Microsoft Copilot in Banking: 5 Business Use Cases UK Banks Are Actually Deploying in 2026

Microsoft 365 Copilot earns its licence fee in a UK bank in five places: relationship manager meeting preparation, credit paper drafting, regulatory change monitoring, complaints handling under the Consumer Duty, and committee and board reporting. Each one is a drafting and synthesis job sitting on data the bank already holds in Microsoft 365. None of them replaces a regulated decision. This guide sets out the five use cases, the controls that make each defensible to the FCA, what the rollout costs, and the 90-day plan that turns licences into adoption.

By Toni Dos Santos, Co-Founder, Spicy Advisory — previously at BPCE, France's second-largest banking group. Last updated 31 July 2026.

Why Copilot Behaves Differently Inside a Bank

Microsoft 365 Copilot is not a chatbot bolted onto Office. It is a reasoning layer sitting on top of the Microsoft Graph — your Outlook mail, Teams meetings, SharePoint documents, OneDrive files and Excel models — with the tenant's existing permissions applied. That architecture is why it works well in banking and why it fails badly when the underlying estate is untidy.

Two consequences follow, and both matter more in a regulated firm than anywhere else.

First, Copilot's answers are only as good as the bank's document hygiene. A relationship manager asking for a client summary gets a genuinely useful briefing when credit files, call reports and meeting notes live in structured SharePoint libraries. The same prompt in a bank running on shared drives and email attachments returns a confident, thin summary that misses the covenant breach recorded in someone's inbox.

Second, Copilot surfaces what a user can already open. It does not grant new access. But most banks have SharePoint sites that were over-shared years ago and never audited, and Copilot makes that discoverable in seconds rather than in a subject access request eighteen months later. Every rollout we have run in financial services has started with a permissions review before a single licence was assigned. Our guide to AI and data residency for UK enterprises covers the tenant-level questions that sit alongside this.

“The banks that get value from Copilot in the first quarter are not the ones with the best prompts. They are the ones that cleaned up permissions, picked two workflows per function, and told people exactly what they are allowed to do with it.” — Toni Dos Santos, Co-Founder, Spicy Advisory

The Five Use Cases at a Glance

1. Relationship Manager Meeting Prep and Account Reviews

This is the fastest payback in most UK banks, and it is the one RMs adopt without being chased.

A commercial or business banking RM covering 60 to 120 clients spends a significant part of the week reconstructing context: what did we discuss last quarter, what did the client ask for, what has changed in the file, what is the current utilisation. That reconstruction happens across Outlook, Teams recordings, SharePoint credit files and the CRM.

What Copilot does: in Word or Teams, it assembles a pre-meeting briefing from the last twelve months of correspondence, meeting notes and stored documents for a named client, structured against the bank's own review template — relationship history, current facilities, open actions, changes since last review, questions to ask.

A prompt pattern that works: “Using my emails, Teams meetings and files relating to [Client Name] from the last 12 months, draft a pre-meeting briefing using our annual review template. Cover: relationship summary, current facilities and any changes, open actions from the last meeting, matters raised by the client, and five questions I should ask. Cite the file or email each point comes from.”

The instruction to cite sources is the part most teams miss. It converts an unverifiable summary into something the RM can check in ninety seconds.

The control: exposures, limits and pricing are read from the core system, never from Copilot. The briefing is preparation, not a record. Investec, working with Microsoft, has reported bankers saving up to 200 hours a year using Copilot for Sales in this kind of workflow, which is roughly what we see when RM prep is the first use case deployed properly rather than the fifth deployed casually.

2. Credit Analysis and Credit Paper Drafting

Credit is where banks are most nervous about AI, and the nervousness is well placed. It is also where the drafting burden is heaviest.

A credit paper for an SME facility is largely structured narrative: business overview, sector context, financial performance commentary, covenant analysis, risks and mitigants. The analytical judgement is a small fraction of the page count. The rest is writing.

What Copilot does: in Excel it summarises movements in management accounts and flags variances worth explaining. In Word it drafts the narrative sections of the credit paper against the bank's template, pulling from the financial model, the RM's file notes and prior papers. It does not produce a rating, a probability of default, or a recommendation.

The line that must not move: creditworthiness assessment of a natural person is classified as high-risk under Annex III of the EU AI Act, which matters directly for UK banks with EU operations or EU customers. The UK has taken a different route — no horizontal AI law, but the Consumer Duty and SM&CR apply in full. Under both regimes, the same discipline holds: the model drafts prose, an accountable human makes and owns the credit decision, and the file shows which is which. Our explainer on the Digital Omnibus and the revised AI Act timeline sets out what changed for firms operating on both sides of the Channel.

Practical guardrail: require the analyst to mark AI-drafted sections in the working file before submission to committee. Not as a compliance ritual — as a way of knowing, six months later, which parts of a paper were reviewed with fresh eyes.

3. Regulatory Change Monitoring and Policy Drafting

The FCA publishes consultation papers, policy statements, Dear CEO letters, portfolio letters and handbook updates at a rate no compliance team reads in full. Most banks handle this with a horizon-scanning subscription and an analyst who triages.

What Copilot does: summarises a new publication against a standard structure — what changed, who it applies to, effective dates, obligations created — and then compares it to the bank's existing policy set held in SharePoint to draft a first-pass gap analysis. With Copilot agents in SharePoint, this becomes a repeatable process rather than a one-off prompt: an agent grounded in the policy library, available to the whole compliance team.

Why this one converts sceptics: compliance teams are the most resistant population in a bank and the fastest to change their minds, because the output is verifiable. A gap analysis is either right or wrong against a document they can open. There is no ambiguity to argue about.

The control: outputs must be source-linked to the specific paragraph, and a named compliance officer signs the analysis. Copilot narrows the reading; it does not conclude. The FCA's Mills Review, published on 6 July 2026, is explicit that firms stay accountable for AI-assisted outcomes and that senior managers must be able to show the reasonable steps they took. We covered its implications in detail in our analysis of the Mills Review for UK financial services.

4. Complaints Handling and Customer Contact Quality

Under the Consumer Duty, banks have to show they are delivering good outcomes, not just following process. Complaints are where that evidence is thinnest and the manual effort is highest.

What Copilot does, in three steps:

1. Summarises the case. Call transcripts from Teams, prior correspondence and case notes condensed into a factual chronology with the customer's actual complaint isolated from the surrounding narrative.
2. Drafts the response. A final response letter against the bank's approved template, in plain English, addressing each point raised rather than the handler's paraphrase of it.
3. Flags what a human must look at. Indicators of vulnerability, potential systemic issues, and cases where the complaint touches a product outside the handler's scope.

The control that is not negotiable: nothing auto-sends. The handler reads, corrects and owns the letter. The Financial Ombudsman Service assesses the firm's conduct, not its tooling, and “the model drafted it” is not a defence anyone has successfully run.

The second-order benefit: once complaint summaries are structured and consistent, thematic analysis across a quarter becomes possible in an afternoon rather than a project. That is Consumer Duty outcome-testing evidence produced as a by-product of doing the work.

5. Committee Packs, MI and Board Reporting

Every bank has a monthly cycle where finance and business management assemble the same pack: performance against plan, portfolio movements, risk indicators, commentary. The numbers take hours. The commentary takes days.

What Copilot does: in Excel, it identifies and explains the movements worth commenting on. In PowerPoint, it converts the approved MI into a draft committee deck against the standing template. In Word, it drafts the pre-read summary that most committee members will actually read instead of the deck.

The control: Copilot never calculates the numbers that go to a committee. Figures come from the MI system or the finance model, verified as they are today. Copilot writes the narrative around verified figures and drafts the pages. Our walkthrough of Copilot workflows for Excel and PowerPoint covers the mechanics, and AI workflows for finance teams covers the wider reporting cycle.

Where it goes wrong: teams that ask Copilot to do arithmetic on pasted data and then present the result. The failure is rarely caught in the room. Keep the calculation layer where it already is and use Copilot for the layer above it.

What Separates Banks That Get Value From Banks That Just Buy Licences

The gap between banks running the same software is not technical. It is three decisions.

They pilot before they scale. Barclays deployed Microsoft 365 Copilot to around 15,000 colleagues before committing to a rollout across 100,000 employees globally, integrating it into the bank's own colleague productivity tool rather than deploying it as a standalone app. Computer Weekly's reporting on the rollout notes that the licence sits inside an existing workflow, which is precisely why usage holds after month three.

They do not build what they can buy. Société Générale spent more than a year running SoGPT, its own internal generative AI assistant, before decommissioning it in favour of Microsoft Copilot in January 2026 — employees complained the in-house tool was falling behind, and the maintenance cost was not recoverable. That is not an argument against building; it is an argument against building the layer where a vendor ships weekly.

They train by role, not by tool. A generic “intro to Copilot” session produces a spike in usage and a collapse by week six. Training built around the five workflows above — with the bank's own templates, its own prompt library and its own approval rules — produces habits. This is the whole argument of our piece on AI training that actually sticks, and it is more true in banking than anywhere else because the cost of casual use is higher.

“Nobody in a bank has ever been fired for buying licences. Plenty of programmes have quietly died because nobody owned the workflow. Name the owner before you name the budget.” — Toni Dos Santos, Co-Founder, Spicy Advisory

The Governance Layer UK Banks Need Before Rollout

None of this requires a new rulebook. It requires the existing one applied to a new tool.

The Bank of England and FCA's most recent joint survey of AI in UK financial services found 75% of firms already using AI, with a further 10% planning to within three years, and only 2% of use cases fully autonomous. The direction of travel is assisted work with a human in the loop, which is exactly where the five use cases above sit. A fourth edition of that survey is being run in 2026. If you are building a governance framework from scratch, our UK AI governance framework guide is the practical starting point.

What It Costs and How to Build the Business Case

Microsoft lists Microsoft 365 Copilot Business at £16.10 per user per month on an annual commitment, currently discounted to £13.80 per user per month until 30 September 2026, or £19.32 billed monthly. An existing Microsoft 365 plan is required. Enterprise agreements and larger tenants price differently, and Agent 365, Microsoft's governance layer for AI agents, became generally available in May 2026 as a separate line.

The licence is not where the business case is won or lost. At £16 per user per month, a 500-seat deployment costs roughly £96,000 a year. One hour saved per user per week at a fully loaded £45 per hour is worth around £1.17m. The question is never whether the arithmetic works in a spreadsheet — it is whether usage survives contact with a busy quarter.

That is why we advise banks to fund the business case on two or three measured workflows rather than a blanket productivity claim. Measure RM prep time before and after. Measure credit paper first-draft turnaround. Measure complaint handling time and quality-assurance pass rates. Three defensible numbers beat one aspirational one, and they survive the finance director's review. If you are comparing platforms before committing, our ChatGPT Enterprise vs Copilot vs Gemini comparison covers the trade-offs for regulated firms.

A 90-Day Rollout Plan for a UK Bank

The single most common mistake is skipping days 1–30 because the licences are already bought. The permissions audit is not a technicality; it is the thing that decides whether the first month generates confidence or an incident.

How Spicy Advisory Helps UK Banks Adopt Copilot

We are not a Microsoft reseller and we do not run awareness theatre. We work with banks, building societies and financial firms across the UK on three things:

• Readiness and adoption audit. Where you actually sit — current and shadow AI usage, permissions exposure, SM&CR and Consumer Duty gaps, and the two or three workflows where Copilot pays back fastest and safest. Start with the free AI self-audit or a free audit call.
• Role-based training for regulated teams. Hands-on AI training for financial services built around your approved tools, your templates and your approval rules — for RMs, credit, compliance, operations and leadership. Delivered in person in London and across the UK.
• Governance and scaling strategy. AI strategy consulting to build the operating model the Mills Review implies: named ownership, human-in-the-loop checkpoints, prompt and evidence libraries, champions, and metrics that hold up in front of a risk committee.