A duplicate row, date stored as text or trailing space can quietly break a lookup and change the report that follows. Microsoft has added a Copilot skill designed to find common spreadsheet problems, but its most useful feature may be the review step: it proposes corrections and lets the user decide which ones to apply.
What Microsoft released
Microsoft's October 2026 Excel update, published 27 October and updated 29 October, says the Clean Data skill is rolling out to general availability across Excel for Windows, Mac and web. A user can invoke it with @clean-data or select it from the Skills menu.
The skill checks relevant data for extra spaces, inconsistent text and number formats, inconsistent dates, duplicate values and values stored as the wrong data type. Microsoft says it explains what it found, why it matters and proposes corrections in the workbook. The user can review the proposed fixes and select which changes to apply.
Microsoft's built-in skills documentation repeats an important limitation: Clean Data is still rolling out and may not be available to every customer. It is enabled by default in Copilot in Excel, but availability still depends on the account and deployment.
A review-first cleanup workflow
1. Preserve a baseline. Save a copy or create a named version before cleaning. Record the workbook, sheet, reporting period and source files. If a proposed correction is wrong, the reviewer needs more than Undo as a recovery plan.
2. Define what “clean” means for this dataset. A repeated invoice number may be an error; two legitimate transactions with the same customer may not be. A date displayed differently may still be valid. Write down the identifiers, units, allowed values and duplicate rules before accepting suggestions.
3. Review one issue type at a time. Start with spaces and data types, then dates, formats and possible duplicates. For each proposed fix, inspect the original cell, nearby rows and the source record. Do not accept a bulk change merely because the explanation sounds reasonable.
4. Recalculate and test dependencies. After approved changes, check formulas, lookups, PivotTables, charts, filters and totals. Compare key outputs with the baseline and investigate every material difference. A technically consistent value can still be the wrong business value.
5. Keep a short change record. Note what was corrected, what was deliberately left unchanged and who reviewed it. For finance, audit, health, education or regulatory work, follow the organisation's existing control and approval process rather than treating Copilot as the approver.
6. Export and inspect the PDF. Use the approved Excel export or print workflow. Reopen the exact PDF and check totals, labels, column widths, page breaks, repeated headers, footnotes and clipped content. Confirm that hidden or filtered rows were handled intentionally and keep the editable workbook separately.
What the skill cannot prove
The announced checks are useful examples, not a complete guarantee of data quality. Clean Data cannot determine whether the source was authorised, a figure reflects reality, a duplicate is legitimate or a report meets accounting, legal or statistical requirements. Microsoft also warns elsewhere in its skills guidance that AI-generated content may be inaccurate or incomplete and does not replace organisational review or approval controls.
After an authorised report has been exported, FileStation's PDF Reader can support a separate page-by-page visual check. The CLEAR document check offers a short release routine. FileStation does not connect to Copilot, validate Excel formulas or certify the cleaned data.
The practical improvement is not automatic perfection. It is a faster way to surface likely problems while leaving the consequential decision—what to change and what to publish—with the reviewer.