The first conversation
We ask about the decision rather than the technology. Which recurring decision do you want to improve, who makes it, how often, and what does it cost when it goes wrong?
That usually takes half an hour and frequently changes the shape of the project. It also occasionally ends it, when the honest answer is that a report, a rule or a process change would solve the problem without a model.
We would rather establish that in the first conversation than three months in.
What we ask to look at
- Which systems hold the relevant data, and roughly how far back
- Whether outcomes are recorded - the single most important question
- A sample of real data, under an appropriate agreement
- How the decision is made today, and how well that works
- Where a prediction would have to appear for anyone to act on it
The sample matters most. Quoting without seeing data means pricing in contingency for problems that may not exist, or quoting optimistically and revising later. Neither serves you well.
A short assessment before any build
For anything beyond the smallest project we propose a fixed-price assessment first: a few weeks, a defined deliverable, and a decision point at the end.
- What data exists, its condition, and what is missing.
- How well your current method performs, measured properly.
- An early model on real data, with an honest statement of what it achieved.
- What that improvement would be worth, using your numbers.
- A recommendation - proceed, proceed differently, or stop.
That last line is not a formality. Recommending not to proceed is a normal outcome, and a supplier who never does it is not assessing anything.
What we will tell you plainly
| If we find | We will say |
|---|---|
| Outcomes are not recorded | Start recording; come back in a year |
| A rule would work as well | Build the rule, not a model |
| The improvement is real but small | It may not be worth the running cost |
| The data cannot be joined | That is the first project, not this one |
| It looks genuinely worthwhile | Here is the scope, cost and what it needs from you |
We have turned down projects on each of the first four grounds. It is a better outcome for both sides than a build that disappoints.
What we need from you
Access to real data early, someone in the business who will own the output, and a willingness to measure the current method honestly. Those three do more for a project's chances than anything we bring.
If you want to talk about a specific problem, use the form on this page - tell us what you are trying to predict and roughly what data you hold, and a senior engineer will come back with an honest view.
The most useful thing we can do in week one is tell you if this will not work.