Stories Case study: The HR department of a Fortune 100 technology company

When paychecks depend on Excel formulas

A regulated payment's path from a spreadsheet to a dedicated system, with an audit trail and five times the throughput

The HR department of a Fortune 100 technology company calculated a regulated payment to employees in one spreadsheet. Fewer than five people used it, and every change in the regulation meant editing formulas by hand in a rush. Sometimes the only way to the right number was to override a cell.

I was the architect and back-end developer on the small team that replaced it with a web application. The rules were in one place, with an audit trail, an approval flow, and employee data pulled in instead of typed in. HR estimated they reviewed about five times more cases a day, and a regulation change stopped meaning hand-edited formulas.

Users
fewer than five people
First version
a few months, in 2020
Throughput afterwards
about five times more cases a day

The challenge

This client was a department inside a large company, not a small standalone business. The spreadsheet, though, had the exact characteristics I see at small companies: a handful of people around one file, with the rules being partly in formulas and partly in someone's head.

The calculation was complicated for good reasons. It had many inputs and many exceptions, and a regulation stood behind every one of them. When the regulation changed, the formulas had to change with it, by hand, before the next case. When the formulas couldn't express a case, someone overrode the value. Every one of those edits was a compliance risk, and a shared folder gave no way to tell who had changed what, or to stop someone from changing what they shouldn't.

What we did

  1. 1

    Speed up the process that already worked

    The process was fine. The tool had reached its limit. So the first version did exactly what the spreadsheet did, faster and safer, and nothing more. Employee data came from the central database instead of being typed in, and the people who used it told us what we got wrong before it cost anything.

  2. 2

    Put the rules where they can be checked

    The rules were in one place, handling exceptional cases, too. A regulation change became a dated rule, introduced ahead of time and applied from the day it took effect. The system refused overrides that weren't allowed and logged every change with who and when. A change could be reverted, and a four-eyes rule became an approval flow: one person enters the case, another approves it. Roles decided who could see or edit which fields.

  3. 3

    More, once the basics worked

    Once HR saw it worked, they wanted more. The system started sending email notifications from templates and generating the official documents that had been produced by hand. An Excel export stayed, because Excel is still the right tool for the analysis they did on the side. The first version took a few months. The extras, and finding out which extras mattered, took the rest of a year.

What changed

HR estimated they reviewed about five times more cases a day. The time went from typing data in to checking that the data was right.

Regulation changes were introduced ahead of time and automatically applied from the day they took effect, so nothing had to be rushed any more. Every number had a trail, with who entered it, who approved it and the previous value. That's a much easier thing to show an auditor than a spreadsheet in a shared folder.

What we learned from it

We only replace a process that already works. We sped this one up first, and added new things only once the transformed process ran correctly. The other way round, we would have automated the confusion.

We don't write software for things that rarely happen. Five minutes once a week isn't worth automating, unless a regulation says it must be traceable. Then it is.

In 2020 the first version took a few months because people wrote every line of code. Today AI does the typing and an engineer does the deciding, and the same first version takes weeks. What doesn't shrink is finding out what the department actually needs, which is why the first step is that we spend half a day with the people who do the work.

The department isn't named, and the calculation is described only as a regulated HR payment. On a call I can say more.

Does compliance depend on a spreadsheet?

Bring the file and the rules behind it. You leave with a straight answer on whether it's worth fixing, which may be no.