The Deployment Ladder: Five Rungs from Pilot to Production

Organisations do not jump from pilot to production. They climb, and most fall off the same rung.

The Deployment Ladder came out of a repeated observation: the companies that got AI into daily operations did not have better models than the ones that did not. They had a sequence. Each rung produced something the next rung needed, and nobody was allowed to skip.

Rung one — Evidence

Before any tooling decision, write down the baseline: how long the workflow takes today, how often it is wrong today, what it costs today. Teams resist this because the baseline is usually embarrassing and always contested. Without it, every later claim about improvement is opinion.

Rung two — Ownership

One named person in the line function, not the innovation team, owns the outcome. The test is simple: if the system produces a wrong answer overnight, whose phone rings? If nobody's does, the programme is not a deployment.

Rung three — Evaluation

Define what good looks like before the system is built, in a form somebody can adjudicate. Three things need to exist on paper:

• A test set drawn from real cases, including the awkward ones
• A threshold that releases the workflow to production
• A named adjudicator when the system and a human disagree

Rung four — Integration

The workflow moves inside the systems people already use, with the old path switched off. Parallel running feels prudent and is the most reliable way to guarantee a permanent pilot: as long as the manual route exists, it will be used under pressure.

A deployment is not finished when the system works. It is finished when the alternative is gone.

Rung five — Standardisation

Only now does the group standard get written, and it gets written from something that already runs: the evaluation format, the ownership model, the release gate. Groups that write the standard first end up with a policy nobody can satisfy and a portfolio of demos.

Where to start

Pick the workflow where the baseline is worst and the owner is willing. Ambition is not the scarce resource. A named owner and an honest baseline are.