Start with the spreadsheet that runs your company

Most organisations run on a spreadsheet nobody wants to touch. Dozens of tabs, formulas built on top of other formulas, links to files that have since moved, and logic that only one or two people ever fully understood. At least one of them has usually left. The file is fragile and barely documented, and it is also the source of a number the business depends on every month.

Nobody rebuilds it, because nobody can prove a replacement would return the same answer. So it stays in place, quietly critical, and the risk sits there with it.

That spreadsheet is the right place to start. Not the first thing to replace, the first thing to start with.

The usual advice makes you wait

The standard sequence is to build the data platform first. Clean the pipelines, put governance in place, stand up the warehouse, and only then add intelligence on top.

It is a defensible plan, but it puts every bit of value at the end. A platform build runs for months, often more than a year, and the return is promised for the final phase, which is the phase most likely to slip. In the meantime the fragile spreadsheet keeps running the business, because replacing it was always scheduled for later.

There is a lower-risk order.

The same pieces in a different order, with the value moved to the front.

Load the meaning, then the file

Start with one important process. Bring in the definitions behind it, the documents that describe it, and the rules people currently keep in their heads. Then import the spreadsheet alongside them.

The order matters. A spreadsheet handed to a general chatbot on its own produces confident but unreliable answers, because the model has no idea which column is authoritative, how your business defines a term, or which rule takes priority. When you load the definitions first, the file is read against your agreed meaning instead of the model's assumptions. The definitions are not decoration. They are what make the answers usable.

Four steps, and the hard part is the meaning, not the technology.

How the answers can be trusted

This is the part that decides whether any of it is worth doing, so it is worth being precise.

The brain answers from the meaning you gave it, not from a guess about your columns. Every answer can be traced: it shows the rows, the tab and the definition it used, so a person can check the working in under a minute rather than taking it on faith. Where the file is ambiguous, where two tabs disagree or a definition is missing, it says so instead of covering the gap with a confident number. And it can run entirely inside your own environment, on your own infrastructure, so your most important file is never uploaded to someone else's cloud.

Four properties that turn an answer into something you dare to use.

There is an honest limit worth stating plainly. The answers are only as good as the meaning you provide. Vague definitions produce vague answers, which is exactly why the definitions come first and why a person still signs off on anything that matters. The brain does not remove judgement. It makes the basis for judgement visible.

The brain does not remove judgement. It makes the basis for judgement visible.

What you actually get

The value is concrete and it arrives early.

A controller asks why last month's margin came in below forecast. Instead of opening the file and working backwards by hand, they get an answer that points to the three customers who drove it, a surcharge rule that had been applied differently across two tabs, and the definition of margin the company agreed to use. They click through to the rows and confirm it. What used to be an afternoon of reconstruction becomes a question and a check.

From there the same file answers plain-language questions, surfaces trends across time that nobody had the hours to pull out, and exposes the places where the old logic disagreed with itself. Knowledge that lived in one person's memory becomes explicit and open to inspection. None of that required a data platform. It required the meaning around your most important process, and the file you already had.

Build the platform once the value is proven

When the value is clear, the case for the foundation makes itself, and it is a stronger case than it was before.

You have not skipped governance. You have started it. The definitions and rules you agreed in order to ask good questions are the first governed assets your organisation owns, and they were defined at the point where they paid off rather than as a year-long prerequisite. Now you build the data platform to give those definitions a reliable, quality-controlled source, and to retire a manual file that breaks when its author leaves.

The work does not restart. The definitions, the rules and the corrections you captured while proving the point become the specification the platform is built to serve. You invest in the foundation already knowing what it has to do, because you have watched it deliver. That is a de-risked build, not a leap of faith.

Start small, in the right order

"Start small" is worth very little as a slogan and a great deal as an order of operations.

Bring the meaning and the one file that matters. Prove the value in ways people can verify for themselves. Let that proof justify the foundation. Keep the brain through every step, because the brain is the asset and the platform is only a better way to feed it.

The spreadsheet everyone is wary of touching is the safest place in the company to begin. The file keeps working while you do it, nothing is ripped out, and by the time you build the platform you already know exactly what it has to deliver.