- il y a 1 jour
Smart Gains: How Are AI Agents Unlocking Productivity?
Catégorie
🤖
TechnologieTranscription
00:00Good afternoon, everyone. Glad to be here today with you today on this panel on
00:05Agent TKI and how it enables new levels of productivity. My name is Nicolas
00:11Mechler. I'm a senior partner at McKinsey and I lead our customer experience
00:17practice globally. Today, to explore this new frontier, I am delighted to be joined
00:26with four distinguished panel experts today. Christophe Vermont, who is the
00:35Chief Technology and Transformation and Technology Officer at AXA in France, who
00:41is at the forefront of driving digital transformation in the insurance
00:45industry. Vincent Colgrave, Head of AI, Strategy and Transformation at Science
00:51Co. A seven billion chemical leader, Vincent enables every employee to be
00:56explorers of growth by driving transformation and rethinking key
01:02customer journeys powered by Gen AI. Christian Buckner, Senior Vice President
01:10Altair, a leader in providing software and cloud solution for simulation, IoT, high
01:16performance computing, data analytics and artificial intelligence. And finally,
01:21Morali, Swaminathan, Chief Technology Officer at Freshworks, renowned for
01:28building intuitive AI powered software that makes IT and customer support teams
01:32more efficient and effective. Maybe before deep diving into the conversation, let me
01:42take a moment to set the stage a little bit on addressing some of the key
01:47questions around agent AI. In a way, making sure, you know, you probably have been there in
01:54the last days and heard about it a lot. These are new autonomous systems that help
01:59drive changes and act to achieve specific objectives using AI and in particular Gen AI to
02:09interact in complex environments. Thanks to the novelties of Gen AI enabled by Gen AI, these have really
02:20powered new horizons for companies to drive significant impact. Personally, one of my clients
02:28that was enabled to implement this at scale, so not a demo, not a pilot, but the real production
02:36environment, we managed to reduce through a conversational agent, a cost by almost 50%, 44% in
02:44that case, while at the same time boosting satisfaction, customer satisfaction by six
02:51points. So for the first time, you achieve actually both. However, in multiple McKinsey studies, we've seen a paradox. A
03:01lot of
03:01people are jumping on that using customer generative AI to transform, but very few see the impact and the productivity
03:12in the bottom line.
03:14So we'll explore this today. And indeed, let's start digging deeper into a little bit of how your industries are
03:22impacted.
03:22And maybe let me start with you, Christian. Christophe, sorry. How do you see the impact of Agent AI disrupt
03:29the insurance industry?
03:32Thank you for your question. So I don't know if this will disrupt, but definitely that will transform our business.
03:38Yes. Let me give you first very quick introduction about AXA. So we are a global leading insurance company. We
03:45are insuring 100 million of customers in every kind of risk they are facing, property and casualties, life and savings,
03:54health and protection with a clear mission to act for human progress by protecting what matters.
04:00And this mission is really driving our transformation around AI, being human centered. And I think in the agentic topic,
04:10that's an important point. So we were really early adopters of AI, of course, for many years, predictive AI. We
04:18entered the Gen AI world two years ago with Secure GPT, which was our platform used by all our employees
04:25across the globe.
04:26And now we are into scaling AI for our core business capabilities. Let me give you a few examples. Now
04:35we are capable to develop and roll out conversational agents to support our people in doing their daily jobs.
04:43For example, in France, we have an agent called Smart in AXA supporting our distribution channels, the tight agents. So
04:52the similarities are interesting. So that they can access, I mean, in a quicker, easier way to the information around
05:01our products, our coverages, the way we manage claims.
05:05So that they can really gain time in answering customer requests. This is something visible.
05:11Another example with our customers, we have an online coach, an agent as well, providing advices and breadth practices on
05:21how to reduce risks on the road. These are very concrete examples.
05:25Another area is how we provide simplification features to our people. For example, in our call centers and claims management
05:35team, we are now capable to analyze a customer request from voice or text to provide insight on what could
05:44be the next step and summarize conversation again so that our people can focus on expertise and customer satisfaction.
05:52And maybe the last one is on prevention. We are combining spatial data and AI to better prevent wildfires, risk
06:02of wildfires, and being more proactive in reacting and taking care of difficult situations in case of natural events. For
06:10example, floods, we used that in Valencia a few months ago to measure the water aids and react and act
06:18where this was the most important.
06:20Now we're entering the world of orchestrating all this, making sure that all those capabilities can work together by a
06:28system of agents so that we become, of course, more competitive, we sell better our products, and more importantly, we
06:37simplify and magnify the customer experience in a way.
06:40So let's be clear, insurance business, it's a matter of advice, of care, it's a matter of expertise, so we
06:48won't become a synthetic company. We won't become a fully agentic driven company. Customers will continue to request human interaction.
06:57Our experts will continue to assess risk, our distribution channel will continue to provide advice, but this is clear that
07:04in the very near future, AI agents will be capable to handle simple requests, to support our people even more
07:11in managing their jobs.
07:14So we are moving into this world. The next step is how we industrialize this, and we are setting things
07:23up in that perspective with this clear principle of being human centered. This is very important for us.
07:29Thank you, Christophe. Indeed, almost naturally, insurance is full of agents, right?
07:36True, tight agents, by the way.
07:38Without play on words.
07:41Human before all.
07:42Human agents first.
07:44Maybe we switch to Vincent.
07:46How is ScienceCo actually pushing the boundaries for this agent world in the, let's say, industrial world and chemical world
07:56that you know better?
07:58Thank you for your question.
08:00I think first I will do the same and just define very quickly what is ScienceCo.
08:05So first, our name, Science Company.
08:08I think it's an easy way to remember that our core business is about science.
08:13To give you a bit of data points, like one of the two electrical cars that is out there have
08:19our components.
08:2070% of everything that is flying in the sky, from small vitals to big airplanes, have technologies of ScienceCo.
08:28Or 50% of the hemodalyses that are done in the world, leveraging the technology in the membranes that we
08:36support our customers to develop.
08:37Just to give you that as an idea that the complexity we need to manage on a day-to-day
08:42basis is realities of value chains, of realities that our teams need to face and to handle.
08:49And this is a perfect ground for Gen.AI, for AI, for Agenting.AI to come into force.
08:55But to answer your question, I would say simply first, we are very humble.
09:00Because we were talking together before we started that session and I think we need to recognize that there is
09:06a lot that we all need to learn and to discover.
09:09So what we define at ScienceCo is what are the key core business as well elements capabilities that we wanted
09:15to augment and to really see how we could leverage the technology.
09:19And I will zoom on one. Our CEO calls it very simply our sales buddy.
09:24We have 2,000 people that are on the ground every day to meet our customers and to really understand,
09:30as it's about science, what are the properties, what are the problems, what are the unmet needs that needs to
09:37be solved.
09:38And this is where developing and having to go up to the very specific details of the issue on a
09:46specific application somewhere that can sit in Europe, in the US, in China or in Asia requires to digest quite
09:53a lot of data.
09:53And this is where we have been deploying and working very hands-on with the teams to ensure that we
09:59were able to deploy the first buddies, the first really agents being able to support our teams and able to
10:07hunt, able to be faster at the way they work with our customers and to be even more precise into
10:13the data that they're bringing to life and to the offer that the company gives.
10:17So that's just the first stage. Then we see that we could start to link that towards our science, the
10:23core development that we do in our labs and so on.
10:25And here we see as well opportunities, but we're just starting to scratch the surface there.
10:30And we know that it's going to be quite a lot of it's going to be a journey. So again,
10:35we are very humble, very excited.
10:38I think we learn every day we fail. I mean, it's a very active and dynamic world.
10:43But to conclude on the sales body, we identified already 200 millions of potential sales.
10:49And I'm saying that because this is something that is into our system.
10:53So when you look at that and you consider that six months ago, we had nothing and we're coming from
10:58an industry that started its life like 165 years old.
11:02I think, you know, it's a promising journey.
11:06Thank you. When, you know, thinking about implementing these agents, obviously the technology and the IT angle comes in the
11:17conversation.
11:18So we're wondering also what needs to be in place. What are the IT readiness elements that need to be
11:26there?
11:26And I'd like to turn a little bit to you, Morali. What are the most, you know, promising use case
11:31that you've developed?
11:32But also how do you address these IT implementation challenges?
11:38Hello, everybody. Nice to be here. It's a very energetic conference.
11:41And like my panelists said, we are scratching the surface of AI and agent tech platform.
11:49So Freshworks, to introduce yourself again, we are an enterprise grade alternative to Salesforce and Service Notes of the world,
11:58providing uncomplicated support for IT and customer service teams.
12:03And we do that through our AI-powered solutions. And our AI-powered solutions have become a lot more agent
12:12tech.
12:13What does agent tech mean? So we are trying to not just answer questions, but we're also trying to perform
12:19actions on behalf of our users.
12:23So there are multiple customers from different industries that use Freshworks products.
12:28We have about 73,000 customers around the world.
12:31I'm sure there are many in the room and many in France. We have over 2,500 customers in France
12:36using our products as well.
12:38So whether you are trying to answer a customer support ticket, or you're trying to book a travel booking, or
12:46you're trying to get a refund, or you're trying to get help from IT help desk, that's where Freshworks gets
12:53deployed.
12:53And in the agent tech AI world, we have an AI agent studio, which allows you to build no-code
13:01AI agents.
13:04That means anybody, our business users don't need to be technical.
13:10They can deploy these AI agents quickly in order to get the maximum ROI out of the AI.
13:18So we want to make it AI as easy to adopt as possible, and that's where we deployed.
13:24We have created a platform for our customers to use AI in their regular business process.
13:30So that way they can deflect support.
13:33The biggest benefit comes in use cases, when you have an AI agent deployed on customer service,
13:39it lets you deflect support tickets so that only the complex ones get to your support engineers.
13:48Similarly, if you are a human agent looking at tickets in the background,
13:53we have an AI co-pilot, which helps you assist with resolving those tickets.
14:00That means we'll provide you with helping respond faster, show you similar tickets,
14:06help you with coming up with resolution notes.
14:09So anything that makes you resolve tickets faster, reducing costs,
14:13and then improving the productivity for your agents.
14:17And then finally, we have the AI insights piece,
14:20which is primarily for service desk managers and IT directors to look at what's going wrong.
14:26If there are any anomalies, we proactively flag it so that they can go ahead and fix it
14:31before it becomes a big problem.
14:33So all of these, we have customers like Hobbycraft,
14:36which have deflected over 30% of their support tickets through deploying our AI agents.
14:44Which means that now they have a lot more time to work on complex tickets
14:48and improve customer satisfaction by at least 25%.
14:54Impressive deployments indeed across many, many situations.
14:58Maybe turning to you, Christian.
15:01So how does Altair address this agentic opportunity?
15:05And if we switch to the IT part of the question,
15:09how do you ensure the right security guardrails are in place for your data, your client's data?
15:17How do you address this?
15:18Sure. So it's a great question.
15:21At Altair, honestly, we build a comprehensive data and AI portfolio of products.
15:27And as we were looking at agents and we were looking at how agents are being rolled out at our
15:31customers,
15:32the funny thing is that we realized a lot of the same problems that have existed in data analytics projects
15:39for decades,
15:40we've been doing this for decades, have continued to exist in the world of agents,
15:44the same that they've existed in the past.
15:46They're still siloed data.
15:48They're still messy data.
15:49They're still enterprise existing legacy systems that need to be connected to that are 30 years old
15:54and communicate in formats that new agentic frameworks have no idea about.
16:00And so what we went forward in approaching our agentic platform
16:04and how we can provide tools for organizations that want to build agents
16:09is we started with these fundamental problems.
16:11We started with how do you connect to the enterprise systems that will power the agents the same way they
16:17power humans,
16:18that will power agents that actually understand the underlying data structures.
16:22And so the core first issues that we solved were in connecting to these legacy and established systems
16:28and getting the data out of them in a way that's actually meaningful for the rest of the business.
16:33And then we moved up from there in providing a semantic layer across those things
16:37that then can expose that data to agents in a way that those LLMs and those agents actually understand
16:44and humans understand in a way that people can understand using concepts,
16:48not rows and columns and obscure column names and all these attributes that machines understand
16:54but human agents don't understand.
16:56Now we've taken this complex data landscape, this messy data landscape,
17:00and described it in a way that the organization and the agents can actually work with it.
17:06And that fundamentally means that you can then build much more powerful automation on top of that underlying foundation.
17:15And I think to your point at the very beginning that a lot of organizations are in this very first
17:19stage
17:20where they're experimenting, they're going out, they're finding open source tools,
17:24they're implementing those open source tools maybe on a department by department basis.
17:28But those projects, I think Gardner said, have like a 60% chance of failing.
17:34And the reason is because they're isolated in that particular use case.
17:38They don't take into account the underlying complexity that's always existed in data analytics.
17:44And so we put together a stack we feel like helps you get from that fundamental messy silo data infrastructure
17:53that is as complex as 40 years of development can make it
17:58and then build this next generation of automation on top of it.
18:03And it's those two steps that are necessary to actually implement agentic automation at a broad scale
18:09and go beyond those initial POCs and projects.
18:13And so you asked about guardrails.
18:15Guardrails are a critical element of that and that fundamental data layer provides the access control
18:21but also the context, regulations, compliance for the organization
18:25into that semantic layer, that underlying data layer.
18:29And you asked about how you can control the development across these different organizations.
18:34You want specified tools for specific skill sets within that organization.
18:39You want low code tools for those who want to be building agents
18:43but don't understand Python and R and all those tools.
18:45And you want code tools for those that are data scientists and can focus on those things.
18:51But by having both in the same platform, you're enabling the entire organization
18:55to get the benefit of data, of agents and build in the way that makes sense for them
19:00and collaborate towards the future.
19:02So you can have somebody, an operator on the factory line or in the business line,
19:07build in low code a tool and then have a data scientist look at that
19:11and deploy it into production using the guardrails that exist in the centralized platform.
19:16And so for us, it was important to have each one of these layers
19:18so that the organization can go from these baseline POCs
19:22but build on a foundation that allows them to scale up to 100, 1,000, 10,000 agents in the
19:29organization.
19:30So clearly, a certain number of enablers are required, I think we all understood,
19:36to get to a level where agents can start to become effective
19:40and you can release them in a way in front of your employees or in front of your customers.
19:45Now, switching precisely to these, in many of our recent research,
19:50we've seen that one of the critical elements to put in place
19:55will be to make sure your employees or your customers are also prepared to what's coming.
20:02So in a way, how do you create this environment where people trust what's coming,
20:08you know, accept and adopt the systems?
20:13By the way, I'm talking independently employees or customers.
20:17And I would like to turn to you, Vincent, a little bit.
20:20What steps are you taking to ensure your teams internally and stakeholders
20:27actually, you know, aligned, engaged and moving along?
20:32Yeah, I think this is the fundamental question.
20:35And you talk about the trust of the employees, but it's a general topic about, you know,
20:40going beyond the hype, going beyond, like as well, some people are scared about the technology
20:45and a very simple question about, am I going to lose my job?
20:48How is it going to affect me?
20:50So on our side, we spend a lot of time sitting down talking.
20:54We were really proud of the fact that we had an agreement with our unions.
20:57And, you know, again, we have people on the shop floor and plants.
21:01We have people in the labs.
21:02We have people on the road.
21:03So you need to ensure that you have a common way of looking at it
21:07and agreeing on the points where you know you will disagree.
21:10And I'm not going to open too deep that door, but there is as well ethical, you know,
21:14questions that are raising about how do you handle the technology.
21:17So you need to ensure that you have the right space to have that dialogue.
21:22And at the same time, you know, we are businesses.
21:24So you need as well to develop and ensure that you have a right track record
21:29and that you're in capacity to really explain what's the value behind what you do.
21:33Because I think the most, the first issue you can face is, wow, it's nice.
21:38It's sexy.
21:39It works a bit to your points on one POC somewhere, but yes.
21:42Okay.
21:42What do you do?
21:43You cannot scale it or you are not able to extract like really core value for the business.
21:48So I think this as well, this set of governance and ensuring that you enable business value
21:53is something that is really at the core of the way we set up our team in Sansco
21:58is to ensure that we constantly push on that direction and that we, even if it's frustrating,
22:05even if you need to have a bit of tough calls, but you have those discussions and you ensure what
22:09you do,
22:09you do it to ensure that your company moves in the right direction.
22:12So it's a mix of all of that.
22:14That's why it's a journey.
22:15It's a fascinating one, full of emotions, of learnings and so on.
22:19But we do believe that's the only way we do.
22:22And I think, Christophe, you said it very well earlier.
22:24It's human first.
22:26And I think if you stick to that line, you should find the answers.
22:29Excellent.
22:30Actually, turning to Christophe, in our preparation, you spoke to me a lot about what AXA has been doing
22:39on moving along your employees on this journey and preparing your customers for it.
22:45Can you tell us a little bit about it?
22:47Yes, because I believe, of course, this is a technological transformation.
22:54I mean, we talked a lot about it.
22:56There is a lot to do from a technological standpoint.
23:00There is a lot to do in our infrastructure.
23:03For example, we started a quite ambitious cloud program many years ago in order to be ready.
23:09You mentioned the topic around data, having data quality, a good access to data,
23:15renovating our systems, our core systems is a significant effort.
23:21But my belief is that the most difficult effort is not there.
23:25It's indeed the way we manage the change.
23:29Technology is moving very fast and we have this even today showing how fast this is going.
23:38But this is clear and we should acknowledge that the time to absorb and the ability of our people to
23:46abort those changes is not so fast.
23:50And this is where the effort has to be and indeed at AXA France, but it is true around the
23:59group in AXA.
24:01We started a program two years ago called Nadia on new ambition for data and AI in AXA France with,
24:11I would say, a few pillars.
24:13First is the strategic intent.
24:17So AI, Gen AI.
24:19At GenTIC is coming also at the very heart of our strategic discussion at MC, XCOM level.
24:26And we put this as one of the three pillars of our strategic plan, the current one.
24:32And of course, the next one will be even more maybe.
24:36So there is a real need of very strong leadership.
24:40Second is how we find the right balance between acculturation and continuous improvement.
24:50And we indeed put this technology in the hands of our employees to learn and let them get accustomed to
24:57it.
24:57But on the other hand, I think it's very important to focus.
25:01And this is where you are saying we need to extract value as we are businesses.
25:06And to do so, my belief is that we need to choose the areas where really the change will be
25:13and the value will be.
25:14And we have a quite bold ambition on rethinking our processes from an end-to-end approach, rethinking the way
25:21jobs will change.
25:22And this is a dialogue we have to happen between IT guys, AI and agenting AI guys, technology guys,
25:33so that the change is really understood and we go in depth.
25:38This is, I think, very important.
25:40Third is indeed how we manage the cultural change.
25:44And Nadia is a lot about that.
25:47Explaining, communicating, training everyone in the company about what it means to have the data of quality,
25:55what it means to work with AI, what those possibilities are.
26:00We've set a complete set of training to do so from experts to, I would say, more expert users.
26:09We are animating a community of data experts.
26:15We have 2,000 data and AI experts reference in the group and we have a proper animation for that.
26:21So we are pushing this cultural change and this is an everyday, I would say, work we are doing.
26:28And fourth, and you mentioned it as well, is how we anticipate the impact on HR and skills.
26:35Of course, job will change for good.
26:38This is our belief because at the end of the day, those agents will handle tasks, maybe low value added
26:46and our people will have the ability to focus on expertise, take care of our customers, which is what our
26:53mission is fundamentally.
26:55So there is a real opportunity, but indeed there are fears, there are doubts that we need to deal with
27:01and we need to anticipate that.
27:03And we started the dialogue with unions very early as well to a point where we crystallized at the end
27:09of last year an agreement with our unions
27:12where very regularly, every three months, six months, we give full transparency of what we're doing and we are involving
27:20them directly in the project.
27:23They are the first users of some of our use cases, so that the change is properly managed and we
27:28have positivity in what we are trying to do.
27:32Thank you. So in a way, if we step back, we've discussed all the great forays you're doing in this
27:40new implementation of agents.
27:43We've discussed the IT challenges and how you're coping with it, the employee part and how you're dealing with some
27:52of your customers.
27:53Obviously, there's a missing part in this conversation, which is the economic equation.
27:58So in a way, how do we drive, how do you manage to drive either directly for your companies or
28:06for your clients an economic equation that has a positive ROI for it?
28:11And maybe, Christian, with Altair, how do you see that? How do you communicate that? How do you see your
28:18clients benefiting from your vast platforms of services that exist?
28:24Sure. So I think if you looked at the ROI that can be generated from agentic programs, you would probably
28:34see a spectrum from it was a complete waste of time
28:38and we spent a bunch of money messing around to really automating and creating impact in the organization and then
28:46producing that return on investment as a result.
28:49And I think if you looked at the one end of the spectrum where it was a complete waste of
28:54time and we spent a bunch of money and time doing it,
28:56you would find that organizations looked at this opportunity in a toy experimental way.
29:06They started to look at the open source options that are out there and they messed around more with the
29:12system integration and architecture of trying to spin together 15 different open source tools into one combined platform instead of
29:20focusing on the business processes that can actually produce real impact.
29:25And I think what we've seen is that those organizations, frankly, I'm a platform provider and so maybe I'm a
29:32little bit biased in this, but those that have gone to a platform have been able to forget about that
29:36system integration component and instead focus more on the how do we actually implement this in the business itself component.
29:47And I think that's the real important piece and it's all the things that you guys mentioned also at the
29:51very beginning.
29:52It's not just technology, but removing that technology element allows you to look at how to actually implement agents at
30:00scale in the organization, how to manage the change, how to reduce the fear that exists in the employees.
30:09And I think if you can do that, then you're able to pull the lever instead of just messing around
30:18with the system integration in the beginning.
30:21And at our organization in particular, what we found is that it's especially helpful to have what we call a
30:26center of excellence.
30:27And that center of excellence works with organizations to do an AI assessment and that AI assessment isn't about technology.
30:35It's about finding the lowest hanging fruit to go after and implement a really impactful project.
30:43And that first impactful project then becomes a way to get buy in on project two and project three and
30:51project four.
30:52And that rollout is what creates the massive ROI.
30:56The first one gets you in the door, not us, but gets the organization, the people who said, I want
31:01to start implementing agents, gets them in the door for actually implementing.
31:06And then adding that across the organization and putting it in a centralized way.
31:10And so those two combinations of things, implementing a platform so that you can have that foundation for building agents
31:16at scale and allowing organizations like us to deal with the complexity of every two weeks, there seems to be
31:22a new agentic framework out there that you have to figure out how to use.
31:26We'll be the ones to figure that out.
31:27Is it MCP?
31:28Is it A2A?
31:29What is it?
31:30We'll do that.
31:30We'll add those supports so that you don't have to figure that out.
31:34We'll be the ones who figure out the skill set of system integration.
31:38You'll be the ones that figure out the skill set of creating impact.
31:42And we can help with that second piece, but it really comes down to the business to manage the change,
31:47manage the people and put it into place.
31:50And so I think that's where we found the most success and the right end of the spectrum.
31:54So indeed, and just to compliment, McKinsey just published a paper, actually we're releasing it for VivaTech, that most of
32:04the usages were generic and horizontal so far of GNI.
32:08We're now switching to vertical cases that are going deep in the business based on this, you know, let's say
32:15access to data that has been now provided.
32:17But now it's the time to go deep on employee journeys or customer journeys that actually you can digitize with
32:25agents.
32:28Muraili, do you want to give a few elements about what Freshworks is doing on proving the ROI to your
32:35clients?
32:35Yes.
32:36I think adding on to everything our panelists said, that's true that you have to, we can't do it all
32:43on day one.
32:44You have to prove yourself in a smaller audience with different use cases before you go big, right?
32:51AI is all about trying it in increments iteratively and going big.
32:56So a lot of the work that we do is we are innovating while we are actually doing the work.
33:05So typically, so most of the AI efforts that we see that I think you talked about employee journeys or
33:12customer journeys.
33:13That's a great place to put AI in place where you pick a specific business problem.
33:21It could be a vertical specific problem. In the case, it could be a vertical retail or a travel.
33:25So you're trying to build, you're trying to provide deflection, you're trying to provide reduction in cost.
33:33You are trying to provide benefit of deploying an AI agent, which primarily comes in at scale, right?
33:41When you have AI agents, you can work with lean organization.
33:43You can support seasonality.
33:46That means you don't have to staff up agents in the back when there is peak loads.
33:51So those are the places where you get a lot of benefit.
33:54But the AI agents you deploy have to be vertically enabled.
33:59Otherwise, it doesn't solve the business problem.
34:01So that's how we operate as well.
34:03So we have built a set of verticalized agents for customers to use.
34:08And we let our customers build their own AI agents as well.
34:12So the ROI comes from high deflection, reduced costs, and also improving productivity of the people who are working on
34:21those solutions.
34:23So it's a cumulative effect.
34:26And it's also about showing the benefit to your organization and then building on that success and expanding it.
34:35So it's land and expand is a very common thing that happens in AI.
34:38So that's how we show ROI of our solutions.
34:42Deflection, productivity, and also augmentation of certain people.
34:48Yes.
34:48Christophe, you wanted to comment on AXA's ROI view.
34:52No, no, but very quickly.
34:54I think the ROI question is an important one.
34:59First, indeed, we need to be humble about that because we are at the very beginning.
35:04There are significant investments to be done.
35:07And thank you.
35:08If you are providing solutions, maybe we can go faster.
35:12But this is true.
35:13I mean, we need to unlock the data.
35:15There is a lot to be done.
35:17But there are low-hanging fruits, clearly, in supporting with, I mean, basic conversational agents first and basic data.
35:27But even there, it's tricky sometimes because you need to make sure everything is working well, that the accuracy is
35:35well.
35:35So there is a real, but there are low-hanging fruits.
35:38The fully orchestrated agentic world will require investments.
35:43And then the question is, where will be the value?
35:46My, in our opinion, is this, of course, efficiency.
35:49But this is also a tremendous opportunity to grow our business, to be better in doing our business.
35:58And this is by itself creating value, as well as customer.
36:01The way we, I mean, magnify, I was saying, the customer experience is also bringing value.
36:07So in many ways, I think those featured agentic features will bring value in the future.
36:14But indeed, I think the point is avoid mushrooming, focus on really where is the value, rethink the process.
36:21And you say that from an end-to-end perspective, so that we really extract the value from an activity
36:28that is, I mean, supported by AI.
36:30And being humble in grabbing the low-hanging fruits and going step by step.
36:35But we are taking the train, as we said in the preparation work.
36:40Maybe, Vincent, a few words on the ROI as well, on your side, from your perspective.
36:46I think quite a lot has been already said and covered.
36:50And that made me think about one element that is critical for us is with this journey transformation on our
36:58side, we try to apply the we care and we dare.
37:00And, Christophe, you just mentioned that it's accelerating.
37:05That's really the, I mean, that's really the feeling we have.
37:09And we have really this vision that on few elements that are really the core businesses one, we know that
37:15the ROI will be massive.
37:17Because if I just stick to science, if I stick to chemistry, there is a reason why today when you
37:22go and chat GPT, you cannot yet solve quite a lot of chemical problems.
37:27Because it's something that yet hasn't been cracked.
37:30So, here we, you know, we look at, I'm looking at the stage, I'm looking at everybody here in the
37:35audience.
37:35So, there is as well a huge opportunity, you know, for the ecosystem to really bring, like, shared value.
37:42And to really try to say, okay, we are going to go there together.
37:45So, this is where you start as well changing the collaboration with your customers, with your partners.
37:50Because you realize that the speed of that technology, those agents, they will not be really only looking at our
37:57company per se,
37:57but they will be interacting with other agents coming from other companies and so on.
38:02So, you start feeling, okay, how do I orchestrate that? How do I ensure that my guardrails, you know, connect
38:07to the other ones?
38:08So, you start to open, you know, more and more questions as you go.
38:11So, this is why, to be able to unlock that value, you need to be very clear about the direction
38:16you want to take.
38:17You need to be very clear about your values, the way you want to get there.
38:21And then you are going to unlock that based on what was shared here in the panel, based on all
38:26this step-by-step value that you unlock up to the point where it becomes like a normal business operation.
38:32But if you start tackling those problems that are way bigger than you, I think this is where you start
38:39rallying a lot of people internally and a lot of people outside.
38:42So, that's really where it triggers a lot of hope and a lot of inspiration for a lot of people
38:47around us.
38:50Thank you. So, thinking back on this conversation, I think, you know, by the way, the question of the panel
38:57was how to unlock the productivity.
38:59How is it and not what productivity? So, in a way, everybody sees the possibilities, even if you pushed a
39:07little bit the boundaries as well.
39:08People think about the cost productivity of, you know, your existing cost base.
39:13But I think, as Christophe said, there's also a revenue opportunity, which is large.
39:18And indeed, agents have a role to play either in direct sales enablement or in customer satisfaction, which at the
39:28end generates more retention and more satisfied customers with cross-selling.
39:34And even where, Vincent, you go to areas that are not cost, not revenues, but unseen areas, unseen, you know,
39:42not cracked yet problems that will probably, you know, unlock.
39:47And so, if we step back on this conversation, clearly, multiple pillars on which, you know, the work needs to
39:53happen.
39:54You know, obviously, a big IT part where I think, Christian, you said a lot about, you know, extracting the
40:04data from where it is a little bit, you know, hidden or blocked or inaccessible or difficult to understand.
40:12So that all the agents that you put on it can have the maximum value.
40:16At the same time, thinking about, you know, how all the protection that is required, the hallucination that could be
40:24possible.
40:25So the guardrails that you need to put from, you know, what the system could generate or the attack that
40:31could come from the outside, the governance around the IT as well is extremely important.
40:35Then we moved into the conversation on your employees, how to bring them along.
40:42It's a threatening move, right?
40:45Indeed, if you don't bring the employees, the culture, the mindset or the union simply along, they will not follow.
40:54They will think my job is threatened and, you know, why should I follow this?
40:57I'm going to resist.
40:59You need to have a story, but you need to bring them along and bring the capabilities up so that
41:03they are not afraid of it.
41:05I think the customers, we discussed a lot about how it could be better for them.
41:10And let's not be naive.
41:11Clearly, it will require investments.
41:14So somewhere there are CFOs in organizations that need to be satisfied and, you know, an external stakeholders that invest
41:24into your companies.
41:25You cannot invest without some views about the returns.
41:30So let's bring the CFO along as well and show the direction and where you believe you are going to
41:37get the value.
41:38So in summary, I think we are all thinking we were using the image in the preparation that Agent AI
41:48is a bit of a train leaving the station.
41:51And we are not completely clear yet, but it's clearly leaving the station now.
41:55So the point is jump on the train, make sure you have the right, you know, pillars in place and
42:01where you're going to go.
42:03But this is about, you know, jumping on the train now and making sure, you know, you build the capabilities,
42:11new capabilities, new type of employees and capabilities in the background that are required.
42:18You jump on the bandwagon and actually follow that revolution.
42:23Thank you very much.
Commentaires