Marnix van Stiphout is chief operations officer and chief transformation officer at ING, the Dutch bank. In an interview published by McKinsey & Company on 1 July 2025, he spoke with Andrea del Miglio and Mohcine Ouass of McKinsey about how ING moved from experimenting with generative AI to running it at scale in under a year. They covered how the bank narrowed its programme, what made it fast, how it handled risk and what comes next.
Why the bank moved
He described three pressures on banks: competition from new entrants, the need to modernise ageing systems, and regulation that keeps growing more complex. Generative AI, in his view, offered fresh ways to respond to all three at once, which is why ING treated it as a priority for the whole bank.
Five areas and no more
Getting from a long list of ideas to a board decision on five priorities took considerable time, he said. The five were hyper-personalised marketing; upgraded chatbots for customer service, along with straight-through processing; know-your-customer (KYC) checks and compliance; faster software engineering; and support for wholesale clients in lending and sustainability. Each was a strategic priority where generative AI could quickly deliver value to customers.
He explained the reasoning behind three of them. In software engineering, AI tools can make legacy code easier to read and maintain, which matters because old technology and bespoke code slow down new products. In compliance, AI can automate transaction analysis and make customer due diligence more efficient, aiming for less fraud, less time spent on checks and a better customer experience. In wholesale lending, it can pull out and summarise ESG metrics from credit and sustainability reports, helping relationship managers and their clients work towards net-zero goals.
On timing, he said execution took less than a year from the decision to focus, and the first live initiative arrived roughly six to seven months after starting from nothing.
What made the speed possible
Van Stiphout set out five factors. The first was discipline on scope: the technology changes quickly and teams can easily take on too much, so ING held to its five areas. The second was freeing up the right people by taking other work off their desks. Third came joint delivery roadmaps, in which business teams and technology colleagues scoped solutions together and agreed milestones and success criteria.
Fourth was outside expertise. ING’s default is to buy proven solutions from the market instead of building its own, and it brought in partners to shape roadmaps, set up the technical infrastructure and deploy the first use cases. It worked with McKinsey on build and deployment and with the large cloud providers on infrastructure. Early, visible results built confidence that a mostly in-house effort might have lost.
The fifth factor concerned risk. Instead of adapting the bank’s existing risk process, ING created a separate, faster route for AI, led by senior colleagues who could decide quickly on solution design and on the guardrails each case needed.
He called board support critical. The board made early decisions on where to play and backed the teams consistently. Each of the five areas had a named board member who followed it regularly, which he described as going beyond simple monitoring. The board also helped people feel safe taking on, or escalating, AI-related risks, and senior leaders stayed close enough to daily work to clear blockers fast.
Starting where mistakes can be fixed
ING deliberately chose early use cases in which errors could be caught and quality controlled. Pricing models and core lending processes were left alone at first. Chatbots made a manageable starting point because teams could test, intervene and tune them before and after launch. Less sensitive customer conversations came first, with customers able to reach a person if the chatbot fell short. Straight-through processing, being more complex, came later.
AI brings new risks, he noted, though it can also improve how existing ones are handled. In KYC, for example, it helps the bank make more targeted requests for customer information. As ING looks at more complex uses, such as autonomous agents supporting decisions on pricing or KYC, the bank has to understand where AI should and should not make decisions. People stay in the loop in sensitive areas.
“You don’t want to scale something that you cannot risk-assess.”
Results so far and the next phase
He pointed to impact in three areas. Generative AI powers the customer service chatbot, raising service quality and passing complex questions to human agents. In marketing, it supports personalised communications and helps predict which customers may leave, so the bank can reach out earlier. In compliance, it reduces the paperwork involved in KYC, making onboarding easier and faster.
Looking ahead, he said the same five factors will guide expansion. He expects generative AI to move from supporting single tasks to automating whole processes, and said ING is already heading towards agentic AI, where decisions are built into automated processes. Getting there, in his view, will take organisational comfort, testing and time.
What it means for leaders
- A short list moves faster than a long one. ING’s five areas came from a board decision, and that decision gave teams licence to set everything else aside.
- For risk and compliance heads, the idea worth copying is a separate, senior-led approval route for AI, so that projects are not queued behind unrelated reviews.
- Choosing first use cases where errors are visible and reversible buys speed without staking the core business.
- Named board ownership of each area turns sponsorship into regular attention, which we read as a less obvious reason for ING’s pace.
- Before agents arrive, settle where machines may decide and where a person signs off.
