The Pilot Before the Platform

The most common opening line on our calls is some version of "we're evaluating platforms." Our most common reply is: don't — not yet. Buy a pilot first, sized so that failing is cheap and succeeding is measurable. Here's why we hold that line even when it costs us bigger contracts.

Platforms answer a question you haven't asked yet

A platform decision is really a bundle of decisions: which workflows to automate, how to route exceptions, what your team reviews, what your customers see. Making all of them at once — before AI has touched a single real ticket — means making most of them wrong. The pilot exists to convert those guesses into observations, one workflow at a time.

What a pilot has to prove

A pilot is not a demo. A demo proves the software runs; a pilot proves the workflow improves. To do that it needs three things most evaluations skip:

• A baseline written down before anything launches
• One workflow, narrow enough to instrument end to end
• A kill criterion — the number that, if unmet, ends the project

That last one matters most. Teams that can't say what failure looks like will call anything a success, and then wonder a year later why the numbers never moved.

If failing the pilot wouldn't be cheap, it isn't a pilot — it's a rollout wearing a pilot's badge.

Six weeks is not arbitrary

Shorter, and you're measuring novelty: the first fortnight of any new tool is unrepresentative in both directions. Longer, and you've built too much to walk away from — sunk cost starts voting in the retrospective. Six weeks gives you two weeks of shakedown and a full month of steady-state data, which is enough to see the pattern without falling in love with the project.

The failure it protects you from

The expensive failure mode isn't AI that doesn't work. You notice that quickly and shut it off. The expensive one is AI that almost works — impressive in the demo, wrong just often enough that your team quietly re-checks everything it does. On a platform contract you discover that in month nine. On a pilot you discover it in week four, with a baseline that proves it, and you've spent a fraction as much to learn the same lesson.

And when the pilot wins

Then buy the platform — genuinely. At that point you know which workflow earns its keep, what review costs, and what number the next workflow has to beat. The platform stops being a leap of faith and becomes a purchase order for something you've already watched work.