um the nature of theinteraction, how often these
requests are coming in, uh, thenature of the results that
you're getting from surveys.
There are a lot of differentdata points that you bring in to

(12:37):
make sure that you're gettingthe answer you're looking for.

Peter Warren (12:40):
Oh, that's excellent.
Well, thank you, Fred, andthank you everybody else for
listening.
Uh, we'll pick this up in parttwo and have a great day.
Bye bye.

Frederic Miskawi (12:47):
Thank you.