Ask for a ranked list, not an AI strategy

The deck problem

Most AI engagements start with a strategy phase. Four weeks later you have a document describing your industry, a maturity model with you somewhere in the middle, and a roadmap with three horizons. None of it tells you which process to change first.

The problem is not that strategy is worthless. It is that the strategy cannot be written until you know which tasks are actually automatable, and nobody can know that from the outside without looking at the work.

What to ask for instead

Ask for a ranked list. Every recurring task in the part of the business you care about, scored on four things:

• Volume. How many times a month does this happen?
• Time per instance. Measured, not estimated by the person who does it.
• How rule-based it is. Could a competent new starter do it from a written procedure?
• Cost of a mistake. This is what decides how much autonomy the agent gets.

Multiply the first two and you have the size of the prize. The third tells you whether it is buildable this year. The fourth tells you what the guardrails have to look like.

The uncomfortable output

A ranked list usually reveals that the interesting use case is fourth and the boring one is first. Invoice processing beats the customer-facing assistant almost every time, because volume is high, the rules are written down somewhere, and a mistake is caught by a reconciliation nobody skips.

It also reveals the ones that should not be built. Expect a good proportion of a first ranked list to fail the twelve-month payback test — usually because the data would have to be cleaned first and nobody has budgeted for that.

If a use case cannot pay back inside twelve months, park it. Writing that down at the start saves an argument at the end.

What it should cost

Two to three weeks and a fixed price. If a ranked list of use cases takes longer than that, the engagement has turned into a strategy project again.

---

Written by the OneRee team. If any of this is wrong for your situation, tell us — we would rather update the page than be right on the internet.