AI can prepare the spreadsheet. You still own the assumptions.
A Business Crumbs finance check with Paul Numbers
A spreadsheet can look beautifully organised and still tell you the wrong story.
For Paul Numbers, the useful question is not just how quickly a report gets finished. It is whether you can explain what the numbers mean before you act on them.
The news
In a case study published on 10 November, Microsoft says its Indonesia CFO, Edu Saputra, reduced his weekly pipeline update from about six hours to one using Copilot Cowork. He still checks figures and assumptions before sharing the update with leadership. The pilot began in July: this is a newly published example, not a new product launch. These are Microsoft's own reported results, not an independent benchmark or a promise for your business. Read Microsoft's case study.
The small-business opportunity is worth exploring. The shortcut is not to skip the checks.
Here is a simple review routine to try with your next AI-assisted management report.
1. Check where the numbers came from
Keep the original export and the reporting period beside the draft. Pick the figures that matter to your decision and trace them back to the records.
Ask whether the report uses the same dates, currency and VAT treatment throughout. Check for duplicate entries, missing transactions and invoices that have not been paid. A tidy chart cannot answer those questions for you.
Paul's check: Can I trace this figure back to its source?
2. Make the assumptions visible
Separate what has happened from what you expect to happen.
For example, imagine a report shows €12,000 in sales, but €4,000 is still unpaid. That is not automatically €12,000 available to spend. A cash forecast also needs to say when you expect the outstanding money to arrive, alongside your opening balance and planned payments.
Those figures are an illustration, not data from Microsoft's case study.
If a forecast depends on a customer paying next Friday, write that down. Then ask what changes if the payment arrives two weeks later. An assumption is much easier to challenge when it is not hiding inside a confident paragraph.
Paul's check: What needs to be true for this forecast to hold?
3. Name the person making the decision
Before a report prompts a purchase, hiring decision or change in spending, give someone responsibility for reviewing it. Record the question being decided, the evidence checked and any uncertainty that remains.
For a small business, that might simply be a short note from the owner or finance lead. The point is not to build another approval maze. It is to avoid treating an AI-generated recommendation as an instruction.
Paul's check: Who has reviewed this, and what decision does it support?
Try one report, not your entire finance function
Choose a recurring internal report. Write down how long it currently takes, including checking and correcting it. Test an AI-assisted version using an approved tool and only the information necessary for the job. Do not paste confidential financial or personal information into an unapproved service.
Compare the results over a few reporting cycles. Count corrections as well as minutes saved. A faster draft is useful only if the finished report is reliable enough for its purpose.
Today's crumb: automate the preparation, not the accountability.
Which figure do you always double-check before making a business decision?


Comments