MoneyLIVE Nordic Banking: Acquisitions, Mergers and AI, Oh My
By W.B. King
During the first quarter of 2026, the NCUA approved 27 credit union mergers, which at that rate could match the 157 transactions that took place in 2025. According to a Bank Director Report, during this year, credit unions completed four bank acquisitions. Across the pond, Shan Lodh, Head of AI at the London-based Hyperlayer, recently shared thoughts on how FIs can use AI to keep the customers they aggregate after acquisition.
“A lot of the AI conversation is rightly about discovery and cheaper acquisition, but that is only half the task for banks. Customers rarely switch banks outright,” he noted in a written piece for MoneyLIVE, prior to holding a fireside chat at the organization’s Nordic Banking Conference in late October 2026 that took place in Copenhagen. “They quietly quit instead, opening secondary accounts where they focus their spending, engagement and data.”
Hyperlayer bills itself as offering a programmable banking platform that enables FIs to innovate at pace without replacing the core systems that underpin stability. Its configurable design enables banks, credit unions, wealth managers, and merchants to launch innovative financial products quickly and cost-effectively, while maintaining compliance and control. The fintech’s technology is behind the award-winning UK consumer money management and family banking app, HyperJar, and is a member of the highly competitive global Mastercard Start Path program.
“Customers will use conversational AI to ask for sophisticated money management and personalization tools that the bank will need to act on,” said Lodh. “Most banking systems weren’t built for that, so will need to focus on developing accounts that run on rules the customer sets, within the bank’s guardrails, because that’s how to keep those hard-won relationships over the long-term.”
Finding and Comparing Options
Agreeing that agentic search, commerce and payments often get used interchangeably, Lodh said there are ways for FIs to separate these—three steps in one journey.
"Agentic search is an agent finding and comparing options. Agentic commerce is that agent choosing and buying. Agentic payments is where money moves, and that’s where the bank comes in. In September PayPal announced that Meta’s agent will search and pay across its merchant network behind a Confirm button. The one asking to pay is no longer the person who owns the account,” he continued.
“Networks, wallets, and agent platforms are building the rails and screening ahead of the payment. For account-based payments the bank still decides whether to approve it, and it is usually the one answering when an agent gets it wrong. Whatever happens upstream, the purchase ends with a request to the bank.”
Deterministic Enforcement
When it comes to consumers trusting AI with their money, Lodh said trust will only come from “deterministic enforcement.” While an AI model can propose a direction, the model shouldn’t decide what the AI agent does. That decision, he offered, belongs in a deterministic layer at the account, with limits the bank sets.
“Banks can keep using probabilistic models for risk scoring, but the mandate itself is a hard rule. Anything an agent asserts is untrusted input, including instructions slipped in by a malicious web page or message,” he shared. “A firewall checks the merchant, the item, the mandate, the funds, and the customer’s history before money moves, and every decision is recorded so it can be replayed for a customer or a regulator. The model can change as often as the bank likes while the limits stay put.”
Know Your Agent
Hyperlayer, he explained, views an agent as a delegate with a mandate. “Banks already handle delegation, such as a parent setting limits on a child’s card. The difference is that an agent has no legal standing of its own, so the mandate stays the customer’s and the bank needs a clear record of what was granted: who was permitted to do what, with which funds, under what conditions and until when,” he noted. “That mandate should be scoped, revocable, and enforced as money moves.”
When thinking of AI agents, Lodh said: “Each agent needs its own identity on the system of record, so the bank knows which agent is acting, for which customer, and with what permissions,” he continued. “Know Your Agent asks of agents what Know Your Customer asks of people, though it is still emerging and has no common standard. Network programs tell the bank who is asking. The bank still has to decide what the money may do.”
