ING’s approach to innovating wholesale banking with AI
By Ioanna Lykiardopoulou
AI is becoming integral to modern banking. From advanced fraud detection to chatbots and personalized services, it now supports both front- and back-office operations, providing financial institutions with a competitive advantage.
As adoption expands, wholesale banking represents a critical frontier for AI — one defined by large-scale transactions, complex products, and regulatory pressure. To understand how this shift is unfolding, we take a closer look at how ING is incorporating AI into its wholesale operations.
We spoke with two executives from ING:
Kacy Harding, Chief Product Officer, WB Analytics
Bahadir Yilmaz, Chief Analytics Officer
They provide insights into early and emerging use cases, how ING evaluates opportunities and measures success, the challenges of AI development, and the vision for where the technology could ultimately take the sector.
AI's role in wholesale banking
For a global bank like ING, the integration of AI is not just a technological upgrade; it's a fundamental shift in how operations are managed and how clients are served. “At ING, we see generative AI not just as a technology — but as a catalyst for transformation,” says Yilmaz.
“In wholesale banking, we’ve moved beyond experimentation to delivering real, scalable impact across client services, lending, and sustainability," he notes. "From automating customer due diligence and enhancing ESG insights to launching our WB Client Services Chatbot, GenAI is helping our teams work smarter and serve clients better.”
Harding sees AI as a “critical enabler” that allows the bank to seamlessly leverage large amounts of data quickly and in a scalable way.
The need for speed and scalability is inherent to wholesale banking, which caters to large clients such as corporates. “We’re dealing with highly complex data,” Harding explains. A key example is regulation, as financial rules and institutional structures vary widely across countries and sectors.
“AI really helps us churn through information and identify the pieces we need to serve our clients,” she says. It also enables ING to “piggybank” on the years of experience and domain knowledge of its front- and back-office experts.
“We work together with the front and back office on how to interpret and process data, and how to build the AI models,” Harding notes. Thanks to such benefits, she highlights that AI is driving the move from incremental improvements in analytics to breakthrough innovation.
Three AI use cases
Know Your Customer (KYC)
Know Your Customer (KYC) is essential for verifying client identities and protecting society and financial institutions against risks such as fraud and money laundering. It’s also “a highly manual process, and one that needs to be repeated regularly,” she notes.
That’s why ING has turned to generative AI to streamline the Customer Due Diligence (CDD) side of the process. By gathering data from unstructured source documents, GenAI can reduce touch time, manual file handling, and alert handling — enabling analysts to spend less time on data collection and more on customer analysis and outreach.
Launched on July 1st of this year, a GenAI project providing CDD transaction profile reports has already demonstrated its value in practice. It saves front-office analysts an average of two hours per client and is now in use across all 19 countries where ING operates.
ESG insights
For ING, a key element for loan approvals in wholesale banking is the verification of a client’s ESG profile.
“As part of this process,” Harding says, “we need to conduct multiple compliance checks, ensure all required questions have been asked, and verify the accuracy of documents such as climate reports and credit applications.” One of the main challenges is that documents come in different formats, from texts and images to tables.
To improve efficiency, she explains, ING has been using the support of Large Language Models (LLMs). “They extract relevant information from those documents and transfer them into a system, where our colleagues can verify that we followed all of the compliance steps, and that we are able to verify and audit that we have cited that information from our client.”
Harding notes that AI has been instrumental in streamlining document processing, reducing an average time of six hours to a few seconds.
Centralizing knowledge
Harding’s team is currently experimenting with LLMs that can help centralize knowledge for internal use. The primary goal is to provide back-office experts with a single, up-to-date, accurate source of information.
“We want to make our internal content more accessible and easier to interpret, so that a service agent can quickly and concisely synthesize relevant information and respond more effectively,” Harding says. She believes that this will also have a significant impact in client service and customer engagement, as it will be possible to cross-reference information related to the specific conditions of a particular customer.
“We’re starting small and I do anticipate challenges around hallucinations,” she says. The team keeps testing the model and challenges it with wrong answers or difficult questions to evaluate its performance.
“So far, it hasn’t hallucinated, which makes us more confident that we’re following the right approach.”
Behind the curtain
Identifying potential use cases
“Our philosophy is to democratize the inclusion of different challenges from across the wholesale banking department,” Harding says. This means that any employee can submit particular issues or challenges they’re experiencing. These submissions are then flagged against key business objectives.
At this point, a dedicated team called “Discovery Squad” takes over. The squad reviews the highest-priority challenges through a defined vetting process. They define the problem, assess its technical feasibility and business viability, and secure risk approval. Then follows the proof-of-concept development process.
“One of the critical criteria we put in place when making the high-priority problems list is whether we can have access to healthy and clean data,” Harding says. “If we don’t, but still believe we want to work on a particular use case, we need to ensure that, if there is a data gap, those attributes aren’t something that can throw the algorithm off, or bias our results.”
Measuring the success of AI
To evaluate the success of AI solutions, ING uses three main metrics, starting with quantitative metrics. These include the projected cost savings and efficiency gains, AI’s expected accuracy in solving the problem, and the capital investment needed.
“With AI, as with any breakthrough innovation, it’s more challenging to balance your evaluation scorecard,” Harding says. “We need to make a high upfront capital investment in order to achieve cost and efficiency savings.”
On the qualitative side is employee feedback. “This will also include client feedback when we get to the stage that our products are live on the market,” Harding adds.
The third metric has to do with regulatory audits. “Can we ensure that there is transparency in what we're developing? Are we able to prove the answers that the model is coming up with are accurate? And are our systems compliant?”
Challenges
According to Harding, there are multiple layers of complexity, starting with regulatory compliance. “There are requirements both per region and per country, which also change depending on the sector you’re working in,” she says. “You really have to be on top of your game in understanding how to work within the system.”
In addition, beyond the data quality itself as an underlying source of how the products and models are built, there is also the challenge of identity and access management. “We need to make sure that sensitive information and results can only be viewed by the approved people and roles within the organization.”
At the core of AI development is also the requirement for transparency, auditability, accountability, and explainability. “This underpins the entire process, from how the model has been built to how it performs and what kind of results are being generated,” Harding says.
Towards optimal action prediction and agentic AI
Regardless of the challenges, Harding shares her enthusiasm about AI future potential in banking. She believes that as we experience the tangible benefits of AI and build greater trust in the technology, its role will shift from driving efficiency and competitive advantage to enabling deep personalization and automating complex financial processes.
“Just as analytics evolved from looking at historical data to suggesting the next best action,” she says, “AI will leap forward to more predictions and recommendations of optimal actions across the entire client journey.” For example: what is most optimal for a specific client, in a particular sector, under certain market conditions.
This requires models that are transparent and explainable, but also context-aware — integrating internal data with external sources, such as competitive benchmarks, market shifts, or regulatory developments.
“I think that could be a very exciting next phase,” Harding says.
“As we enter the next chapter, our ambition is clear: to expand GenAI’s reach, embrace agentic AI for front-office productivity, and continue building responsibly — with strong governance, auditable systems, and a relentless focus on customer value,” adds Yilmaz.
🏛️ Would you like to stay up to date with ING's latest news and innovations? Follow its global page on LinkedIn, or check out the LinkedIn profile of ING Nederland.
💡 Want to learn more about the KickstartAI community and our mission to accelerate AI adoption in the Netherlands? Check out our website, listen to our podcast, or follow us on LinkedIn!
