Interview: How ING reaps benefits of centralising AI

At ING, artificial intelligence (AI) is not a new phenomenon. The wholesale and retail European bank has been using AI for a decade and has centralised its AI development efforts to scale product development across multiple countries while maintaining strong connections with business divisions.

As Computer Weekly has previously reported, ING is focused on the use of AI in five key areas of the business: know your customer (KYC), call centres, in wholesale banking to improve customer due diligence, in retail for the hyper-personalisation of offerings, and inside tech for engineering.

Discussing the centralised approach, Marco Li Mandri, who used to work at Vodafone and is now ING’s head of advanced analytics strategy, says: “A clear consequence of how we are organised is that, from the beginning, we invested in having one platform. This is the platform that we use to build most of our AI, all of our GenAI [generative AI], and that we are positioning for agentic AI.”

According to Li Mandri, ING’s centralised approach to AI development has resulted in a high success rate for pilot projects, with 90% moving to production compared to the industry average of 30%.

The bank has standardised on cloud-hosted AI models from preferred partners, which are then made available globally, allowing ING to scale. He says the platform is centrally managed with risk controls, guardrails and real-time monitoring.

Regarding ING’s approach to digital sovereignty, Li Mandri says: “You can design for this with many of the cloud platforms. They have EU-based servers where the data does not leave the EU, but there are still some dependencies to provide continuity of the service.”

The AI tools ING has deployed are being actively used and, according to Li Mandri, employees interact with these tools 30 to 50 times per week in some markets. For instance, ING is rolling out AI productivity tools, such as Copilots for software engineers and Microsoft 365 Copilot, to enhance efficiency across teams. In addition, it has trained 5,000 employees on data fluency and GenAI to ensure they can effectively use and challenge AI systems.

ING has deployed generative AI-powered customer-facing chatbots in multiple countries, automating 75% of customer queries. The bank is focusing on high-value problems, such as mortgage processing and anti-money laundering, where AI is being used to augment human capabilities rather than to replace workers.

When asked about how the bank rolled out GenAI-powered chatbots to its customers, Li Mandri says customers can request to speak to a real person at any point during the online chat conversation, providing a human in the loop. ING has also implemented a data-driven quality assurance framework to avoid AI hallucinations and malicious manipulation of prompts.

“This is the platform that we use to build most of our AI, all of our GenAI [generative AI], and that we are positioning for agentic AI”

Marco Li Mandri, ING

Given that ING is one of the biggest mortgage lenders in Europe, there is an opportunity to use agentic AI to handle parts of mortgage application processing as Li Mandri explains: “We are using an agentic AI system in mortgage processing to help back-office employees conduct all the checks they need to do.”  

The agentic AI system handles the task of extracting information from the various documents involved in a mortgage application and can perform checks on these. “It will not replace the human,” says Li Mandri, “but it’s like the human will have a mini team  and, if we are able to do that, it will have quite some impact on the time to approve mortgage applications.”

Like many organisations using AI to improve workflows, ING is ensuring there is always a human in the loop. Article 1 of the EU AI Act covers human oversight in high-risk AI systems – from a financial services perspective, this covers AI systems that are intended to be used to evaluate the creditworthiness of or establish their credit score of people.

Discussing compliance with the Act, Li Mandri says the bank already had a robust compliance framework: “When we did the analysis of the EU AI Act, we realised a lot of it was already embedded in the bank. But new technologies like generative AI introduce new risks, so we have 140 different risks that we specifically vet when we deploy GenAI systems.”