Your Spreadsheet Workaround Is About to Become a Skill

I asked a controller at a $14M distribution company to narrate the revenue tab of her close file. It took twenty minutes. Nine manual overrides. A row labeled "adj to tie" that had been sitting there since 2023. One hardcoded margin percentage for a single customer, because that customer's rebate posts to COGS a month late and nobody ever fixed the mapping. She knew every one of them cold. None of it was written down.

On June 26, Microsoft shipped Copilot skills in Excel: teachable, repeatable routines for the work finance does every month. Building models, closing the books, refreshing reports, running variance analysis. There's a sample library, you can author your own, and partners like Ramp and Vena will distribute theirs through the marketplace. Brian Jones of the Excel product group described a skill as something that guides Copilot through the steps instead of starting from scratch each time.

That will work. That's the part worth sitting with.

What gets encoded is whatever the file does today

AFP's 2025 survey found 96% of finance professionals plan in spreadsheets and 93% report from them weekly or daily. Under $20M in revenue, that isn't a stopgap on the way to a real system. The spreadsheet is the system. It carries logic that never made it into the ERP and probably never will.

Training a skill on that file doesn't capture the process you meant to run. It captures the one you actually run, compensating steps included. To Copilot, the override in row 43 and the revenue cutoff test look the same. Both are steps. Both happen every month. One is policy and one is scar tissue, and a skill will execute them with identical confidence, faster than anyone can read the output.

A patch survives because a person remembers why it's there. Encode it and the memory becomes a routine, and the routine outlives the only person who could have told you it was temporary.

Sort the steps before you teach them

Walk your close file line by line and put every recurring step in one of three buckets.

A rule comes from outside your spreadsheet. GAAP, a contract term, a covenant definition, a board-approved policy. Teach these first. They're stable and the documentation value alone is worth the hour.

A mechanic is movement without judgment. Pull the trial balance, roll the schedule forward, format the variance columns, tie the subledger. Automate these hard. This is where the time actually goes.

A patch exists only because something upstream is wrong. The hardcoded margin at that distribution company was a patch. Fixing the rebate mapping in the ERP took about forty minutes once someone looked at it. Had she trained the skill first, that override would have run every month forever, and it would have run under the label "monthly close" rather than "known defect."

The sort takes a couple of hours and it produces something more useful than the automation: a written list of everything your close depends on that isn't in a system of record. Most companies I work with have never had that list. It's usually the shortest path to a faster close, whether or not you teach Copilot anything.

Excel is about to learn your close exactly as you run it. Decide now which parts you want it to remember.