Start with an argument,
not an architecture

The first ninety days usually go on evaluating tools and drawing target architectures. Both feel productive and neither produces anything the business can use, which matters because the budget conversation comes at day one hundred.

Pick the argument, not the dataset

Somewhere in the business there is a recurring disagreement that better numbers would settle. Which customers are actually profitable. Whether the discount policy is working. Where the returns are coming from.

Start there, because it gives you a finish line somebody outside the data team recognises. Starting with a dataset gives you a finish line only you can see.

Weeks one to four: follow it backwards

Take the disagreement and trace it to the numbers behind it, and those to the systems they come from. You will find the real state of your data faster this way than through any audit, because you are following the path that actually matters instead of surveying everything equally.

Expect to find at least one spreadsheet, one field that two systems disagree about, and one report whose logic nobody can explain.

Weeks five to nine: build the narrow thing

Ingest only the sources that argument needs. Land them raw, model them properly, define the two or three metrics involved once, and put them somewhere people can see.

Resist widening. The temptation to add adjacent sources while you are in there is strong and it is how ninety-day plans become nine-month plans.

Weeks ten to thirteen: settle it in public

Take the answer to the people who were arguing. Sometimes it confirms one side, sometimes it shows both were measuring different things, occasionally it shows everyone was wrong.

All three outcomes are wins, because all three demonstrate that the platform can end an argument. That is the thing that gets you the budget for the next ninety days, and no architecture diagram has ever done it.

What you will have at the end

A working slice rather than a foundation for everything. Raw history for a few sources, a handful of governed metrics, and one settled argument. That is a much better position than a chosen toolchain and a plan.