By Martin Tombs, Field CTO EMEA, Qlik
There’s been plenty of discussion about how AI could transform financial services in the UK. Many banks are already testing, if not using, the technology to support customer service, fraud detection, credit assessments, compliance and everyday administration.
But much less attention has been paid to what happens when a bank later wants, or needs, to change the AI companies it works with.
The UK government is reportedly currently assessing the economic and security risks British businesses could face if they lost access to the most advanced AI models. At the same time, rating agency Moody’s has warned that banks’ reliance on a small group of AI and cloud providers could leave them exposed to outages, disruption and higher prices.
This does not mean financial institutions should be wary of adopting AI. But they should consider how much freedom their current choices leave them. Decisions that make sense today could create vendor lock-in that becomes difficult and expensive to address later.
AI can become difficult to untangle
On paper, replacing one AI model with another might sound fairly straightforward. In practice, the task becomes much more complicated once the technology is widely used across a bank.
The model may have been adapted for particular tasks, connected to large amounts of company and customer data and built into processes that employees use to make decisions. Its outputs may feed directly into other systems, while some functions could depend on tools supplied by the same cloud company that hosts it.
A replacement would need to work with the bank’s existing systems, understand the same data and produce answers that meet the required standards. The bank would also need to check that the new arrangement meets its security, privacy and regulatory requirements.
These dependencies do not appear overnight. They build as more teams adopt AI and find new uses for it. A bank can gradually become more tied to one provider than it realises, making any eventual move slower, more expensive and more disruptive.
Make switching part of resilience planning
Banks should prepare for that possibility before they face pressure to move. The ability to change AI provider should form part of how they think about operational resilience overall.
Banks already test backup systems, keep track of important suppliers and plan how essential services would continue after a technology failure. Similar planning can help them prepare any future changes of AI provider.
There are several reasons why a bank might choose to move. A service could become unavailable, prices could rise or contract terms might become less favourable. A model could also fall behind newer alternatives, while changes to regulation, data rules or the bank’s own requirements could make another option more suitable.
Preparation begins with a clear picture of how AI is being used across the business. Banks need to know which models they rely on, where those models run, what information they can access and which services depend on their answers. This should include AI features built into software from other suppliers, where the underlying model or cloud provider may be less obvious.
The next step is to put those assumptions to the test. Banks should consider which services would be affected by a move, whether another model could take over, how long the process would take and whether the replacement would produce answers of a similar quality. Testing this early gives them time to close any gaps before a change becomes urgent.
Keep control of the data
Data is where an exit strategy becomes practical. If a bank cannot easily find, understand and move its information, connecting it to another AI model will be difficult.
The challenge is that bank data is rarely stored neatly in one place. It may sit across several systems and cloud platforms, use different formats and labels or lack details explaining what it means and how it was collected. Before moving to another model, a bank needs a clear view of what data it holds, where it came from, how it has been changed, who can use it and whether it is appropriate for the task.
Clear records also make it easier to compare how different models use the same information and understand why their answers may differ. This is particularly important when AI supports decisions involving lending, fraud or customers. After changing a model, the bank must still be able to explain which information influenced a decision and show that the correct checks were carried out.
Banks should also understand the trade-offs involved in using features available from only one provider. Specialist tools can deliver real value, but important data, rules and records should remain accessible if the bank later decides to move.
Build in room to change
Banks can keep their options open by testing different models for important tasks, keeping their data separate from the tools analysing it and using widely supported formats where possible. Contracts should also explain how data can be retrieved, what support will be available during a move and what happens when an agreement ends.
AI will continue to develop quickly, with new models emerging and different tools proving stronger at different jobs. Those that build in room to change will be better placed to take advantage of that progress. A credible exit plan gives them the confidence to invest in AI today without giving up their choices tomorrow.
