AI and Bank Balance Sheet Management: Key Lessons

AI and Bank Balance Sheet Management: Key Lessons | UBA UK

Artificial intelligence is advancing faster than most banks can govern it. For treasury and risk teams, the question is no longer whether AI will reach the balance sheet, but where it adds real value, and where human judgement must stay firmly in charge.

That question shaped the final session of UBA UK’s “Mastering Balance Sheet Resilience” webinar series on 30 July 2026. Professor Moorad Choudhry of the BTRM faculty completed a series that began with liquidity risk and moved through interest rate risk in the banking book, closing with a practical assessment of AI in bank asset-liability management (ALM).

What role should AI play in bank treasury?

A supporting one. Professor Choudhry’s starting premise was that generative AI, and agentic AI used with care, can make the Asset and Liability Committee (ALCO) more effective in its pursuit of an optimised balance sheet. AI is there to amplify the treasurer, not to replace them.

He also warned against treating banks as identical. Institutions differ in size, business model and risk appetite, so what makes sense for one may be wrong for another. Any decision on AI should start from the bank’s own circumstances.

Where can generative AI deliver real gains today?

In narrow tasks with well-defined inputs and outputs. Professor Choudhry set out several practical examples:

• ALCO reporting: cutting pack production from days to hours and producing a one-page dashboard of key risk indicators so that boards see any breach at the start of the meeting, not on slide 200.

• Horizon scanning: summarising regulatory publications, such as the PRA’s consultation paper CP5/26 on liquidity, for senior executives.

• Stress testing: running more scenarios, more quickly, on current balance sheet data rather than month-end figures.

• Funding and FX hedging: generating optimised plans and hedge recommendations for ALCO to challenge.

The caveat is verification. When Professor Choudhry asked an AI tool to revise its summary of CP5/26, it dropped the paper’s most important requirement. A summary that is “97% accurate” is not good enough for a regulator or a rating agency.

Should AI execute treasury decisions on its own?

Not yet, in most cases. Agentic AI acts rather than advises: it can monitor limits, propose hedges and route them for approval. When the audience was asked how comfortable their institutions were with AI executing decisions autonomously, views were evenly divided.

Professor Choudhry’s own preference is clear: AI recommends, and the treasurer decides. Hedging a banking book is “as much art as science”, involving judgements on timing, prepayment and rebalancing. One agentic use he did find compelling was enforcing approved risk limits, so that they simply cannot be exceeded.

What should banks put in place first?

An AI policy, followed by clear governance. Banks should define the outcome they want before building or buying any model. The practical starting kit discussed includes a register of every model and agent, second-line validation before go-live, a named owner for each model, human-approval thresholds, a shared prompt library, AI incidents routed through operational risk, and a manual fallback for every critical treasury function. Existing model risk guidance, such as the PRA’s, provides a sound baseline.

Why does this matter for banks connecting Africa and the world?

One of the sharpest audience questions asked how banks can preserve the treasury judgement built through past crises as experienced staff retire, particularly where succession pipelines are thin, as in many African banks. Professor Choudhry saw this as a global challenge: if AI takes on all junior work, the next generation will not learn by doing. That is a strong argument for starting with the clearest use cases rather than automating everything.

For UBA UK, which supports companies and institutions doing business into and out of Africa, this balance matters. Sound balance sheet management, informed by the best tools and led by experienced people, underpins the reliability our clients depend on across trade finance, correspondent banking and treasury services. The series reflects our commitment to sharing expertise with partners and the wider banking community in the UK and across Africa.

Frequently asked questions

What was the third part of the UBA UK treasury webinar about?

It examined how artificial intelligence can support bank balance sheet management, covering generative and agentic AI, practical treasury use cases, governance and the risks of relying on AI output.

What is the difference between generative AI and agentic AI in banking?

Generative AI produces content such as summaries, reports and recommendations when asked. Agentic AI goes further: it pursues a goal on its own initiative, using tools and taking actions across several steps.

Which treasury processes benefit most from AI?

Webinar participants ranked ALCO pack and dashboard production, and liquidity and interest rate stress testing, as the processes with the most to gain. Regulatory horizon scanning and hedging recommendations were also highlighted.

Will AI replace treasury and risk professionals?

Professor Choudhry’s view is that the core skill will not change: understanding the data and making the decision. AI can provide faster and more current information, but accountability remains with people.

How should banks govern AI models in treasury?

By putting an AI policy in place first, registering and validating every model, assigning named owners, setting human-approval thresholds and applying existing model risk management guidance.

To learn more about UBA UK’s correspondent banking, trade finance and treasury services, email info@ubauk.com

AI and Bank Balance Sheet Management: Key Lessons | UBA UK