What a two-week AI diagnostic actually covers

The problem with starting with a pilot

A pilot commits you to a solution before you have measured the problem. It is scoped around a technology someone wants to try, it runs for months, and its success criterion is usually that it worked, not that it paid back. The diagnostic exists to answer the prior question: of everything that could be automated or forecast in this operation, which one is worth doing first?

Week one: how work actually runs

We sit with the people who do the work. Planners, buyers, warehouse leads, the analyst who maintains the spreadsheet everyone depends on. We map the process as it runs, not as the procedure document describes it, and we measure it: how long each step takes, how often it is wrong, what the error costs. Most of the value of the diagnostic is in this week, because the map usually surprises the people who commissioned it.

We also look at the data. Not to build anything, but to answer whether the history exists to support each candidate: is there two years of sales at SKU-location level, are lead times recorded, are exceptions logged anywhere?

Week two: scoring and ranking

Each candidate gets scored on three things. Impact: what it is worth if it works, in stock, hours or service. Feasibility: whether the data and the process can support it now. Adoption: whether the people involved would actually use it, which is the criterion most AI projects fail on. The scores produce a ranked list. The top item gets a build plan: scope, data needed, how it lands in the existing process, and what it would take to run it after hand over.

What you receive

• A process map of how the work runs today, with measured time and error at each step.
• A ranked list of candidates with impact, feasibility and adoption scores, and the reasoning behind each.
• A build plan for the top item, including what it needs from your data and your people.
• A data readiness note: what exists, what is missing, what would need to be captured before the second item is possible.

How to judge whether it was worth it

The diagnostic is fixed-fee and the output is yours. If the ranked list tells you the top item is something you can do with a spreadsheet change and no AI at all, that is a good outcome, and it happens. If it tells you the data is not ready, that saves a six-month pilot. The test is whether you could hand the document to another firm and have them build from it. If not, it was a sales pitch, not a diagnostic.

What it is not

• It is not a technology assessment. We do not arrive with a platform to recommend.
• It is not a pilot. Nothing is built in two weeks. Building starts after, if the list says it should.
• It is not a strategy deck. The output is specific enough to act on the following Monday.