Why our sprints run on sanitised data

The most common reason an AI project has not started is not budget or scepticism. It is that nobody can get access to the data yet.

A four-week sprint that needs live data does not take four weeks. It takes four weeks plus however long the data access review takes, which in a bank is measured in months and sometimes does not conclude.

So our sprints run on sanitised sandbox data with no connection to live records. That constraint costs something real: you learn less about edge cases in the actual data, and some findings have to be confirmed later.

What it buys is that the project starts. A stakeholder sees working agents on recognisable data inside a month, and the risk conversation then happens about something concrete rather than about a proposal. In our experience the second conversation is far shorter, because the questions have specific answers.

It also changes what procurement is being asked to approve. Not "give an outside firm access to customer records", which is a serious ask that deserves a serious review. Just "let them work on a copy with the identifiers removed".

The trade is slightly less fidelity for months less waiting.

If you have a use case that has been waiting on data access, write to hello@windmill.digital and we will tell you whether a sandbox version of it is worth running.