In the short time since generative AI became available to all, ING has moved rapidly to adopt, deploy, and scale the technology. Today, gen AI helps the bank’s frontline teams serve customers better, simplifies internal processes, and reduces costs. The Dutch bank moved from experimentation to deploying gen AI at scale in less than a year, and it currently uses gen AI in five critical areas, including software engineering and know-your-customer (KYC) and compliance, in addition to its customer service.
The Dutch bank’s chief operations officer and chief transformation officer, Marnix van Stiphout, credits his organization’s ability to deploy and scale gen AI so quickly to specific management techniques, such as freeing up the right people to work on the initiative. He spoke about the journey recently with McKinsey Senior Partner Andrea del Miglio and Partner Mohcine Ouass. Their interview has been edited for length and clarity.
ING’s AI transformation
McKinsey: Marnix, generative AI is still a very new technology, yet ING has already deployed and scaled it broadly. Was there a particular imperative for ING to move this quickly?
Marnix van Stiphout: Like all banks, we need to compete with new entrants, and we need to modernize our systems while keeping up with increasingly complex regulatory environments in the markets we serve. In gen AI, we saw a new opportunity to address all these imperatives in creative ways that had not previously been available to us.
McKinsey: How did you ensure a focus on your gen AI program?
Marnix van Stiphout: It took us quite a bit of time to progress from analyzing many different ideas into a decision at the board level, where we could say definitively that these are the five things we’re going to do. We ended up deciding to focus on five strategic domains: hyper-personalizing our marketing, upgrading our chatbots for better customer service and straight-through processing, improving KYC and compliance, speeding up our software engineering, and supporting our wholesale clients in their lending and sustainability journeys. We felt these domains were areas of strategic priority for the bank, where gen AI could deliver customer value fast. Deploying gen AI programs in these areas was meant to create a foundation on which we can build future gen AI capabilities.
McKinsey: Can you say a bit more about the last three of those five strategic domains? It’s not hard to understand speeding up software engineering with gen AI, but what were the reasons for choosing that as a strategic domain? And how can gen AI help with legal and regulatory compliance or with wholesale lending?
Marnix van Stiphout: Yes, obviously developer productivity soars when you can use AI-powered tools to create transparency on legacy code or to enable more effective maintenance and generally make our modernization efforts go faster. This is especially true in our industry, because banks tend to have mountains of legacy technology and bespoke code to connect it all, which creates a lot of drag on innovation, especially of new products.
As for compliance, gen AI can help us speed up and improve our KYC activities and enhance our efforts to prevent financial and economic crimes. By automating transaction analyses and improving efficiency in customer due diligence processes, we can reduce fraud and abuse, spend less time doing so, and improve the customer experience at the same time.
And with wholesale lending, gen AI can help do things like enable net-zero steering by pulling relevant sustainability metrics together for relationship managers and clients. This is done through AI-powered extraction and summarization of ESG [environmental, social, and governance] metrics from available credit and sustainability reports.
Moving to AI at pace: What it takes to win
McKinsey: Once you decided those five areas, how long did it take you to build, deploy, and scale these new tools?
Marnix van Stiphout: Less than a year. Our decision to narrow our focus to five strategic domains marked the start of our journey, and once the direction was clear, execution followed quickly. From zero to the first real initiative, we were probably six or seven months away.
McKinsey: How did you do that? It’s a big win to get most new initiatives going in that kind of time frame, even without having to adopt a new technology.
Marnix van Stiphout: It’s really five factors that let us do it. First, and perhaps most critically, we stuck to the five focus areas we identified in the experimentation phase. It’s so easy to take on too much, especially in this case, because the technology is advancing and developing so quickly throughout the whole process. Second, we committed to freeing up the right people to do the work. That means you have to take other things off those people’s plates, and that can feel hard to do. But if you don’t, you can’t move quickly enough. Third, we focused on opening up the road maps to enable delivery. The business worked together with our AI and technology colleagues to scope out the solutions [and to] define delivery milestones and success criteria. Fourth, we brought in external help when we needed it. With gen AI, the technology is developing so quickly that you really need support from experts to deploy successfully and fast. And then lastly, we streamlined risk assessments at a higher level. Our risk assessment process was not ready for gen AI. Rather than adapting the current process, we created a new “highway” AI risk process led by more senior colleagues to allow for fast decisioning on AI solution design and required risk guardrails.
McKinsey: Let’s unpack some of that a bit more. How did you choose the five areas you focused on for initial gen AI deployment?
Marnix van Stiphout: We were deliberate in choosing lower-risk use cases so we could move fast and feel confident that any missteps would be fixable. We chose use cases where we can still decide how to reach the right quality, and if something does go wrong, it’s repairable. That allowed us to go full throttle in those areas. We were certainly not going to take on pricing models or core lending processes to start, for example. Of course communication with customers itself is sensitive and critical, but we’re talking about interactions via chatbots here, and if you don’t get it exactly right at first, you can intervene, you can test, and you can control quality. Straight-through processing is a bit more complex, but that’s not where you start. You apply the new systems first to less-sensitive customer communications where they can still call in if the chatbot isn’t finely tuned enough yet when we launch it.
McKinsey: On the subject of bringing in outside help, how did you decide where you needed it? There are so many providers out here, all looking to jump on the same gen AI opportunity.
Marnix van Stiphout: Having partnerships has been incredibly important. My thinking is usually that if we can get the right AI solutions from the market, we should get it from the market. This is novel technology, so we had better use what has already been developed. Additionally, we felt we wanted to bring in experts at certain points throughout the process—helping define the road maps, stand up the technical infrastructure we needed, and deploying early use cases. We knew where we wanted to go, but we also knew we needed help to get there fast. In addition to partnering with McKinsey to build and deploy, we partnered with hyperscalers to access and set up the required infrastructure.
I don’t think there is any other way we could have done this, and I don’t think anyone would disagree with me on that. It’s not just about getting the work done. It's also about showing the organization what can be achieved quickly. Having real examples early on—tangible proof points—gives people confidence that this is real. If we had chosen to build on our own or with less outside help than we brought in, we would have risked people losing confidence in the endeavor.
The role of leadership in deploying AI
McKinsey: Can you talk about your leadership team, the board of ING? Did you feel you needed their buy-in to pursue this level of initiative?
Marnix van Stiphout: We felt that having the board’s support throughout was critical. The board made early calls on where to play and then supported teams with clear direction and consistent backing. Once we decided the five areas where we would deploy gen AI, we matched the relevant board member to each of those focus areas and had them follow that piece of the program on a regular basis. That board-level attention went beyond monitoring. With all these new ideas—even within the five domains—there’s uncertainty and risk. People needed to feel safe taking those risks or bringing those risks to their leaders.
At the same time, our senior leaders stayed close to the day-to-day work of building these gen AI programs, ensuring fast decision-making and hands-on help was available when needed. If a team hit a blocker, we had leaders who could pick up the phone and clear it. That made a big difference to our speed. Many of the challenges the teams encounter are new; this is an area where leadership can weigh in and solve these challenges for and with their teams.
Managing AI risk
McKinsey: One of the five success factors was streamlining risk processes at a higher level. How are you managing the risk involved with delivering AI?
Marnix van Stiphout: It is good to realize that AI not only brings new risks but also helps us manage our existing risk in a better way. For example, in the KYC space, we leverage AI to be more targeted in what information we request from customers. However, we need to understand the limits of where AI can make decisions in our bank. This is especially relevant as gen AI matures and we explore more complex use cases, such as autonomous agents that support operations teams with decision making, for example in pricing or KYC. You don’t want to scale something that you cannot risk-assess. So in sensitive areas, we’ll always keep humans in the loop.
Impact of AI thus far and what’s next
McKinsey: You have had these scaled-up gen AI tools in place for some months now. How is it going?
Marnix van Stiphout: The results are already visible. For example, gen AI powers our chatbot, improving customer service quality while routing more complex questions directly to human agents. In marketing, gen AI helps us build personalized communications and helps us predict customer churn, allowing more proactive and personalized outreach. And in compliance, gen AI reduces the documentation burden in KYC processes, making it easier and faster for customers to onboard.
McKinsey: Will you adapt the five success factors you mentioned as you apply gen AI programs to more areas and scale them further? Do they need to evolve along with the technology and its uses?
Marnix van Stiphout: The formula we used will remain consistent. Strategic focus, creating impact, engagement between management and the board, smart use of external partners, and a relentless approach to delivering quickly where we believe value can be realized fast—these will continue to support us. This isn’t about scaling a tech tool, it’s about rewiring how we work.
McKinsey: Now that you’ve made these big advances, what’s next? How do you see your use of gen AI evolving?
Marnix van Stiphout: Gen AI will become more embedded over time, shifting from task support to process automation. We’re still in the phase of complementing human activity—helping our people do more, faster, better. But we’re already seeing the move from task to process with agentic AI, and decision making will become embedded in these automated processes. That shift will require the organization to get comfortable, though. It’ll take a bit of time and a lot of testing, but I’m sure we will get there.
