The first three days

Three days after we agree what needs to be right, you get a deck. No data has been touched at that point. It is a concept: what we think is going on, the route we would take, and what that route costs.

It is deliberately cheap to make and cheap to turn down. Three days is long enough to think properly and short enough that neither side has built anything they will feel obliged to defend.

It is also where either of us can say this is not a fit. If what you need is not what we build, we would rather tell you in the first week than the sixth month, and where we know of someone closer to the problem we will say who.

If the deck lands and we both want to carry on, the fortnight starts.

Week one

In week one our engineers get onto your systems. Not a sample, not an export somebody tidied up first, the systems the work actually runs on.

So expect us to find the same record stored under two spellings, a field everybody quietly stopped filling in, and a rule that lives in one person's head and nowhere else. We would rather meet all of that in week one than in month four.

We need access to the system of record, to whatever feeds it, and to the history. That can run inside your own environment or under a processing agreement in ours, whichever your security people will sign. We have built it both ways and we will not argue for either.

Credentials are the easy half. The harder half is one named person who can tell us what the data means. Not a sponsor. Someone who knows why that field has been empty since the migration.

You will hear from us during the week rather than at the end of it, and most of what you hear will be questions. Some of them will be about problems in your own data that nobody has put in front of you before. We raise those the day we find them, because a fortnight is too short to save bad news up for a summary.

Week two

Week two we build. The smallest thing that answers the one to three things you set at the start, and nothing beyond it.

You end the fortnight holding two things. The first is a proof of concept running on your own records. Your question goes in, an answer comes out, and you check it against what you already know, which is the only test worth running. It is narrow by design. It is not production, and you should not put anything that matters through it yet.

The second is the three-day deck, written again. Same questions, answered this time with your data underneath them instead of our assumptions. What we found. What we could not answer, and why. What it would take to run this properly, and what that costs. Where the first deck was wrong, this one says where and by how much.

Then you decide, and either answer is fine. If the fortnight convinced you, we move to a contract. If it did not, we stop there and nothing is owed.