- il y a 2 jours
Addressing the Gen AI Paradox
Catégorie
🤖
TechnologieTranscription
00:01I'm Aleksandr Sokorewski, and I'm leading Quantum Black.
00:05That is 5,000 people in 50 countries that apply AI as you speak.
00:12So you better stay and at least get the wisdom of this group.
00:16Now, before we go and discuss the future,
00:21what we will try to do in the next 20 minutes
00:24is to answer one of your most difficult questions.
00:28How come that something that sounds like a magic
00:33has very limited to do with your lives and with your enterprise at this stage?
00:39Okay?
00:40Now, before we go there, we probably should go a bit into the history
00:46because it always helps to understand the future.
00:51Let's go back into May 1997.
00:56Any ideas what happened in May 1997?
01:01To give you a hint, 11 of May 1997.
01:06It was the first time in human history
01:10that machine beat reigning world chess champion
01:16according to the standard rules.
01:20Now, this is a well-known fact, right?
01:23So all of you know Deep Blue beat Gary Kasparov.
01:26I think, however, the learning of this moment
01:30and the implications are not well-known.
01:34And if you have an opportunity to talk to Gary
01:37or actually to listen to him,
01:39he will tell you the following as somebody who was there at ground zero.
01:44The reason that machine won is very, very simple.
01:50It just makes less mistakes.
01:54It doesn't matter if you try to play fast or slow.
01:56It still makes less mistakes.
02:00And I think the implication is
02:02within any closed environment, if you wish,
02:06and game is a closed environment,
02:08for the last 30 years, machine beats human beings.
02:12Number one.
02:14Number two, Gary calls himself
02:17the first ever knowledge worker
02:20that was displaced by a machine.
02:24And you would expect him to say that he was fired,
02:28but he actually says that he was promoted.
02:32And the reason he believes he was promoted
02:35is because it allowed him to pose,
02:39to step back,
02:40and to think what really we as human beings
02:45are standing for.
02:47What does it mean in terms of developing our skills?
02:51What does it mean in terms of developing our critical thinking?
02:57What does it mean in terms of applying our will?
03:01And then you could continue from there.
03:03Now, we all got promoted 30 years ago, almost,
03:09and trying it hard.
03:11And if you look at our research and our deployments,
03:1588% of enterprises
03:18are claiming to do some sort of digital and AI transformation.
03:23I think the others are doing,
03:25they simply didn't participate in the survey.
03:27So you kind of see everybody's trying to do sort of AI,
03:31gen AI, you name your favorite transformation.
03:34Now, I think the interesting thing
03:36that only 5%,
03:395% are able to transform at least one domain.
03:44So think of marketing, operations, R&D.
03:49And now, help me understand,
03:51how do you really reconcile it
03:53with what Gary says?
03:54It's a very closed environment.
03:56Most of these processes, right?
03:58They kind of, you define as close as it is.
04:00Machine is better than us,
04:01but we cannot see any bottom line impact.
04:05Now, it's getting more interesting.
04:07Just follow me.
04:07If you go into our applications,
04:11and let's assume that any professional
04:12has a raw productivity of one.
04:17If I were to give you some AI tools,
04:19I'm going to give you a 20% boost.
04:22If I were to give you some of the agentic tools
04:25and you could deploy a few processes,
04:27you're going to get, let's say, around 2x.
04:31But if one human being is able to operate
04:37digital factories, the factories of agents,
04:39thinking about 20, 15 agents
04:42that each one of us is operating,
04:44it's got 20x.
04:46Now, here's the deal.
04:48It wins 20x, but no bottom line impact.
04:52So who is getting the benefits?
04:55And you know what?
04:56Our pets are extremely happy
04:59because all of a sudden we have more time
05:01spending time with them outside.
05:04And also the coffee machines are busier
05:06because in 20% of the time,
05:08machines are helping us to do stuff
05:10and we're able to drink coffee.
05:12Now, on the more serious note,
05:15we try to go after this paradox.
05:17How come it is so broad
05:20and so discussed
05:22and we are trying really so hard
05:24to implement AI
05:25but there are no bottom line impacts.
05:28And I think the paradox comes
05:30from the tale of two worlds, essentially.
05:33So we all see enterprise-wise deployment.
05:37So think about you write your email
05:39or you create a picture
05:41or you summarize the meeting.
05:43Now, it's all great.
05:44We're all doing it.
05:45But actually, the impact is quite fragmented,
05:49diffusive, and hard to track.
05:51At the same time,
05:53if we really go into the very deep vertical,
05:58end-to-end processes,
05:59so think about R&D,
06:01think about store operations,
06:03you choose your favorite.
06:05In most of them,
06:07it's still extremely fragmented
06:09and we didn't fall true
06:10what would it mean for the future.
06:13So, and therefore,
06:14there are no bottom line impacts here.
06:17Now, the reason why it's happening,
06:19if you go one level deeper,
06:21you realize that first and foremost,
06:23let's be fair,
06:23technology was not great
06:25and had a lot of issues.
06:27The governance,
06:29I don't need to tell you.
06:30It's all fragmented.
06:31We all try to do our own thing.
06:33Not necessarily trying to have
06:35the top-down view reinvention.
06:37Not always works.
06:39And then again,
06:40human beings are difficult to change
06:42because we're all happy campers
06:44with what we are.
06:46And for every dollar of technology,
06:48we're not necessarily always put in
06:50three to five dollar
06:51on changing human beings
06:52because that's much more expensive.
06:55Now, the era of agents arrives.
06:59And hallelujah,
07:00we think it's going to solve the technology.
07:03And indeed,
07:04to a certain degree,
07:05it solves and it looks like magic
07:07and it works like magic.
07:08But you know what?
07:09It creates more questions
07:10around technology.
07:12Be it around architecture
07:13to allow the data to flow.
07:16Be it around orchestration
07:17because agents are not living in silos.
07:20They would like to get all the data
07:21from different systems
07:22in your organization.
07:24If we think about the governance,
07:27just the sheer fact
07:28of you not thinking about
07:31taking an old
07:32and a chronistic process
07:33and embedding AI there,
07:35but really stepping back
07:37and rethinking it
07:38is difficult.
07:39It requires creativity.
07:41Even think about it.
07:43Such a simple thing
07:44like interface with AI.
07:45I'm not typing to you here.
07:49We are talking, right?
07:50It's not developed.
07:52You use websites.
07:54They tend to be completely
07:55anachronistic
07:56because who said
07:57that I need to follow
07:59a designer journey
08:01in selecting my goods?
08:03Maybe I have a different way
08:04of selecting goods.
08:05Maybe I have a different way
08:06of thinking.
08:06and this is a very simple example
08:09of what reinvention might mean.
08:11And then last but not least,
08:13the change management
08:14is difficult.
08:17It's so difficult
08:19to manage human beings.
08:21Think about what happens
08:23when you get a team
08:24where you have human beings
08:25and virtual members
08:27working together.
08:28So think about the chaos
08:30that could be created.
08:31So how do you govern
08:32this organization
08:33where you have sometimes
08:35more virtual members
08:36than your physical members?
08:38And again,
08:39coming back to the questions,
08:40what is our role?
08:41What is the role
08:43of human being?
08:45Now, so far,
08:47it sounds scary
08:48and kind of philosophical.
08:49It's great.
08:50But we decided
08:51to do a move.
08:53And based on hundreds
08:55of different transformations
08:57that we run,
08:58we are very proud
09:00to launch at Vivatec
09:03a new product suite
09:05Agents at Scale.
09:09Now,
09:11the idea of Agents at Scale
09:13is essentially
09:14to take 100 years
09:15and we're going to turn 100
09:18as Mothership McKinsey & Company
09:19next year
09:20of experience
09:22and put it into software.
09:25So you immediately
09:26get a depository
09:27of across all industries,
09:29all functions
09:30of newly refought processes.
09:33And you not only get
09:35the process
09:35that is kind of fascinating,
09:37but you're actually
09:38going to get Agents
09:39that power these processes.
09:41So day one,
09:43you could rewire
09:44your processes
09:45and launch the Agents.
09:47Now,
09:48we also understand
09:49that if we all
09:50operate the same,
09:52it's very hard
09:53to compete.
09:53It's very hard
09:54to understand
09:55what's my secret sauce.
09:57And therefore,
09:57what we're presenting
09:59is also a proprietary
10:00developer tools
10:02to actually create
10:03your own Agents
10:04based on the secret DNA
10:06of your organizations
10:07with your special data
10:09in IS.
10:12beyond this,
10:13it's also coming
10:15with an architecture blueprint
10:16in order for you
10:17not just to implement it today,
10:19but also be able
10:20to scale your system
10:21in the future.
10:23And then,
10:25if you think
10:26about different systems,
10:27you actually need
10:28independence
10:29from looking
10:29in different vendors.
10:31You actually need
10:31all the data
10:32that might sit
10:33in very different silos
10:34in your organization
10:35brought together.
10:37and therefore,
10:38we also introduce
10:38an orchestration layer
10:40that allows different agents
10:41to collaborate
10:42with each other
10:42so they could
10:43create new tasks,
10:45they could solve
10:45new problems.
10:46You could really get
10:47the richness
10:47of this world.
10:49And then,
10:50last but not least,
10:51again,
10:51coming back to humans,
10:52we are difficult to change.
10:53We require change management,
10:55we require upskilling,
10:56we're all complacent
10:57with what we are,
10:58so we come with
10:59our reward recipe,
11:01how do you really
11:02change human beings?
11:04So,
11:05ladies and gentlemen,
11:06agents at scale
11:07are currently online
11:09and available for you
11:10to transform
11:11your organizations.
11:13Now,
11:13beyond the marketing pitch,
11:15I would like to hand
11:16over to Stéphane,
11:17who will take you
11:18through a few exact cases
11:20that have been done
11:21recently,
11:22here and now,
11:23to show you the power
11:24and how the real magic
11:25looks like.
11:26Stéphane,
11:27over to you.
11:28Thank you.
11:29Good morning.
11:32Good morning,
11:33Stéphane Booth.
11:34I am a senior partner
11:35leading Quantum Black
11:36AI by McKinsey
11:37in France.
11:38Very happy to be
11:39with you this morning.
11:40As previously mentioned
11:42by Alex,
11:43the value creation potential
11:44of agentic AI
11:46goes far beyond
11:47productivity gains.
11:49Agents can turbocharge
11:51operations,
11:52they can boost revenue.
11:54Capturing this potential
11:56requires, however,
11:57more than simply
11:59inserting agents
12:00into existing
12:02business workflows.
12:03It calls for
12:04business process
12:05reinvention
12:06around agents.
12:08Redesigning processes
12:10from the ground up,
12:11exploiting agent-specific
12:13strengths,
12:14such as
12:15real-time adaptability,
12:18task parallel execution,
12:20elasticity,
12:21or resilience
12:22under uncertainty.
12:24This is not
12:25science fiction.
12:27Quantum Black
12:28has helped
12:28leveraging
12:30the agent-at-scale
12:31platform
12:32to transform
12:33business processes
12:35for several
12:36forward-looking
12:37companies.
12:39As a first
12:41real-life
12:42example,
12:44let's take
12:44the case
12:45of a digital
12:46factory
12:46composed of
12:47more than
12:48100 agents
12:50created for
12:51a bank.
12:51The challenge
12:53here was
12:54to modernize
12:54a highly
12:55complex
12:56legacy
12:57application
12:57composed of
12:58more than
12:59400 pieces
13:00of software,
13:02several millions
13:02of lines
13:03of code.
13:04The project
13:05was initially
13:06estimated
13:06to more
13:07than 600
13:07millions
13:08of euros.
13:10We have
13:10built
13:11a digital
13:12factory
13:12composed
13:13of more
13:13than 100
13:14agents
13:15grouped
13:16into
13:16squads
13:17supervised
13:18by human
13:19beings
13:20with a
13:21ratio
13:21of 1
13:22to 20,
13:22one human
13:23being
13:24for 20
13:24agents,
13:25and thanks
13:26to that,
13:26we have
13:26streamlined
13:27the modernization
13:28effort by
13:29more than
13:3050%
13:31and reduced
13:32the duration
13:33of the
13:33modernization
13:34by 30%.
13:36But,
13:37of course,
13:38a system
13:39of agents
13:40are not
13:40limited
13:40to software
13:42engineering.
13:43As a second
13:44real-life
13:44example,
13:45let's take
13:46the case
13:46of the
13:47credit
13:47risk
13:48writing
13:49process
13:50for a
13:52credit
13:52risk
13:52memo
13:53writing
13:53process
13:53for a
13:54bank.
13:55It's a
13:55very complex
13:56process
13:56involving
13:57a dozen
13:58of different
13:59information
14:00sources,
14:01taking
14:01typically
14:02several
14:02weeks.
14:03We have
14:04here developed
14:04a system
14:05of agents
14:06leveraging
14:07the agent
14:07at scale
14:08platform
14:08to fully
14:10re-engineer
14:10this process.
14:11The result
14:12is also
14:12impressive.
14:13the effort
14:14has been
14:14reduced
14:15by 30%
14:16and the
14:17overall
14:17duration
14:17of the
14:18process
14:18by 20%.
14:19But,
14:20agents
14:21are also
14:23very powerful
14:24to boost
14:25revenues.
14:27In that
14:27case,
14:28for a grocery
14:29retailer,
14:30agents
14:30embedded
14:31in their
14:32e-commerce
14:33platform
14:34are leveraging
14:35real-time
14:36data
14:36to surface
14:37cross-sell
14:38and up-sell
14:38offers,
14:39generating
14:40more than
14:4110%
14:42basket
14:43value
14:44increase.
14:45So,
14:45as you
14:45can see,
14:47agentic
14:48AI is
14:49no more
14:50science
14:50fiction.
14:51We have
14:51now a
14:51concrete
14:51example
14:52of
14:52companies
14:53leveraging
14:54that
14:54opportunity.
14:56Thank you,
14:57Stéphane.
14:58And,
14:59we started
15:00with the
15:00history.
15:01We probably
15:01will also
15:02try to
15:02close with
15:03the
15:03history.
15:03But,
15:04let's
15:04go to
15:04the
15:04point.
15:05The
15:05reason
15:05we'll
15:06launch
15:06it,
15:06it's
15:07not just
15:07to help
15:08you to
15:08unlock
15:08economic
15:09value,
15:09but it
15:10actually
15:10gives you
15:11more time
15:12to think
15:13who we
15:13are and
15:14who you
15:14are as
15:15human
15:15beings.
15:16And,
15:17what we
15:17would try
15:17to do
15:18by this
15:18exercise
15:19is really
15:19to try
15:20to reverse
15:20the formula
15:21defined by
15:22Edison.
15:23That is
15:241% of
15:24inspiration
15:25and 99%
15:27of perspiration,
15:28meaning sweat
15:29and hard
15:30work.
15:30So,
15:31by introducing
15:32an agent
15:32at scale,
15:33we hope
15:33to allow
15:34you more
15:34time to
15:36focus on
15:37your
15:37inspiration,
15:38on your
15:39will,
15:39and on
15:40many
15:40other
15:40things
15:41that
15:41just
15:41human
15:41beings
15:42can
15:42do
15:42while
15:43living
15:43some
15:44of
15:44the
15:44more
15:44procedural
15:45things
15:45to
15:46the
15:46machine.
15:47And,
15:47therefore,
15:47we invite
15:48you on
15:48the amazing
15:49journey
15:49of
15:50agent at
15:51scale.
15:51You've
15:51got a
15:52white
15:52paper here.
15:53You have
15:53our
15:53context
15:54to really
15:55experiment
15:55with it,
15:56to develop
15:57it.
15:57We're going
15:57to open
15:58source part
15:59of it
15:59and to
15:59join the
16:00journey.
16:00Very
16:01much
16:01looking
16:01forward
16:02to this
16:02journey
16:02and for
16:03your
16:03feedback.
16:03Thank
16:04you
16:04very
16:04much.
16:05Thank
16:06you.
Commentaires