Ukbusiness.ai
Get our free newsletter

Explaining Your AI’s Financial Decisions

A quarter of business leaders now admit they can’t explain what their own AI is doing with company finances and regulators are starting to notice. This issue covers what that gap could cost you under UK GDPR, an honest review of Starling‘s new AI banking assistant and the September events worth your time.

Prompt Engineer

If an AI tool has flagged, scored or decided something and you need to explain it to a client, investor or the ICO, ‘the algorithm said so’ won’t cut it. This prompt turns that output into a plain-English explanation you can actually stand behind. Before you paste anything in, strip out any customer or employee names and swap in generic labels like ‘Customer A’.

“Act as a compliance and communications advisor helping a UK small business explain an AI-generated decision in plain English.

Here is the situation:

  • The AI-generated output or decision: [paste the output, score, or recommendation, with any personal data removed or replaced with generic labels]
  • What I know about how the tool reached it: [e.g. what data it uses, what it’s designed to flag, any settings I changed]
  • Who I need to explain this to: [Insert audience, e.g. a client, an investor, the ICO]

Please write a short, plain-English explanation covering:

  1. What the AI decided or flagged, in one sentence
  2. What information it likely used to reach that decision
  3. What role a person played in checking or approving it before it was acted on
  4. A clear, honest answer to the question ‘how do you know this is right?’

If there are gaps in what I’ve told you, flag them rather than guessing. I’d rather know what I don’t know than sound confident about something I can’t back up.”

AI Updates Roundup

AI is being used for a lot of financial admin this year. This week’s news asks whether anyone’s keeping an eye on it.

Overview

  • A quarter of business leaders can’t explain their own AI’s financial outputs.
  • Starling’s AI assistant now helps business customers with admin and fraud checks.
  • The government may step in if voluntary AI safety rules fall short.

A new survey found 85% of small businesses now use AI for sensitive financial tasks, yet a quarter of leaders can’t explain what it’s actually doing.

  • So what? If you can’t explain your AI’s numbers to a client or investor, it can become a compliance problem fast.

Starling has begun rolling its AI assistant out to business customers, offering help with everyday admin like VAT tracking and spotting invoice scams.

  • So what? Less time spent on transaction admin frees up hours for the parts of the business that only you can run.

The UK’s AI minister has said the government would consider regulating powerful AI models if today’s voluntary safety rules don’t turn pout to be enough.

  • So what? Nothing changes today, but it’s a reminder not to treat the current light-touch rules as permanent, especially if AI touches customer data.

“Blindly trusting a tech “black box” with sensitive financial data is a massive legal and compliance gamble.” – Zohra Huda, editor of Startups.co.uk

When You Can’t Explain Your Own AI

A quarter of business leaders can’t explain what their AI is actually doing with their numbers. If that’s you, it’s worth fixing before someone asks.

Overview

  • Under UK GDPR, you must be able to explain how your AI reaches decisions on financial data
  • Check which AI tools can show their logic and add a human check where they can’t

The problem isn’t the AI. It’s that a quarter of leaders can’t explain how it reached its numbers when someone asks. Editor Zohra Huda put it plainly: founders are letting AI run their accounting, but if a stakeholder asks how the figures were calculated, many can’t answer.

Why this matters more than it sounds

This isn’t just an awkward moment in a meeting. Under UK GDPR, you’re required to explain how automated systems process personal data. If your AI tool is making decisions about someone’s expenses or credit risk and you can’t explain the logic, that’s a compliance gap. The ICO can fine businesses up to £17.5 million or 4% of global turnover for breaches.

What to do about it

You don’t need to quit your AI tools. You need to be able to answer one question for each one: ‘If a client, investor or the ICO asked how this number was reached, could I explain it?’

Start with the tools touching your money:

  • List every AI tool with access to financial data, even the ones that have crept in through an accounting platform’s ‘smart’ features.
  • For each, check whether it can show its logic. Some tools have an audit trail or explainability feature buried in the settings.
  • Where it can’t explain itself, that’s your highest-risk tool. Either find out why it made a specific decision before you rely on it again or add a human check before anything goes out the door.

The Takeaway 🚀

Pick your one AI tool with the most access to financial data. This week, find out if it can show its working. If it can’t, that’s where you start.

Quick Review: Starling Assistant🏦

Starling has built an agentic AI assistant directly into its business banking app, capable of acting on instructions rather than just answering questions, such as setting aside money for VAT and flagging potential invoice fraud. For UK sole traders and small firms getting to grips with Making Tax Digital, having this sit inside the account you already use is a genuine timesaver.

Key Takeaways

What’s Good ✅:

  • Sets aside funds for VAT automatically
  • Flags potential invoice fraud before payments go out
  • Free for Starling’s UK business customers, with no separate subscription

What’s Less Good ❌:

  • Accounting software integration and receipt data extraction are not live yet
  • No support for Microsoft 365, Xero, QuickBooks, Dynamics 365 or Salesforce
  • Only useful if you already bank with Starling

Impact on Your Business 📈: For sole traders, tradespeople and small agencies handling frequent supplier payments, this cuts the manual slog of tax admin and adds a layer of fraud awareness without extra software or cost. Invoice fraud cost UK businesses more than £41 million in 2025 according to UK Finance, so a built-in check inside your banking app is a welcome layer of protection, even though it will not catch everything.

Financial ROI 💰: The tool is aimed at cutting time spent on transaction categorisation, VAT preparation and chasing payment queries rather than promising a fixed cash saving. Because it comes free within your existing Starling account, any time saved on admin is effectively pure upside.

Price Point 💳: Free for Starling’s UK business account holders, with accounting software integration and receipt extraction due later in 2026.

Upcoming Events & Conferences 📅

Cyber risk and AI now sit on the same agenda for most boards and this free session unpacks what that means for deployment controls.

techUK members get hands-on demos of BigQuery, NotebookLM and Gemini Enterprise, with governance guidance built in.

If you’re an SME leader based outside London, this is the one to prioritise: 200+ free sessions including a ‘Beyond the Hype’ track built for exactly this audience.

  • Big Data LDN 📊
    • Date: 23–24 September 2026
    • Location: London
    • Cost: From £49 (free registration closed 27 August)

Vendor comparison is the real draw here, useful if you’re actively shopping around, though the scale leans more corporate than SME.

A niche one for law firms, LawTech and conveyancing businesses specifically, offering coordinated access to regulators rather than a general audience session.

Networking 🤝

Best suited for technical teams building or running cloud-based AI systems who want peer exchange on AWS architecture and tooling.