Claude for Excel in a UK finance team: where it saves time and how to set it up safely

Claude for Excel saves a UK finance team the most time on the writing and explaining around the numbers. That means variance commentary, notes on reconciliation breaks and plain-English walkthroughs of a model. The numbers themselves should still come from Excel's own formulas. Before you rely on it, name a reviewer, set a sign-off rule and list the data that stays out.

Claude for Excel is Claude, Anthropic's AI assistant, working inside Excel itself rather than in a separate chat window. It is one part of Claude for Microsoft 365, which also works inside PowerPoint, Word and Outlook. When Claude changes your workbook, the edited cells are highlighted, so you can see exactly what it touched.

Two other features matter for finance work. First, context carries across the four apps in one conversation, so commentary drafted in Excel can move into a Word report or a PowerPoint board pack. Second, a process that works can be saved as a skill, which is a stored workflow the whole team can run the same way in all four apps.

If Claude is new to you, start with our plain guide to what Claude is. This guide stays with the spreadsheet. For finance work outside Excel, such as policy questions and contract triage, see our post on what Claude takes over in the finance team.

Claude for Excel is generally available on all paid Claude plans, according to Anthropic's product page. Claude for Word and PowerPoint are generally available too, while Claude for Outlook is still in beta. The page does not list the Free plan. You sign in with a Claude account or through your existing cloud provider. Claude for Microsoft 365 is offered on Amazon Bedrock, Gemini Enterprise Agent Platform and Microsoft Foundry.

The table sets out the paid plans as listed on Anthropic's pricing page, in US dollars.

Plan	Listed price	Who it suits	Controls relevant to finance
Pro	$17 a month billed annually, $20 monthly	One person trying it out	Memory on by default
Max	From $100 a month	One heavy user, 5x or 20x Pro usage	Memory on by default
Team, standard seat	$20 per seat a month billed annually, $25 monthly	Teams of 2 to 150	Memory off by default, no training on your content by default, admin tools, SSO
Team, premium seat	$100 per seat a month billed annually, $125 monthly	Team members who need 5x standard usage	As Team standard
Enterprise	$20 per seat a month billed annually, 20 seat minimum, plus usage at API rates	Larger organisations	Adds domain capture, role-based permissions, compliance API access

Every plan has usage limits that reset on a rolling five-hour window, and paid plans add weekly limits. A heavy month-end can hit those limits, so plan seat types around your close calendar. For the same plans in pounds, see our Claude pricing guide for the UK.

Claude for Excel saves time where the finance team writes or explains, not where it calculates. One publisher's guide for finance teams describes this work as document-based, analytical and repetitive in structure. Four jobs fit that description well.

Variance commentary

Claude can turn a budget-versus-actual sheet into prose that non-finance managers can read. The same guide is clear that the human still owns the interpretation, and Claude handles the writing. You decide why marketing overspent; Claude turns your reason into a clear sentence.

Reconciliation notes

Your own formulas find the unmatched items. Claude then sorts and describes them and drafts the note that explains each break to the reviewer. The matching logic stays yours, so a wrong match cannot hide inside a paragraph of text.

Model checks

Ask Claude to describe how a forecast model's inputs flow into its outputs, sheet by sheet. Use that description as a map for your own review, then trace the key links yourself. Treat it as a second pair of eyes that writes things down, not as an auditor.

Recurring report drafts

The same publisher reports that Claude can draft a recurring report from the numbers, last period's report and the key narratives. It puts the draft at 80% complete, and it describes two hours of drafting becoming thirty minutes. Those are one publisher's estimates, not a promise. They do show where the saving comes from: the first draft, not the final check.

Claude should not produce the numbers in your workbook. The finance guide from AI Codex states that Claude does not crunch spreadsheets reliably. It can appear to, but it makes calculation errors that are hard to spot. The same guide's advice is simple: use your spreadsheet or BI tool for the arithmetic and use Claude downstream of the numbers.

In practice, every total, variance and ratio should sit in a cell formula you can audit. The same applies to a report built in Zoho Analytics or another BI tool. Claude reads those results and writes about them. If Claude suggests a new figure, rebuild it as a formula before it goes anywhere.

Accounting judgement is the second limit. The guide notes that Claude can say what a general accounting standard says. It cannot tell you the right treatment for your specific situation. That decision stays with a qualified person on your team or your adviser.

Compliance is the third. The guide says no AI output should be the final word on whether something is compliant. Use Claude to prepare a question for your auditor or tax adviser, not to answer it.

The guide also names one instruction as the most important: tell Claude to flag uncertainty explicitly. A finance assistant that confidently gives a wrong figure is more dangerous than one that says it is not certain and asks you to verify. Put that instruction into every skill your team saves.

This worked example takes a management accounts workbook with a budget-versus-actual sheet per cost centre. The goal is a commentary column for the monthly pack, reviewed and ready for the board. Follow the steps in order.

1. Calculate the variances yourself. Add a variance column and a percentage column as ordinary Excel formulas. Claude will not calculate anything in this example.
2. Set the threshold. Decide which lines need comment, for example every variance above a materiality level your finance lead agrees. Mark those lines with a formula-driven flag column.
3. Give Claude the reasons. In a notes column, type a few words per flagged line: "timing of campaign spend", "new hire started early". Claude writes the commentary; you supply the cause.
4. Ask for the draft. Open Claude in Excel and ask it to write one or two sentences per flagged line into a commentary column. Tell it to use only the notes and figures on the sheet, and to flag any line where it is uncertain.
5. Review the highlighted cells. Claude's edits appear as highlighted cells. The reviewer checks each sentence against the figure beside it and the reason given, then edits or rejects it.
6. Move it into the pack. Because context carries across apps, ask Claude in PowerPoint to build the variance slide in your template, or in Word to add the text with tracked changes.
7. Save the skill. Once a full month's output has passed review, save the process as a skill so the team runs it the same way next month.

Notice what the example protects. Every number comes from a formula. Every reason comes from a person. Claude only joins the two into readable text, and a named reviewer signs it off.

A review step is the named check a person makes on Claude's output before it reaches anyone outside finance. Set one for every task, before the team starts using Claude for Excel. The table gives a starting checklist.

Task	Claude's role	Who checks	What they check
Variance commentary	Writes text from your figures and reasons	Management accountant	Each sentence matches the figure and the stated cause
Reconciliation note	Describes breaks your formulas found	Preparer, then reviewer	Every listed item exists in the ledger and the bank lines
Model walkthrough	Describes how inputs flow to outputs	Model owner	Key links traced by hand; no new figures accepted
Scenario sheet	Sets up a copy with changed assumptions	Finance lead	Assumptions changed only where intended; totals still formula-driven
Board pack slides	Builds slides in your template	Finance director	Figures on slides match the signed-off workbook

Three rules sit across the whole table. Nothing Claude writes leaves the team without a named person's approval. Any number Claude proposes is rebuilt as a formula. Any line Claude flags as uncertain is checked first, not last.

At Svennis we agree the reviewer and the sign-off rule with the finance lead before anyone opens the add-in. We save a task as a team skill only after that reviewer has approved a full cycle of its output. That order keeps a draft that looked right once from becoming the team's standard.

A data boundary is a written rule about which data your finance team may put into Claude and which stays out. Set it before rollout, because a workbook can hold far more than the sheet you are working on.

Plan choice sets some defaults for you. On Team and Enterprise plans, Anthropic states there is no model training on your content by default. Memory is off by default for Team and Enterprise organisations, according to the Claude release notes. It is on by default for Free, Pro and Max. That is one reason to keep finance work off personal plans.

Admin controls are the next layer. Anthropic's deployment documentation lists admin tools and SSO on Team. Enterprise adds domain capture, role-based permissions and compliance API access. If your organisation already runs on Amazon Bedrock or Microsoft Foundry, staff can sign in to Claude for Microsoft 365 through that provider instead.

Then write the rules your plan cannot enforce:

• Which sheets never go in, for example payroll lines that name employees.
• Which workbook is the source of truth, so a Claude draft never replaces the signed-off file.
• Where the ledger lives. Whether that is Zoho Books or another package, its figures win over any text Claude writes.
• Who may send anything outside the team.

On sending, Claude for Outlook cannot send an email on its own. Drafts open in Outlook's compose form and wait for you to click send. Check your rules on personal data against our annotated guide to UK GDPR for AI.

Roll Claude for Excel out one task at a time, starting with a task that recurs every month. Variance commentary is a good first choice because the review step is easy to see. A staged rollout looks like this:

1. One task, one or two people. Run the worked example for a single close. Keep the old manual commentary alongside it so you can compare.
2. A shared finance reference. The AI Codex guide suggests one Claude Project for finance holding your chart of accounts, reporting templates, finance policies and metric definitions. A Project is a shared workspace in Claude where those files sit as reference for every chat.
3. Save the skill. When a full month has passed review, save the process as a skill. Include the instruction to flag uncertainty.
4. Extend to the team. Move the team onto Team seats, switch on SSO and give premium seats only to people who hit usage limits at month-end.
5. Add the next task. Reconciliation notes or board pack drafts come next, each with its own row in the review table.

Excel is often only half the picture. Finance teams also live in Outlook, Teams and SharePoint. Our post on connecting Claude in Microsoft 365 and Teams to the systems behind it covers that side. If you want a structured path for the people, our Claude Team adoption service sets out how a team gets started.

For a UK finance team, Claude for Excel arrives in a market where most firms are still early. Research commissioned by the Department for Science, Innovation and Technology surveyed 3,500 businesses between February and May 2025. It found that 16% of UK businesses currently use at least one AI technology. In finance and real estate the figure was 21%.

Checking is already the norm among UK adopters. The same DSIT AI adoption research found that about two thirds of businesses using AI report significant input or checking of its outputs. Only 2% reported no checking at all. A finance team that builds review steps in from day one is doing what most adopters already do. The research notes that these figures are self-reported and should be read as estimates.

The barriers in that survey match finance concerns. Among businesses that cited them, ethical concerns came out as the most significant barrier at 80%. High costs followed at 76%, then unclear regulation at 72%. Written data boundaries and a named reviewer answer the first and third directly.

Skills are the other gap. The government reports that only 21% of UK workers feel confident using AI at work. Every adult in the UK is eligible for free AI courses under the AI Skills Boost programme, and some take under 20 minutes. One of those courses is a cheap first step for a bookkeeper before the pilot starts.

You can start setting up Claude for Excel this month with a short list of decisions. None of them needs new software beyond a paid Claude plan.

1. Pick one recurring task, ideally variance commentary, and one or two people to pilot it.
2. Choose the plan. For a team, Team seats give you memory off by default, no training on your content by default, and SSO.
3. Write the data boundary: the sheets that stay out, the source-of-truth workbook and who may send outside the team.
4. Copy the review table above and fill in a named checker for each task.
5. Run one close with the worked example, keeping the manual version alongside.
6. Save the process as a skill only after the reviewer has signed off a full cycle.

If you would rather have the setup done with you, Svennis works in four steps. These are a Workshop, a pilot, a department Activation and then monthly operations. For a finance team that wants Claude Team across everyone, the relevant pilot is Pilot Claude B, Claude Team set up in 30 days. Check seat costs in pounds on our Claude pricing page before you decide.