- il y a 2 jours
AI Gets to Work: Enterprise Transformation Through Automation
Catégorie
🤖
TechnologieTranscription
00:00C'est un grand plaisir d'être avec vous aujourd'hui pour faciliter cette discussion
00:08sur l'entreprise transformation through automation.
00:13Je m'appelle Stéphane Booth, je suis un senior partner à McKinsey,
00:15leading Quantum Black AI à McKinsey en France
00:18et co-leading avec un collègue à l'agentic service line à Quantum Black globally.
00:24Aujourd'hui, j'ai l'honneur d'hôpire d'avoir deux premiers tech leaders,
00:30Michael Park, SVP, global head of AI at ServiceNow,
00:36et Octave Klaba, founder et chairman of OVH Cloud.
00:40Bienvenue.
00:41Merci.
00:43Business process automation n'est pas un nouveau topic.
00:46Dans les dernières 10 dernières années,
00:47many companies have deployed RPA technology to automate some processes
00:53with the first very positive result.
00:56But we were facing some challenges
00:58when we tried to automate complex business workflows
01:04with a high level of variability, exceptions, and so on.
01:09This is changing.
01:10With the surge of Gen AI, and more recently with Agentic AI,
01:16we have now the opportunity to automate complex business workflows,
01:21leveraging the specific strength of agents,
01:24such as real-time adaptability, task parallel execution,
01:29scalability and elasticity, or even resilience under uncertainty.
01:36We are going to investigate this groundbreaking paradigm now.
01:42Let's maybe start by the perspective of OVH Cloud.
01:47So, active in cloud computing,
01:50this question of automation is in fact a two-fold question for you.
01:54On one hand, you have the opportunity to automate your internal processes,
01:59but you are also providing a state-of-the-art tech stack,
02:05allowing your client to deploy automation at scale.
02:08Could you share your perspective on the trajectory of automation at OVH,
02:14and in particular, how you address those two aspects?
02:19Thank you.
02:20Very happy to be here to talk about AI.
02:23Yeah, you're right that when we have AI,
02:26we're thinking about this automation at scale,
02:29but think about cloud.
02:30Cloud is already, you have the control planes inside of cloud
02:35that automate everything.
02:38And what we've done that last 25 years in OVH Cloud,
02:41that we, instead of doing the things for the customers,
02:45we develop the control planes that do the things for the customers,
02:49data planes.
02:50Now, the time is to build the platforms
02:53that will develop the products,
02:56the control planes,
02:57that will deliver the value for our customers.
02:59So we are building this additional layer of the software,
03:06not to develop the product directly,
03:08not hiring people to develop the new product,
03:10but hiring people to develop the platforms,
03:13AI platforms,
03:14that will understand what we want to build,
03:17and it will help us to build,
03:18not just completing,
03:19but really developing the product.
03:21So this is the step that we are, right now,
03:24we initiated.
03:25We will hire additional people.
03:28We already hired a lot of people,
03:29but we will hire additional 150 people
03:32to really to accelerate in the understanding
03:35how to build the models,
03:36not LLM just for LLM,
03:38but how to build the models
03:39that can make the reasoning,
03:42how to make the decisions,
03:44how to make understanding,
03:45and to having the knowledge,
03:47and to, based on the knowledge,
03:50how we can deploy the things.
03:52Thank you.
03:53Michael, ServiceNow platform
03:56is evolving at a breakneck speed.
04:00Could you share your perspective,
04:03your vision on automation,
04:05and on how ServiceNow is helping companies
04:08to address this powerful lever?
04:11Yeah, for us,
04:12AI is just a tool in the journey we're on.
04:15ServiceNow, for the last 20 years,
04:17has actually been automating
04:18service management workflows,
04:19and so the mantra of put AI to work for people
04:23is what we're here to do,
04:24and what AI is enabling us to do,
04:26first with generative AI,
04:28is to simply augment the process
04:30of service management
04:31so that a requester of a service
04:34or a fulfiller of a service
04:35can be helped
04:36with the augmentation of gen AI skills.
04:39That can be conversational search
04:41for self-service,
04:42or on the fulfillment side,
04:45it can be augmenting things like case summarization
04:49or knowledge-based generation,
04:50remedial tasks that people working in service tests
04:53generally don't like to do
04:54to make their jobs easier.
04:56The advent, though, of agentic is interesting
04:59because you can start to actually use the LLM
05:02to use a stochastic method
05:04to understand the problem and solve it.
05:07So I don't think we're ever going to be in a world yet,
05:09just yet,
05:10where an enterprise company is going to say,
05:12I'm going to rely on the LLM
05:14to make all stochastic decisions
05:16on how my work operates,
05:17but to use agentic AI
05:19in a subroutine task
05:20of how a deterministic workflow
05:22that goes from A to B to C to D
05:24can be further provisioned with autonomous
05:27to go from A to D,
05:2950% faster,
05:3130% more efficiently,
05:32and with the end users
05:34and the people that are using the AI more happy,
05:37that's tangible value.
05:39Now, there's a lot of challenges
05:40to deploying that in the enterprise,
05:42but we're definitely heading in that direction quickly.
05:45Thank you.
05:47Octave,
05:47I know that OVH has recently recruited
05:51more than 50 AI experts.
05:55Could you elaborate a bit on the use cases
05:58on which you have been working,
06:00what kind of impact has been generated,
06:05what are the main challenges you are facing?
06:09Yeah, for example,
06:10we use a service now in OVH,
06:13so one of the topics that we work in right now
06:16is how we can help
06:17to help our customers using AI,
06:23using all this data that we have,
06:26but also changing totally the user experience.
06:28we think that the first connection right now
06:31has to be chatbot,
06:33not the call,
06:34not the message, etc.,
06:36but having really the chatbot,
06:38and then because of this chatting,
06:40you have the content,
06:41context with which customer can describe what he has,
06:47when he can make the call,
06:48when he can carry the ticket,
06:50when we can have the helping
06:52from what kind of the knowledge
06:53we have to push him.
06:55One or another step
06:57will be not just giving the information,
06:59but doing the things
07:00instead of giving the information
07:03how to do the things for the customers.
07:05Okay, so this will be another part
07:07on the support.
07:09Another thing that we are trying to automate
07:11is really how we can create
07:12the web pages just from the screenshots.
07:16Okay, you have the screenshot of the web page,
07:18so let's do that in Drupal,
07:20just using the standard APIs.
07:23Okay, so this is the kind of thing.
07:25Another is,
07:26you know,
07:26we manage 500,000 servers,
07:28we manage a few millions of VMs.
07:31How we can automate the AI agents
07:35that will be connected
07:36to all these servers,
07:38all these VMs,
07:39all these Kubernetes,
07:40and then we manage that
07:42instead of having the people
07:44that will connect and fix the things.
07:47Okay, so there was a lot of areas
07:48where we, internally,
07:51we need AI to accelerate our growth.
07:53But what is beauty on our job
07:55is that once we have these tools for us,
07:57we can offer these tools for our customers.
08:00And to help them
08:02to, on this algorithm,
08:05all these databases,
08:06we can take the specific knowledges from them
08:09and to build for them
08:10the specific models.
08:13This is where we're going
08:14in the next 18 months.
08:18Thank you.
08:19Michael, in a typical company,
08:21we have a high number of business processes
08:24with probably a different level
08:26of automation potential.
08:28Based on your experience,
08:30you know,
08:31from a platform provider perspective,
08:33how companies should identify
08:36and prioritize
08:37the best candidate process
08:40for automation,
08:41taking into account
08:42the value creation potential
08:44on one end,
08:45the complexity on the other end.
08:47What kind of KPIs
08:49they could use
08:51to assess
08:53and monitor
08:54the value creation potential?
08:56Yeah, I think right now
08:58we're in between acts.
09:00We're just starting to see
09:02the enterprise
09:02really learn
09:03and understand
09:04how to use
09:04generative AI techniques
09:06to complement
09:06the way work gets done.
09:08And we're in the first stages
09:10of the autonomous AI
09:11where we're seeing
09:12some early stages
09:13of autonomous AI agents
09:14take form.
09:16So the way that I would describe
09:18how ServiceNow
09:19is seeing that
09:20is actually
09:21in the construct
09:21of service management.
09:23If you think about
09:24a basic workflow,
09:25regardless of whether
09:26it's for an employee in IT
09:28or they're trying
09:29to get something done
09:29in HR
09:30or maybe it's
09:31an expense report
09:32or something in finance
09:33or procurement
09:34or it's a call center
09:35where a customer
09:36is coming in
09:37with a request
09:37or a partner needs something,
09:39there's a fulfillment request
09:41and a request.
09:42So the first thing
09:43that we're seeing
09:44is actually
09:45the measurement
09:45of deflection rate
09:47or the rate of self-service.
09:48And we're seeing
09:49tangible evidence
09:50that the use of
09:51generative AI
09:52and also now
09:53some early stages
09:54use cases of autonomous
09:55is actually substantially
09:57increasing the deflection rate
09:58or the rate of self-service.
10:01And the reason
10:01that's important
10:02is the end user,
10:03whoever that is,
10:04whether an employee
10:05or they're a customer,
10:06that deflection rate
10:07is satisfying them faster
10:09because they're getting
10:10the problem solved.
10:11And it's also
10:12reducing the cost
10:14to serve on that need
10:16because there's not
10:16a human involved
10:17in the loop.
10:18So deflection rate
10:19is the first measure
10:20and if you're not
10:21measuring that today,
10:22you should be
10:23because what generative AI
10:24and autonomous will do
10:25is continue to
10:27exponentially
10:28accelerate the rate
10:29of deflection
10:30on things
10:30that we need
10:31to go get done.
10:32The second effort
10:33is if it cannot be solved
10:35for self-service,
10:36it needs to be fulfilled.
10:38And generally
10:39in the fulfillment process,
10:40whether it's
10:41for an employer
10:41or a customer,
10:42there's a level one service
10:43and there's a level two.
10:45So with the use
10:46of generative AI technology
10:48in level one,
10:49you can actually
10:50apply that skill
10:51so that the mean time
10:52to resolution
10:53of a person
10:53that's fulfilling
10:54level one service
10:55can be dramatically
10:56improved.
10:57And so through the use
10:59of generative AI techniques,
11:00we're seeing on average
11:01customer MTTRs
11:03improve by 50%.
11:06So the amount of time
11:07they're trying to spend
11:08resolving an issue
11:09is going down by 50%
11:11because the AI
11:13is actually doing
11:13the research
11:14to present a summary
11:15of the problem.
11:16The AI is actually
11:18writing the resolution
11:20for the agent.
11:21The AI is actually
11:22writing the knowledge-based note
11:24so that people
11:25don't have to spend
11:26time doing that.
11:28So that helps.
11:29But what agentic AI
11:30is now,
11:31what we're starting
11:31to see is,
11:32with agentic AI,
11:33you can actually
11:34use it to replace
11:36level one services.
11:38And so in the future,
11:40what we think
11:40is going to happen
11:40is that level one service
11:42itself will go to zero
11:43or close to zero.
11:45And level two
11:47is where the real value
11:49add is going to be
11:49for the service personnel
11:50because they can,
11:51when it hits level two,
11:53it's super important,
11:54the human being
11:54is in the loop
11:55and they've got
11:56the generative AI
11:56capability to assist
11:57them to solve it better.
11:59So the whole work
12:00in process cycle time
12:01of actually how
12:03a service request
12:04is done
12:04through the application
12:06of generative AI
12:07and then the evolution
12:08of autonomous AI agents
12:10is going to completely
12:11transform that world.
12:13And what's more
12:14exciting about that
12:15is once the generative AI
12:16and the autonomous
12:17solve the problem,
12:18the pattern of the resolution
12:20can be moved into
12:22proactive maintenance
12:23in operations
12:24so that when the pattern
12:27is actually seen,
12:28the AI will actually
12:30execute the command
12:32of that problem
12:33before the end user
12:34even knows it
12:35so they never actually
12:36have to have
12:37the self-service
12:38experience again.
12:39And so in the next
12:40three to five years,
12:41we're going to see
12:41a material evolution
12:43in service management
12:44in this way
12:44where the system itself
12:46will become an autonomous
12:47learning enterprise
12:48for how to proactively
12:50manage service.
12:52Thank you.
12:54The value creation
12:55potential of enterprise
12:57transformation
12:58through automation
12:59goes far beyond
13:01the productivity gates.
13:03Through automation,
13:04we can turbocharge
13:05operation agility.
13:07We can also boost
13:09company revenues.
13:11But to capture
13:13this potential,
13:14it requires more
13:15than just plugging agents
13:18into existing
13:19business workflows.
13:21It calls for reinventing
13:23business processes
13:24around agents,
13:26leveraging the specific
13:27strength of agents.
13:30Octave,
13:31automation is a
13:32Cine,
13:33quite an condition
13:35for competitiveness.
13:37What has been
13:38the approach
13:39of OVH Cloud
13:40to help your client
13:41achieve this potential?
13:43which products
13:44are you providing?
13:46Why have you
13:47developed them?
13:49Yeah,
13:49so,
13:50yeah,
13:51AI
13:51help us
13:53to the productivity,
13:54but also,
13:55you know,
13:56on our side,
13:57we see that
13:57you have one part
13:59of the AI
14:00and productivity
14:01that's coming
14:01from the,
14:02let's say,
14:03the people
14:04in the offices,
14:05so support,
14:06developers,
14:08and the people
14:09that managing
14:09the data
14:10with the computer,
14:11but there's
14:12totally full
14:13new area
14:14with robots,
14:16with drones,
14:17with the physical world.
14:18And this is where
14:20there's a lot of areas
14:21that they have not
14:22explored yet
14:23because it's
14:24totally new
14:25that you can manage,
14:27for example,
14:27a human lead
14:29that is walking
14:30somewhere
14:31and to creating
14:32the value
14:33for the customers.
14:34So,
14:34also on our side
14:35is how we can
14:36build a product
14:37that will help
14:39our customers,
14:40developers,
14:40to develop
14:41this kind
14:42of the product
14:43that has to work,
14:44for example,
14:45in the factory,
14:46in the data warehouse,
14:49on the booths,
14:50in those,
14:51so we are thinking
14:52about all this area
14:53of the product
14:54that it's,
14:54of course,
14:55we have them
14:56and we develop them,
14:57et cetera,
14:58but how to be ready
14:59for the next revolution
15:00is physical world.
15:01And this is why
15:02we developed
15:02somewhere racks,
15:04specific cloud
15:05that we can deploy
15:06somewhere.
15:07We are working
15:07on the 5G,
15:08private 5G,
15:09we are working
15:10with the different
15:11companies
15:11that they offer
15:12the drones,
15:14the physical robots
15:17that they can move
15:18and to see
15:19how we can create
15:20the product
15:20that is easy
15:21to implement,
15:23that it's not just
15:24in the offices
15:25that happens,
15:26but also in the factories.
15:28And for us,
15:28it's really the next revolution
15:30that will totally change
15:32the way
15:32of building the processes,
15:34physical processes,
15:36moving the boxes,
15:37moving the logistics,
15:38and there's a lot
15:40of areas
15:41that they are not
15:41explored.
15:42For us,
15:43the question is
15:43how we can help
15:44that and how
15:45we can help
15:46create a very easy
15:47product that they
15:48can then take
15:50specifically in all ways
15:51and help customers
15:53moving in the right
15:54direction.
15:55Thank you.
15:57Michael,
15:57I was saying
15:58at the beginning
15:58of the session
15:59that the ServiceNow
16:00platform is evolving
16:01at a breakneck pace.
16:03could you share a bit
16:04what are the next
16:06important strategic
16:07milestones in the
16:09product roadmap
16:10and in particular
16:10how you are going
16:11to embed the new
16:14opportunities offered
16:15by Agentic AI
16:16for your clients?
16:17Yeah,
16:18this is moving
16:19so fast.
16:20If we look
16:21to six months ago,
16:22we were all
16:23in Gen AI world
16:24and Agentic AI
16:25was just taking form.
16:26We're essentially
16:27doubling the AI
16:28capability
16:28to the ServiceNow
16:29platform every 90 days.
16:30and so we've got
16:31to have the instance
16:33be able to be
16:33upgradable because
16:34we don't backport
16:35the technology.
16:36But the roadmap
16:37is being reflected
16:38based on what we're
16:39seeing in the
16:40customer demand signal
16:41and right now
16:42one of the,
16:42there's two key areas
16:43right now we're
16:44investing in
16:44that you're going
16:45to see in the
16:45roadmap coming
16:46very shortly here,
16:47actually three.
16:48But one is
16:49governance, okay?
16:50There's an issue
16:51with AI governance
16:52right now because
16:53this stuff is moving
16:54so fast.
16:55The policies and
16:56procedures for how
16:57to govern, regulate,
16:59drive security,
17:00drive compliance
17:00is all over the map.
17:02So one of the things
17:03that we actually launched
17:04is something called
17:05AI Control Tower
17:06and what AI Control Tower
17:07does, it creates a new
17:09class of AI in the data
17:10model of ServiceNow
17:11and it allows you
17:12to actually assign
17:13a set of APIs
17:15or an LLM
17:16or AI capabilities
17:17into that asset class.
17:19Once it's in the
17:19asset class of ServiceNow,
17:21the AI Control Tower
17:22presents a policy
17:24and workflow management
17:25tool that operates
17:27against those assets
17:28to be able to define
17:29very specifically
17:30risk, security,
17:33compliance, and performance
17:34attributes against that
17:36so that it's constantly
17:37reporting on the
17:38performance of those
17:39AI assets.
17:40They don't have to be
17:40ServiceNow AI assets.
17:42And what that does
17:43is it presents a
17:44control tower pane of glass
17:46that allows your CISO
17:47and your legal teams
17:49to actually enter
17:50into the ServiceNow
17:51environment and audit
17:52the assets and how
17:54they're performing.
17:54So we think this is
17:56going to be very important
17:57for anybody that's using
17:58AI because it's going
17:59to help you secure
18:00in the enterprise
18:01how to manage things
18:02and secure things
18:04with compliance.
18:05The other area that
18:06we're all seeing
18:07evolve right now
18:08is very early days
18:09with model context
18:10protocol and
18:12Google's A-to-A
18:13protocol,
18:14the ability to move
18:16agent to agent.
18:17And so this,
18:18we believe,
18:19is going to be a new
18:20way that integrations
18:21actually happen
18:21across workflow,
18:22where as the
18:23agentification of
18:24different applications
18:25occurs,
18:26the necessity to be
18:27able to quickly move
18:29multi-agent integrations
18:31across platforms
18:32to actually administer
18:33work in a more easy,
18:35a faster,
18:36more measurable way
18:37is going to occur.
18:38So we're spending
18:39a lot of time on
18:40something that we call
18:41the AI agent fabric,
18:42which will allow
18:43for multiple agents
18:45from different platforms
18:46to interoperate
18:48with ServiceNow
18:48so that you can
18:50actually transcend
18:51and leverage inference
18:52from many different
18:53places into unifying
18:55a workflow experience.
18:57Imagine a world
18:58where instead of being
18:59just inside ERP,
19:01CRM, HR,
19:02you can actually use
19:03the agents from that
19:04world, integrate it
19:05into ServiceNow,
19:06and it can actually
19:07render in one unified
19:09experience to an
19:10employee where all
19:12of those systems
19:12render as one
19:13simple experience,
19:14but all of the power
19:15of those systems
19:16are called upon
19:17through their agents.
19:18So this, we think,
19:19is also going to be
19:20a very important aspect
19:21of business transformation,
19:23and to what you were
19:24saying, for all of you
19:25in the room,
19:26the most important
19:27thing to think about
19:28is do not take
19:29the existing processes
19:31that you know
19:32that you've been
19:32operating for the last
19:3315, 20 years
19:34and just lift and shift
19:35it onto an AI platform.
19:37You have an opportunity
19:39to truly rethink
19:40how can you use
19:42Agentec to completely
19:43redefine the game,
19:45and this is going
19:46to take courage,
19:47it's going to take
19:47a lot of planning,
19:48it's going to take
19:49a lot of creativity,
19:50but this is the change
19:51in front of us,
19:52and so you have
19:53to see the change
19:53to be the change
19:54in this world of AI.
19:56Very exciting
19:57and promising.
19:58We are reaching
19:59the end of this
20:00fascinating discussion
20:02before closing
20:03the discussion.
20:05As a conclusion,
20:06maybe, Octave,
20:07could you share
20:08with us
20:09your priorities
20:10over the coming months
20:11to further transform
20:12OVH capabilities
20:14with AI
20:15and also help
20:16your client
20:17to accelerate
20:18AI deployment
20:19and maybe also
20:20if you have
20:20some advice
20:21for other companies
20:22who are embarking
20:23on the same journey?
20:25You know,
20:26OVH is a part
20:27of the ecosystem.
20:28We are here
20:29to leverage
20:29the ecosystem
20:30to help
20:31that this ecosystem
20:32reach the customers
20:33as the European
20:35Cloud Platform.
20:36This is our goal,
20:38this is what we are doing.
20:39So, now aside,
20:41also, we need
20:41to add additional products
20:43like AI.
20:44We already,
20:45of course,
20:45we have open-sources AI.
20:47We started to work
20:48on our chat
20:49that we can develop,
20:51we can offer that
20:52for the customers
20:53so we can use that.
20:54There was mobile application
20:56that will come very soon
20:57but also,
20:58how we can bring
20:59all this value
21:00and go into the
21:02customer's data centers.
21:04We can go at edge
21:05because with the
21:07sovereignty market,
21:09with the data sovereignty,
21:10technology sovereignty,
21:11you have also
21:12operational sovereignty.
21:13so the customers
21:14have to be able
21:16to say,
21:16I want my data,
21:18it's compute here.
21:20So, we are also
21:21working that
21:22and AI is one part
21:23of the products
21:24that we are offering.
21:26This is cloud,
21:27there was more
21:28than 40 products.
21:29One of them
21:29is AI
21:30and now our goal
21:31is really to offer that
21:32wherever the customer
21:34wants.
21:36Clearly,
21:37robust cloud foundation
21:39is a critical prerequisite
21:41to deploy AI at scale.
21:43and thanks a lot to Octave.
21:44Michael,
21:45maybe any additional advice?
21:47My only one piece
21:48of advice
21:49for all of you
21:49is don't be enamored
21:51by the fizzle
21:52and the sexiness
21:53of AI.
21:54Begin with the end
21:55in mind.
21:56So many projects
21:57we're saying
21:57are failing
21:58because people
21:59just want to kick
21:59the tires.
22:00You have to have
22:01the business outcome
22:02defined
22:02and then you have
22:03to learn how
22:04to employ AI
22:04to deliver the outcome.
22:05that is what will
22:07create success
22:07with AI.
22:09Right.
22:10Thanks a lot
22:11for your attention.
22:12Thank you.
22:14Thank you.
22:15Thank you.
22:19Thank you.
22:21Thank you.
Commentaires