La série De l’IA au rendement du capital investi (RCI) de CGI présente des discussions d’experts sur la façon dont l’IA stimule le changement au sein des organisations et les mesures à prendre pour obtenir des résultats fiables. Dans cet épisode, l’animateur Frederic Miskawi est accompagné de Nicholas Morel de Google et de Diane Gutiw de CGI pour réfléchir sur les principales avancées de l’IA en 2024 et leurs répercussions dans le futur. Ils explorent comment les organisations peuvent utiliser l’IA de façon responsable et efficiente pour assurer un RCI, le rôle crucial de la gouvernance et les tendances émergentes qui devraient définir 2025.
L’évolution rapide de l’IA générative, particulièrement pour les applications multimodales, a été une caractéristique déterminante de 2024. Ce changement comprenait l’intégration du texte, de l’imagerie, de l’audio et de la vidéo au sein de solutions cohérentes.
« Ce que nous faisons depuis 40 ans en matière de données, d’analyse et d’intelligence artificielle est de tenter de ramener les perspectives vers l’entreprise et l’utilisateur final, explique Gutiw. Avec l’IA générative, nous y sommes réellement parvenus, et c’est là que vous pouvez constater l’évolution. »
Plusieurs organisations ont eu de la difficulté à mettre l’IA à l’échelle en raison d’un manque de gouvernance et de préparation des processus, malgré des capacités technologiques avancées.
Morel observe : « En 2023, la question était davantage de savoir “pouvons-nous y arriver?” Il est devenu évident que tout au long de 2024, la réponse était généralement “oui”. Mais pour ceux qui ont une mentalité de validation de concept plutôt que l’audace de lancer un projet pilote, il a été très difficile d’amener le tout à une plus grande échelle. »
Les stratégies d’IA efficaces reposent sur des problèmes d’affaires clairement définis plutôt que sur des outils et modèles.
« La clé pour atteindre un rendement de vos investissements est de comprendre vos priorités et de continuer à mesurer la valeur que vous détenez », souligne Gutiw. Pour ce faire, il faut aussi avoir « l’intention d’examiner la production et d’aller chercher ce bénéfice réel. »
Une approche de gouvernance structurée permet de s’assurer que les déploiements de l’IA sont évolutifs et offrent une valeur mesurable.
Miskawi note : « Il est facile de déployer une solution fondée sur l’intelligence artificielle. La question est : devriez-vous? Et y a-t-il une analyse de rentabilité, du rendement du capital investi en place? »
Les modèles adaptés à l’objectif visé et les solutions d’IA agentiques devraient permettre de rationaliser les flux de travaux et d’améliorer la collaboration humain-IA en 2025.
L’année 2025 sera celle où les « les vannes vont s’ouvrir alors que les organisations mettront en place leurs politiques et passeront à l’étape de la production, note Gutiw. Je crois que la chose que nous verrons le plus apparaître est une interaction, une collaboration et une assistance accrues entre l’humain et l’IA. »
« Je crois que l’avenir est prometteur, précise Miskawi. Pour ce qui est de l’IA agentique, je crois que celle-ci générera une valeur fort considérable en 2025. »
Les organisations prévoient de surmonter les défis liés aux politiques ainsi que d’exploiter l’innovation et les déploiements à grande échelle.
Morel ajoute : « L’IA générative comporte moins de barrières à l’entrée pour certaines de ces composantes. L’accent sera mis sur des domaines qui ont un impact significatif sur l’entreprise plutôt que de tenter d’appliquer l’IA partout, comme l’illustre l’adage de Mark Twain voulant qu’à celui qui tient un marteau, tout lui paraîtra comme un clou. »
Un soutien unifié de la haute direction est essentiel pour harmoniser les initiatives d’IA aux résultats d’affaires stratégiques.
« Il est impossible d’avoir des programmes d’IA fructueux s’ils sont uniquement poussés par la direction ou proviennent de mouvements populaires internes », explique Morel. Il souligne également le besoin de « développer des capacités à l’interne afin que l’organisation soit prête pour la prochaine vague, et continue de progresser ».
Les collaborations avec des fournisseurs technologiques et de services infonuagiques à très grande échelle aident les organisations à composer avec les complexités des systèmes IA extensibles.
« Un autre conseil pour 2025 est de mettre de l’ordre dans votre organisation : il faut prévoir une gouvernance, une structure et une stratégie de l’IA pour planifier comment mettre le tout en œuvre », ajoute Gutiw.
Miskawi précise : « Ce que vous devriez faire en tant que leaders est de vous concentrer sur deux principes clés. L’un est de générer une valeur plus rapidement pour l’environnement, nos clients et les consommateurs. L’autre est de fournir une valeur plus alignée, plus personnalisée à ces clients. »
- 1. Introductions
-
Fred Miskawi
Welcome to CGI's From AI to ROI podcast, where we discuss how AI drives change across enterprises and government organizations. We explore today's challenges and help you plan for trusted outcomes. I'm your host, Frederic Miskawi. I lead our AI innovation expert services at CGI, and I'm part of our global AI enablement team. And today I'm joined by Nick Morrel from Google.
Joining Nick and I is my good friend, Dr. Diane Gutiw from CGI. Nick, can you tell our audience a little bit more about yourself and your role at Google?
Nick Morel
Very nice to be with you both, today. Always nice to catch up on these types of conversations. My role is to help organizations understand how to navigate our generative AI offerings and how they start with a business outcome or a challenge you're trying to overcome, and how do we work back from that problem into technology. So, I kind of pride myself in being somewhat of a geek translator around taking business outcomes and then bringing it back into the technical world.
Fred Miskawi
Thank you, Nick. Diane, could you please share a little bit about yourself and your role at CGI?
Diane Gutiw
Thanks, Fred. I'm really excited to have this conversation. So, my role at CGI is leading our research center where we've been doing everything from developing our foundation at CGI for responsible use of AI for developers and tools. I also have been advising a lot of clients on AI policy and a lot of the foundation that we need to make sure that we're able to innovate within guardrails.
Prior to that, I was very deep in the traditional AI machine learning world, which I'm happy to see rise up a little bit again, now that generative AI is becoming more familiar for a lot of people. My background is healthcare. I've got a PhD in expert systems, and it's been a really exciting couple of years.
Fred Miskawi
Thank you, Diane. As for me, I've been involved with AI since the 1990s -- from technical architect to user experience expert to metro leader. My role in our global AI enablement team is to bring AI strategy and apply to AI best practices. I do that with some of our largest accounts worldwide. And I drive also the AI-led development acceleration program.
So, let's cover some of the key questions. We've got three key aspects of this podcast. The first phase will be to talk about the year in review, so what's happened until now in 2024. We'll then provide some advice, maybe talk about advice for 2025. And, what will be the value that enterprises should be expecting in 2025 -- that'll be the third segment.
- 2. 2024 in review
-
Fred Miskawi
So, starting with a year in review, Nick, when we're reflecting on this year, what's been the most unexpected AI development? And why do you think it may have caught some of the industry experts off guard?
Nick Morel
That's a very interesting question. So basically what I think is interesting is that when we started off the year, I think the pace of change of all of these different models and how rapidly more and more capabilities are being baked into these models and what the definition of state of the art kept meaning every single time a new announcement came out.
I think the ability to explore multiple modalities was a big unlock for a lot of organizations in 2024. We kind of went into 2024 with a lot of these kind of text-based chatbot type applications for a lot of organizations. And I think that as we were kind of rolling through 2024, a lot of emphasis on how these different modalities, whether it be imagery, whether it be audio, whether it be video, could be brought into the fold.
So, I would say that's probably one of the biggest leaps in 2024. I would also say that we kind of jokingly called it internally the end of the pilot-palooza. So, we had a lot of organizations in 2023 that were dabbling in generative AI and trying to figure out basically where they could -- potentially get the POCs that were driving the results they were looking for. But (they) tended to spread themselves across various areas of the business and not necessarily look at what ROI was going to come from it, but rather more around “could we do it”? And it became quite apparent throughout 2024 that the “can we do it” is generally “yes.”
This is generally not a technology problem. This is usually more a people and process problem. So as organizations were trying to get these into production. I think when you went with that POC mentality rather than a pilot mentality, it was very difficult for them to scale.
Fred Miskawi
Yeah, thank you, Nick. And the keyword being ROI, it's easy to deploy an AI-based solution. The question is, should you? And is the business case, the ROI in place? And Diane, what about you? What's been the most surprising aspect of 2024? And what has potentially been catching some of our industry experts off guard?
Diane Gutiw
Yeah, it's a great question. And I think I echo a lot of what Nick just said. You know, coming into 2024, there was a lot of organizations that just were not prepared for the rapid pace of the technology evolving and the level of access across all of the people in their organizations, personally, and the high demand for getting their hands on these tools.
I think the biggest surprise for me in 2024 would be some of the different types of media and tools that are evolving so quickly. Being able to generate a podcast from a document and listen to interrogation of that document to be able to get value, the ability to create images on the fly and the quality that's increasing so rapidly really surprised me.
Going into 2024, we were seeing some gaps in reliability and robustness that were causing organizations to hesitate a bit on “when will I be ready to use this” for decision making or documents or code development and all of industry. How do we solve the gaps and come up with a configurable solution that's scalable for businesses so they're comfortable moving into production? I would say that's the biggest surprise: how quick that moved.
What about you Fred? Where are you finding surprises in the last year?
Fred Miskawi
I think speed of evolution, the amount of functionality that we're getting has been surprising. I use the following metaphor quite a bit. It's like getting this alien baby that has landed in our hands and we're all trying to figure out how do we best apply that within the enterprise? And all of sudden, one day you realize that that baby can fly and all of a sudden that baby can shoot lasers out of their eyes, reference to Superman, but it's a lot of what I've been seeing in software development acceleration, that's especially true. We went from “this looks like this could be the killer app for generative AI” to “this is by far the biggest ROI in the enterprise today.” The ability to use different tools like Google Code Assist and others, and get some pretty sizable performance productivity improvements, has been surprising to me.
Nick Morel
There was another aspect that I'd be curious to get your take on this as well. I found that when we were in 2023 into 2024, we also had a lot of conversations around models. It was always about models and which models was kind of high on the benchmarks, kind of which one was state of the art. And as we rolled through the year, I think the importance of the platform became more apparent because models, change so frequently and there's always a newer or greater model.
And a lot of organizations were kind of stalling, thinking, “what if I develop this in the next model, can do it natively?” And I think that the reliance on the platform really was a big shift for us at Google, because organizations wanted the confidence to be able to go to production, but to know that things like model interchangeability down the road was going to be something we're solving for. Things like dynamic routing capabilities around, “hey, do I need to use a very high-grade, high-power model to answer these kind of chit-chat style questions or can I deflect that to a simpler model?” So, a lot of these kind of platform considerations are that element that allows companies to scalably go into production and leave the test bench.
I would say to anybody out there, leverage your partnerships. You work with hyperscalers, you work with these model providers, and their partners to be able to have these conversations so that they can try to avoid things that might be avoidable.
Fred Miskawi
Yeah, absolutely. And access to expertise, right? The ability to get advice on particular approaches, avoiding the rapid iterative approach that sometimes we do when we try to figure out particular key pieces of technology that are being released.
But Diane, question for you. So, when we look at these particular specific factors that have made successful AI deployments, what would you tell the audience in terms of those key success factors for a good AI enterprise deployment?
Diane Gutiw
I think the key is knowing what it is you're developing or using the tool to do in shifting that focus from “what do I use this tool for” to “how do I make this tool useful starting with a problem.”
Don't go in thinking I'm going to use this generative AI model to be able to do something. Start with, I have a problem, a high priority specific problem that I need to solve, then look at what information do I need and what's the right tool. A lot of the time we're seeing, you know, the problem is a data integration problem. It's not a machine learning generative AI problem. And then you can scale to adding more tools, doing more over time. So, I think that's truly the key to getting a return on investment is understanding what your focus is, and then continue to measure what's the value you got.
If my focus is to reduce the amount of time for my contact center to be able to resolve issues, did you achieve that? And if not, what do you need to adjust so that you do achieve it? I think that's really critical, rather than just taking a tool and trying different things, which was 2023 and going into 2024. And Nick, what you were saying about models.
What we're starting to see now about fit-for-purpose models -- not all large language models, but some of these scalable small language models that are very focused on a specific industry or a specific use case. I think we're going to see a lot more of that coming out because that helps solve what is it I'm actually trying to do, and what's the tool I need to be able to resolve that.
- 3. Recommendations for leaders
-
Fred Miskawi
And speaking of advice for 2025 and KPIs, Diane, we spent quite a bit of time together to measure the value that these tools are bringing in the ecosystem. What would you recommend leaders in the various industries that we work with, to measure as you're introducing these AI-based solutions?
Diane Gutiw
Well, the first thing that I would recommend, and it's coming out of our Voice of Our Clients, which is CGI's (research) investigation across leaders globally in different industries, is that 79 % of organizations are experimenting with generative AI, but only 26 % are actually moving into production. The key to getting a return on investment is focusing on something with the intention of going into production, with the intention of getting that actual benefit. That's going to be the key.
Other advice for 2025 is get your house in order: AI governance, having a structure and AI strategy for how you're going to implement this. How you're going to scale is going to be critical because of these productivity tools. And I know you and I, Fred, have been doing a lot of analysis in measuring what is the benefit to help advise how we're going to invest in the technology.
It really is important that you have a strategy for how you're going to scale this out, that you're looking at where am I going to get the best value. There's an expectation that they're going to get the hands on the tool. I truly believe it's inevitable. But coming up with a strategic way of moving that forward to make sure it's successful is really key. It's change management.
Nick, you said it in the beginning. It's not a technology problem. It's a policy and a process problem.
My last point on 2025, I think the floodgates are going to open. A lot of organizations have been trying to get their policies in place so that they can begin to adopt and move these things into production. 2025 is going to be the year that you see that start happening and all of the innovation that people are creating in using these tools. We're going to see more of that.
Fred Miskawi
I agree, Diane. And Nick, when you talk to customers that may be a little bit hesitant about AI adoption, what kind of advice do you give them to help them kind of go through that initial step?
Nick Morel
Typically, what we do is when we're working with organizations, there's some key factors of success that help one organization outperform their peers when it comes to AI. And I think that what we typically start with is the idea that this requires some kind of unified C-suite support around these projects, because you can't have successful AI programs that are driven purely top-down or purely from grassroots movements internally.
You need to meet in the middle. So we tell organizations that leaders and CEOs need to lean into this and try to understand how this technology applies to their business and contributes to meaningful results for the organization and to start investing in finding the partner that's going to help them have these questions and build an initial roadmap, help them go through guided ideation sessions and understand the implications of these projects.
To Diane's point, we used to call these ML systems because the word systems was everything that was around it. It's not just your forecasting model. It's how are you integrated into your ERP? How are you going to push it to the people that make the decisions? How do your finance or procurement people get access to this information and trust the data?
So, it's a journey, but I think that organizations need to invest in and understand that the journey is iterative and starts off with some quick wins, investing in some transformational projects. Working and building that muscle internally around how do they better the organization to be ready for the next wave and then kind of just continue to iterate through that but keep themselves honest too. Are we getting the results that we expected? Are we following up on the ROI projections that came from our line of business when they said they were going to adopt this solution? And we're reprioritizing based on business value that might be shifting over time or technical feasibility that might be getting a whole lot easier as these solutions get more sophisticated.
So, I think just having a good practice internally around how do you navigate this from your executive team, directors and the team and get the right people involved so that you can set yourself up for success.
- 4. What's next
-
Fred Miskawi
So, as we're going into 2025, we're looking at the future now. Diane, what do you feel will be the next major breakthrough when we talk about AI within the enterprise?
Diane Gutiw
You know, I think the key in 2025, we're going to inevitably see more small language models, fit for purpose models that are solving common use cases. I think we're going to see higher quality and the reliability of the responses, agentic AI where you're using, I know I've called them little worker bees, not babies with laser eyes, but we're using these little worker bees and focusing them on a task. Having one agent be the gatherer, another be the validator is really what we're going to start seeing in these models with a single source of access for the users.
What we've been doing for 40 years in capturing data and analytics and AI is trying to bring the answers to questions, the insights closer to business, closer to the end user and I think with generative AI we've really achieved that and that's where you're seeing the evolution.
The biggest thing I think we're going to see is more human AI interaction, collaboration and assistance. AI is not in my near future that I can see going to replace anyone. It's people that know how to work with AI, which understand how they can develop code, develop documents and work with AI to help accelerate and improve the quality of what they're doing. that's my forecast for 2025.
-
Fred Miskawi
Thank you, Diane. Nick, where do you think AI will deliver the highest ROI as we're looking into 2025?
Nick Morel
I think it really depends on the organization and how its costs are structured or how it generates revenue. But what we're seeing organizations do is really focus in on the areas that matter. So for example, if you have a very scaled part of your business that deals with customer engagement. We're seeing organizations going further down the conversation around rather than just talking about kind of traditional contact center platforms and stuff like that, but really looking at the whole kind of omni-channel customer experience as a whole and looking at where can I reimagine my customer experience?
We often apply our kind of Google, we call it 10X thinking. And we jokingly tell customers, well, if we all sit together and try to make something 10 % better, well, we're going to make a few optimizations here and there. We might merge two teams together and might say, hey, okay, well, we're on a path to 10 % improvement.
But if we ask ourselves, how do we make this customer experience 10 times better? All three of us are going to say, well, that entails replatforming. Those conversations are the ones that allow you to scale over time and to be able to really tap into those areas that generate revenue.
And I think to Diane's point, generative AI is a much lower barrier to entry for some of these things. So I think there's going to be increased focus on these areas that drive meaningful impact to the business rather than to Mark Twain's point to someone with a hammer, everything looks like a nail, and taking AI and trying to apply it everywhere like Montreal steak spice and thinking it's going to make everything taste better.
Diane Gutiw
Fred, going back full circle to you, what's going to surprise you in 2025 and what do you predict is coming next?
Fred Miskawi
So, as we're looking into this coming year, I do believe you're going to see an acceleration in the value in different areas of the enterprise.
What we should do as leaders is to focus on two key things. One is delivering value quicker to the environment, to our customers, our end consumers, and two, providing better aligned, more customized value to those consumers. By focusing on the timeframes, on the value, we'll justify a backlog of work that will reduce the impact to jobs.
Software development is a good example of that. There's a lot of work to be done. As you were mentioning, Nick, there's entire workflows and value that companies are delivering that can be rethought, improved and customized for the particular markets that they're targeting.
And as part of that process, you've got to embrace these tools. You've got to understand how they work. You've got to understand how best to leverage them without atrophying your own skills, your own capabilities within the enterprise. And that transition, I think, could be difficult.
But I think the future is bright. What we're looking at in terms of agentic AI is going to be a pretty massive amount of value right there in 2025. What makes them different is the ability to work in far more dynamic type of environments. Until now, the applications that we were building and developing and deploying, we had to think about every case and scenario of what could happen in production and then code accordingly.
These models enable us to act a lot more dynamically in changing data conditions within the production environment and take action and learn from it, and we're all learning how to apply that. We're all learning how to use solutions on the market today to get this multi-agent ecosystem going so that we can deliver that value that end consumers are going to be expecting. But for me, I'm focusing on speed of quality delivery and the ability to deliver more value to the end consumers.
Nick Morel
If I may add to that, I think that also for companies like CGI and your customers, think hyperscalers and technology companies are going to make that more and more easy for those conversations to occur. I think that we were in that kind of period where we were just like new technology was being released to market, brought to market.
And to use an analogy is like we were all building a bunch of Lego pieces and basically customers would come to us with a problem, and we would put together all of these pieces of Legos and say, well, here's how we're going to build it today. Three months from now, the recipe has changed. So, packaged solutions that are more systems that are solving a complete problem from one end to another, and less all of these kinds of APIs that you need to build in DIY together.
How can I see these solutions and how can I apply them to my problems with the right partners?
Fred Miskawi
Thank you, Nick. So, my deepest thanks to our guests, Nick Morrel from Google and Dr. Diane Gutiw from CGI.
Today's conversation emphasized, I think, the role that AI has in driving practical and impactful changes across enterprises and government organizations. To our listeners, I want to thank you for tuning in to the From AI to ROI podcast.
Remember that the key to success with AI is not just leveraging its capabilities but doing so with purpose and integrity. Until next time, continue exploring how AI can create value and drive results in your organization. If you enjoyed this episode, please check us out on cgi.com/AI. We'll see you in the next episode.
