The spreadsheet test
Before automating anything, ask whether a well-kept spreadsheet would do. Usually the answer tells you more than the automation would.
Most requests for automation start with a tool. Someone has seen a demo, or a competitor mentioned an AI assistant, and the question arrives already shaped: can we connect this to that? It's a fair question, and usually the wrong first one.1Not always. When the tool is already chosen and paid for, the useful work is making it fit, not relitigating the purchase.
The test
The first question I ask is simpler. If a person kept this process in a spreadsheet, carefully, every day, would it work? If yes, you have a process with clear inputs, clear outputs and a rule in between. That is exactly the shape that automates well.
If no, if the spreadsheet would be abandoned in a week because nobody agrees on the columns, then automation will fail in the same place, only more expensively.2A spreadsheet is a cheap prototype of a database with a human for an API. Most of what it teaches you carries straight over.
What the test finds
Running a process through the spreadsheet test tends to surface three things:
- Undefined inputs. Two people fill in the same field differently, and both think the other is wrong.
- Hidden judgement. A step everyone calls "just checking" turns out to be a decision only one person knows how to make.
- Missing owners. Nobody is responsible for the sheet, so nobody is responsible for the process.
None of these is a software problem. All of them will sink a software project.
If it can be done in a spreadsheet, do it in a spreadsheet. Then automate the spreadsheet.
After the test
Once the spreadsheet version works, and people actually use it, the automation is almost mechanical. The columns become fields, the rules become code, and the person who kept the sheet becomes the person who checks the output.3This is also the right moment to measure: time spent per week on the manual version is the baseline any automation has to beat.
The point isn't that spreadsheets are the answer. It's that a process has to exist before it can be automated, and a spreadsheet is the cheapest way to find out whether it does.
Some processes pass the test in an afternoon. Others take a month of argument about what a column means. Both outcomes are useful, and the second one is usually where the real work was hiding all along.