Subscribe
Share
Share
Embed
What does it really take to scale AI across the enterprise? How do leaders move from ambition to impact? In Winning with AI, world-renowned AI pioneer Andrew Ng and Bain partner and digital re-invention expert Sarah Elk host candid conversations with global CEOs on the frontlines of AI transformation.
Together, they explore the opportunities and obstacles of deploying AI at scale: reimagining operating models, empowering teams, and leading culture change to prepare their organizations for the future.
Winning with AI helps leaders of global organizations navigate times of radical change with clarity and vision, uncovering the very human challenges – and opportunities – that transformation brings.
About the hosts:
Dr. Andrew Ng is an internationally acclaimed thought leader in artificial intelligence. He founded DeepLearning.AI, cofounded Coursera, and serves as an adjunct professor at Stanford University’s department of computer science. He was named one of the Time’s 100 Most Influential People in AI in 2023.
Sarah Elk is Bain & Company’s AI practice leader in the Americas, and a leading voice on digital reinvention. She partners with CEOs and senior executives to reimagine how their organizations operate – aligning strategy, operating models, and culture to turn AI into sustained competitive advantage. Sarah is the co-author of Doing Agile Right (Harvard Business Review Press), and her perspectives on leadership and transformation have been featured in the Harvard Business Review and Forbes.
Bain & Company:
Winning with AI is brought to you by Bain & Company, a global consultancy trusted by the world’s most influential business leaders. With decades of experience guiding organizations through growth, transformation, and leadership development, Bain’s executive insights offer what it takes to lead at scale.
(upbeat music)
- There is not a single
thing that we're doing
that doesn't impact growth
and revenue at the end of the day.
So everything we do, when
you think about productivity,
is about how we can save time,
how we can reinvest and grow the business.
- We are excited to speak
with Carsten Egeriis,
CEO of Danske Bank.
Carsten has an incredible
grasp of technology,
which has shaped his human-centric
agile leadership approach
and how he's accelerated
the AI transformation
at Danske Bank.
- As a technology leader,
he spent a lot of time to
understand the technology himself
to drive forward both the technology
and the people aspects
of his business forward.
I think that's really interesting.
- I'm Sarah Elk.
- I'm Andrew Ng.
- And this is "Winning with AI."
(upbeat music)
Carsten, thank you so much
for being with us today.
We're excited to have you.
- Thank you. Great to be with you.
- One of the things
that really impressed me
as I was learning about the
journey that you've been on
with AI at Danske Bank is
how early you got started.
As I understand it, in early 2024,
you launched a DanskeGPT,
which, by the end of 2024,
you had over 75% adoption internally,
which is incredibly fast.
And then in 2025,
you already had 12 generative
AI solutions rolled out
and being used by your employees.
What gave you the conviction
to lead so early on AI?
- Look, it's always hard to say,
of course, compared to
others, how fast you are,
but we've always been hugely focused
on technology in Danske Bank.
And, as you know,
the financial sector more
generally is a big investor
and user of technology.
- Yeah, well, you must
have figured something out
because I also read and was
impressed that, in April,
you actually made a commitment
to financial results
and putting your money where your mouth is
in terms of the progress
that you've made from an AI standpoint.
Can you talk about the
conviction to do that
and how you thought about being willing
to publicly put numbers on the table?
- Yeah, I mean, the way that
we really thought about it
is that, look, in many ways,
it's very difficult to try to
find a return on investment
on a lot of the technology
investments we're making.
So it wasn't about trying to be scientific
and finding the perfect
return on investment.
It was more about a mindset of,
"Okay, we're going to go to the board,
we're going to go to the market,
we're internally going to
invest a lot in technology
and more than we've ever invested before,"
to ensure that we get the
buy-in and the support
and also to ensure that the mindset
of really being disciplined,
diligent, following up,
testing, learning,
failing fast, moving on,
setting tangible targets
that we were all sort of committed to,
we thought was an important part
of this very significant
investment journey.
- Could you unpack your
investment strategy
a little bit further?
I saw the number you had
announced was 4.5 billion kroner,
about 700 million US dollars.
How do you end up at that number
rather than something
much higher, much lower?
And would you be willing to
unpack for us one layer deeper?
How do you allocate this
700 million US dollars?
How do you govern it?
What's your prioritization framework?
But how do you go about
deploying this capital?
- First of all, I should say
that we came from a place
probably like in '23,
where we're probably investing
roughly three billion a year
in strategic development.
So it's a very significant
increase in overall investment
for us on that scale, 50%
increase, if you will.
And that's nominal number.
And then, as you all know,
if you sort of productivity
adjust that number,
it's significantly more
than the 50% increase
because you can do so much
more with the technology,
you can do so much more
with the software developers
and so on and so forth.
How we thought about it was
that we tried to look at
what are the key areas
where we believe
that we can create meaningful
competitive advantage
or moat going forward.
And where we have very high conviction
is we need the foundational architecture.
We call that our AI City, where
we're sort of building out
and we're doing that in
partnership with AWS.
So those kind of foundational things,
those are the things that we
absolutely have to invest in
and we have to accelerate that very fast
because anything you build on top
in terms of various
different application layers,
et cetera, is going to
be dependent on that.
So that was sort of one,
or you could even say two
layers of the investment,
the architecture, our AI City,
and then the data layers.
Then we sort of thought about,
"Okay, on the front-end
user application side,
what are the most important areas
where we believe that we
can scale meaningfully?"
And those would be areas
such as customer assistance.
So customers contacting us via call center
or via the branches.
How do we do all that on
the mobile phone instead
and build out our financial
assistant on the mobile phone?
And we can come back to what
that is already showing us
because we're already seeing
customer adoption quite quickly.
Another example on customer journey
would be the credit process.
We gather a lot of financial information;
we do a lot of ratios; we
do a lot of calculations.
And only when that's done,
then you have somebody that actually sits
and actually provides
a human judgment view
on whether or not and
how much and so forth.
So we went through both
the architecture layer,
the data layer,
but then also the
important customer journeys
or employee journeys where we thought
that those journeys are large enough
and scalable enough to
warrant investments.
And then going back to
your question around,
"So how did it become 4.5 billion?"
Well, there was a bit of going bottoms-up,
looking at those areas
and how much we believe
we can meaningfully scale
the resource we have and invest.
And then, of course, also,
how much do we lean in?
Obviously, there has to
be some sort of balance
between how we invest, how we
pace ourselves, how we ensure
that the investments are
robust enough and pay back,
how much we can actually scale
the organization all at once.
And then, of course, also testing that out
with board and with market
expectations and so forth.
So it's, of course,
a constant triangulation
between those things.
- And in addition to the investment,
you had also announced
a two billion kroner,
about 300 million US dollar savings
or productivity gain target.
The examples you gave just now
to improve the customer experience
sound like they're not just
cost savings potentially
or productivity gains,
but also growth by serving
customers much better.
I'd love to hear how you think about that
in terms of this investment paying off
and at what timescale and to what amount,
to the extent you have a view already.
- So I've tried actually
to completely move away
in my narrative, both
internally and externally,
around that this is a cost game.
Of course, there's productivity,
but at the end of the day,
exactly to your point,
there is not a single
thing that we're doing
that doesn't impact growth and
revenue at the end of the day
because all of these opportunities,
the ultimate implication
is that you save time
for your colleagues to
spend more time with clients
or you make some meaningful
automation on processes
that allows you to free up
investment to invest in growth.
So everything we do, when
you think about productivity,
is about how we can save time,
how we can reinvest and grow the business.
But of course, when you make
commitments to the markets
around this, our view is
that it's much more credible
to explain that in cost terms,
at least now, early days.
So what exactly are the
automation benefits,
the productivity benefits in cost terms,
rather than trying to then translate
that into the revenue and growth piece,
which is always going to be
a little bit more tenuous.
But, of course, there
are very real examples,
like, for example, we've gone
over the last couple of years,
on average, from doing less
than five customer meetings
to almost eight customer meetings here
in the retail network in Denmark.
Just as one small
example, it's not only AI,
but a lot of that is things
like having assistance available
that frees up time,
things like having Copilot
that can actually take meeting minutes
and provide the meeting
minutes to the client
and not having to write them in hand.
And by the way,
it also provides a better
customer experience
because you're much more
present with a customer.
- Yeah, that's very interesting.
You make me wonder if
so much of the narrative
in the broader world around
AI is on cost savings
because that's how CEOs have to explain it
to Wall Street or to investors.
Even if at the back of your mind,
you actually are thinking growth
much more than cost savings,
but it's just harder to
explain to the market.
- Yeah, it's interesting.
There's several CEO
dialogues that I've had
where the cost pressure has
actually come from the board
and the CEO is very growth
and productivity-minded
in the same way that
you're describing, Carsten.
I guess how have you
navigated board alignment
on the goals that you've set
and come through that process?
- Our board is very supportive
in terms of our focus
on technology and the
transformation around augmenting
and accelerating our
strategy with technology.
But, of course, we also all agree
that in order to sort of stay in control
and be able to continue to invest,
you need to be able to
have the credibility
and the credibility comes
with constantly having proof
points on how deployment,
adoption, impact on customers works.
And therefore, it's not a
"this or that"; it's a "both."
But there's no question
that in terms of the energy
in the organization and the
excitement around it comes
from the ability to do
more with customers,
to grow, and of course
also to make work more fun.
It's not like productivity or
cost savings is a bad word.
If it frees up a lot of dull,
boring, back-and-forth rework,
which in fact is what the technology does,
it also gives a much better quality
of colleague and work experience.
- I love what you described
in the data and technology foundation
and sort of the test-and-learn process.
I would love to learn more
about what that looks
like sort of in practice.
One of the things that
we've seen is teams go
to do process redesign
and they design the future state process
without sufficient testing
and learning across,
"If I get to a reasoning
step that isn't working
as well as I want it to be
from an agentic standpoint,
should I go do the data transformation?
Should I think about human-in-the-loop
and the AI adoption piece on
the human side differently?
Should I put deterministic
code in front of the agents
and only send part of the use case
through to the reasoning step?"
And all of this triangulation
across AI adoption,
what's actually happening
in agentic reasoning
and the data underneath
that feeds all of that
is a sort of an art to come through
and problem solve to make it high-quality.
What has your experience been
as you've enabled your teams
to drive faster test and
learn in and around agentic?
- We've had mixed experiences.
So a really good experience
where we've been able
to scale is the example
of our financial assistant
in our mobile bank
where just over 30% of queries now come
through the financial assistant,
the digital assistant in the mobile bank.
And we have a resolution
rate of above 80% now, e.g.
So questions that are not
passed on to a human being.
And that's only happened
over eight, nine months.
So I think about like it
took us 25 years in Denmark
to go from 600 retail branches
to 50 retail branches.
It only took us nine months
to have a significant amount
of customer queries be answered
through a financial assistant.
So that's an example of a process
where I think it's very scalable
and it has been able to be
successful quite quickly.
And we continue, of course,
to augment that model so that
it can do more and more things
like setting up a customer meeting
and all sorts of other things.
An example where we have clearly
seen much more challenges
in our, what we call
exception-based services
in our operational areas
where we have hundreds
of different smaller,
more subscale processes.
It could be like a customer
calling in about a issue
with taking a lien on the
mortgage or you have a situation,
it could be divorce, it could
be death or other things,
where it's smaller
exception-based services
that help customers and clients
with those type of questions, queries.
And there we have also looked at,
"Okay, how can we automate
that more with AI,
with agents and agentic flows?"
And the technology can clearly do it.
So we don't have any question
that that is going to be possible,
but the data and the data links
required are just not there
yet in terms of having all the data,
all the linkages ready
to be able to support
a lot of smaller subscale processes.
It'll happen over time, I have no doubt,
but it's just a good example of a process
that you can fairly quickly
scale and which works well,
but also examples of where we
have not been able to scale
and where we need to sort
of rethink how we do this.
And again, going back to
where we probably need to make
more foundational investments
before we can continue.
- So you mentioned some
of the architecture
or the vendor choices you're making
for how you're building out
that data infrastructure.
Do you have any organizing principles
or ways of prioritizing or philosophies
for how you lay this foundation
for all the applications
you want to build on top of?
- Yeah, I mean, this
has been a key question
and continues to be a key question for us.
So I constantly get surprised
of new applications,
new models, new approaches
that come out where in fact,
I thought that there was
more competitive advantage
in building things or
developing things where then
with what comes out, you can see,
"Okay, it's open source; you
can take that open source,
you can pretty quickly put
that into your own harness,
you can pretty quickly
use the skills available
and all of a sudden you
can build a pitch book
or whatever it is."
It doesn't mean that the data piece
and the linkages still aren't important
before you can really scale it,
but it just means I think
a little bit differently
about where there was advantage
and where there isn't.
But going back, just like,
"How do we choose the
technology, the partners?"
The way we've chosen to do it
is that we've got two
very important partners
in AWS and then in Infosys
where we have a significant
amount of our engineers working.
Of course, we have many
other partners as well.
We're sort of in the context of, I guess,
global banks and midsize banks.
So we're Denmark's largest bank
and one of the largest banks
in Nordics, but in the
context of global banks,
we're a midsize bank and
we still have limited
and scarce resources in the
context of the biggest banks.
And we think very much around,
but we still have enough size
and scale to partner up
with really great technology partners
that can help give us a global view
of what they're seeing and
how they're thinking about it.
So for us, it's really important
to have close strategic
partnerships with these providers
and then together with them
to constantly test and learn.
And to your point, to
try different providers,
to try different models.
We have, of course, made
some fundamental choices back
to the choices around who
we're using as cloud provider,
who are using as engineering support,
and several other examples as well.
- As you build up this foundation,
I remember at Davos,
you said you think about
a 10, 15-year payback on
some of these investments,
but I'd love to hear how you think
about moat or lack of
moat in certain cases,
but how do you approach
building your own moats,
assuming that's something
you're focused on?
- It's really something we
are really talking a lot about
and really focused on.
Like I said, I think
on the technology side,
there is some level of moat
in building out the infrastructure
and building out the data products
and building out the interlinkages
and how you use that data.
I don't think that that
is that easily replicable.
And moat, I think today
with the pace of change
and the transformation going on,
is not something that's there.
I don't think about it as being there.
I think about it may even be there
for a couple months, e.g.
that you have a few months
or six months or 12 months of head start.
And in that time period,
you can free up more time
to do more customer meetings
at a better cost to serve,
or you can deliver a better technology
or service to clients,
which means that you can just get
that extra bit of market share.
The biggest but also
most intangible, I think,
competitive advantage is
the leadership and culture
around driving this
technology transformation.
And that goes back to how
can you actually drive?
We're 155-year-old bank;
we have 20,000 employees.
Some of them are AI-native,
but most of them aren't.
So what is the leadership
and cultural transformation required
to get the entire workforce
to adopt the technology?
And there, I think it's
many different things,
and maybe it sounds trivial,
but I really think it's
many different things.
It's things like constantly showing demos
of what the technology can
do that excites people.
It is obviously various
different types of training,
and there I don't think
traditional training is necessarily
that relevant because the
technology moves so fast.
So we have things like AI
champions and superusers
in our different parts of the business
that walk around with AI champion hoodies
and help various people in their teams
on showing them how the
technology can be used.
We do hackathons, we do GenAI week.
So my point is it's many,
many different things
to set the tone from the top,
to create the culture,
to create the excitement.
It's many, many different things,
and we're trying to do
many of those things.
But that, I think, is
the true differentiator,
is if you can get the whole organization
to adopt and deploy at pace.
Linked to that is, of course,
risk management more broadly,
compliance, risk, third-party
risk management, cyber,
all those things that
just take such a long time
in banking because
we're heavily regulated.
So those are some of the areas
where I think about competitive advantage,
and some of them are more lasting.
And as I said, some of them are actually
maybe only a very short period of time,
but can still make a difference.
- But ephemeral advantage
lets you gain market share.
That market share could be
quite durable still, right?
- Absolutely.
- What you said about people is great,
but do you have a view on how
the workforce will evolve?
There are a lot of the
things you're doing now
to bring the team with you,
which sounds fantastic.
Tone from the top, change management,
demos, champions, all that sounds great.
How much do you have a view on where the-
Maybe there's never a final
end state in this world,
but a few years from
now, do you have a view
on what the new operating model will be
and how org charts, team
structures, how that will change?
- I certainly have different views on it
or hypothesis around it,
but I'm also a little bit careful
around what jobs are going to change,
what jobs are not going to be there,
what jobs are going to be
there because it creates,
I think, a lot of unnecessary nervousness
that I don't think
actually is that conducive
and supportive to the transformation.
It doesn't mean that we should
put our head in the sand
and say, "Look, we're
going to be as many people
working in customer services,"
because we're probably not.
Customer services is probably an area
where we will see less
people picking up the phone
because more people
will, as I said before,
be contacting us via
our financial advisor.
So, of course, there are some areas
where I think it is more obvious
that the technology will
replace some of those jobs.
At the same time, we'll be
needing more people working
with agent and agentic
flows, working with data,
working with cyber risk, at
least in the foreseeable future.
Now, yes, I have views
and we also have processes
around strategic workforce
planning where we try
to associate with the various different,
we call them AI big wins that we have,
and then a constant view
on how you can help our
colleagues, train them,
and move into the areas
where there's more need
for what humans are great at:
building connections with clients,
thinking about big issues and
opportunities and so forth.
- I think the ceiling for banking
and financial services is very high.
A lot more people would benefit
from having a lot more financial advice,
financial planning,
various financial services.
It's just that we've
not found economic ways
to give all these people
the advice and the services
that I think they would benefit from
because it's so expensive
to deliver these services.
So I think of an exciting journey ahead
for the whole banking industry
as we lower the cost to
serve, use more automation,
save bankers more time.
- I think it's a really
great point, right?
Because yes, there is a lot of customers
that like to use the likes of Robinhood
or Avanza, Nordnet here in the Nordics,
so digital platforms for
trading, for example.
But most of our clients also
like to have advisory discussions,
exactly to your point, Andrew,
around their mortgage and how
they should think about it,
around their pension, around savings,
around how to build a healthy economy.
And that is not just done via technology;
that's also done with advisors
that have strong experience
that can really help you guide
these important financial decisions.
And I think you're absolutely right.
The more we can free up that time
for those individuals to
spend time with clients
and then letting AI take care
of some of the more mundane tasks
that are just not as value-adding.
So I think that's a really good point.
- I've been impressed
about how eloquently you
speak around technology.
APIs, you're very facile with
lots of technology terms.
For CEOs that are earlier on this journey,
I'm curious how you spend your time.
You're clearly across
the AI platform build,
your AI big wins.
You seem to have a pretty detailed grasp
about where you are, where you're headed,
how that's quantified, where
the teams are getting stuck.
How much time are you
spending on this topic?
What does that look like?
How would you describe that
for a CEO that's earlier in the journey?
- I would say that in general,
when I look across banking
and my colleagues across
banking, I think everybody's
in fact leaning quite
a lot into technology.
And personally, I spend
probably most of my time now
on both the tech,
but also the leadership
associated with tech.
And those two things,
as we talked about
earlier, go hand in hand.
So you can't talk leadership and culture
without talking technology
and AI and vice versa.
Sitting down with the vendors,
sitting down with the partners,
I think is just absolutely
critical, but also learning.
I spend a lot of time listening
to podcasts and learning
about the technology,
trying out the technology,
of course, visiting our tech partners
and really immersing myself
in terms of what's going on.
I have a hard time seeing how-
My background is risk management,
and one of the things I
love about risk management
in the context of banking is,
to really understand the
machine room of banking,
you need to understand the processes,
the controls, the technology,
how it all fits together.
And so, if you sort of think
first principles on banking,
I think you need to be able
to understand all those bits
and pieces of how it works
together to also understand
how it can transform customer experiences
and colleague experiences.
- How has your background
in risk changed your path
or perspective on AI transformation?
- Look, it's hard to say because it's hard
to sort of imagine myself
having taken another route.
But as you say, I've
spent a lot of my career
in risk management and risk
management is at the very heart
of what we do in banking, right?
We help allocate capital
investment and finance-wise,
just because we are, of course,
a systemically important
organization that is also
a hugely leveraged organization
vis-a-vis the economy.
We're the biggest bank in Denmark
with a balance sheet that's
larger than Danish GDP.
So really, both understanding risk
from how you manage the
balance sheet, the bank,
understanding the various
outcomes and ranges of outcomes,
worst-case scenario, best-case scenario.
And within all of that is,
technology is critically
important because technology, of course,
is what helps you run stress
tests, run scenario planning,
be able to give a client a better view
on different product types.
But, of course, technology
can also create risk
and makes you more vulnerable.
I don't think the last couple years
we've had any physical
bank robberies in Denmark,
just to take that simple example,
which most people can relate to.
Now it's through financial
crime, it's through fraud,
it's through cyber hacks that
those kind of attacks happen.
- I guess one specific
question before we wrap
that I think a lot of CEOs are struggling
with is the AI platform and the decision
to invest significant
money in an AI platform.
What advice would you
have to other CEOs looking
at that decision or just getting started?
- I don't know if I have any advice
because I find it's hugely
challenging as well.
We're also testing,
learning, but, as I said,
what we are doing is we're
really working closely together
with some of the absolute
leading technology firms globally
to do this together in partnership.
And so my very simple logic
is, if you have great partners,
great people, and you're
constantly looking
at what's working, what's not working,
and if you have the agility
to pivot and course correct,
I think you can move forward
and you can transform.
We're focused on competing
in the marketplace
and we just need to do
it a little bit better
so that our customers see us
as being just a little bit better.
- Well, what you've
done is very impressive,
so congratulations on the journey so far.
I think it's really exciting to see
where you're going next
and the capabilities
that you're building to
take advantage of the tools.
- Thanks, probably won't
surprise you, Sarah,
that we don't think like that at all.
We just think about how to make sure
that the next thing we do is going
to get us just a little bit-
Do it just a little bit
better and faster, right?
I mean, it's just like constantly,
"What's the next thing," right?
- Yeah, my experience when
I did all the research
for the "Agile" book in 2019, 2020
was that the companies
who were the most agile
and the most adaptable
were actually the ones
that were most unhappy with where they are
because they've set such big goals
and they're testing and
learning against them,
and so they see how much
further they have to come.
And so I'm not at all surprised
by your response to that.
- No, no, no. I'm with you.
- It's great. It's a good sign.
- Thanks. It was interesting.
Good fun.
Thanks and really good
meeting both of you.
- Thanks for your time today.
(upbeat music)