Home › Blog › Complete AX transformation guide

The complete AX transformation guide — from assessment to rollout

AX transformation is a question of sequence, not tooling. Skip the sequence and even a strong tool stalls at the pilot; keep it and an ordinary tool produces results. Here are the four stages.

All 4 stages Failure points Links to the cluster

The short answer

AI transformation in one sentence: find the repetitive judgement and work that can move to a machine, move it, and spend the recovered time elsewhere. It is closer to process redesign than to a technology project.

Which is why the causes of failure are rarely technical. Misjudged readiness, no prioritisation, no named owner, no measurement — those four repeat. The stages below block them in order.

Stage 1 — Assessment: where you actually are

The purpose of the first stage is not to make a plan but to turn the current state into numbers. Across work, data, technology and organisation, establish what is ready and what is missing.

The most common misjudgement here is "we are not ready, so later". Most organisations start below half, and that is normal. What matters is not the total but which axis is lowest — that axis is the starting point.

The assessment must produce numbers: monthly volumes, minutes per item, people assigned. Without them you cannot prove results at any later stage.

Stage 2 — Prioritisation: what goes first

Frequency times duration — The simplest criterion and the one that fits best. A hundred times a month at five minutes usually beats twice a month at four hours.

Is it reversible? — Make the first project one you can recover from. Attempt the first on irreversible work and the organisation will not authorise a second.

Does the data already exist? — Anything requiring a data clean-up first is a poor first project — results take too long to appear.

Can an owner be named? — Do not start work nobody’s name is on. Projects run on "somebody will pick it up" stall without exception.

Stage 3 — Pilot: small, and measured

The purpose of a pilot is learning, not success. So keep the scope small and fix the measurements in advance: what improvement justifies rollout, and what result stops it.

Writing the stop condition down first matters most. Without it, a pilot with no result gets pushed indefinitely, and that cost is what blocks the organisation’s next attempt.

One more thing the pilot must record: review time. Leave out the time people spend checking AI output and the savings are overstated, and the arithmetic breaks after rollout.

Stage 4 — Rollout: a different problem from the pilot

Policy and data classes — What one team did not need becomes mandatory company-wide. Define which data may be entered, as classes.

Access control — SSO, audit logs, permission separation — usually top-tier only, so budget for it in advance.

Training and a request route — With nowhere to request a new tool, people will not request one; they will use a personal account.

Running cost — Usage charges rise with headcount. Pilot cost times a headcount multiplier usually underestimates it.

Keeping measurement alive — Keep watching the same metrics after rollout. When measurement stops, the results lose their evidence.

The rest of this cluster — reading on by stage

Stage 1, assessment — The twelve AI readiness checks · Running an AX assessment yourself · How to calculate AI ROI

Stage 2, prioritisation — Six things failed AI adoptions have in common · Why AI projects stall at the pilot: the data

Stage 3, pilot — Where to start with process automation · How meeting-note time was actually reduced

Stage 4, rollout — The AI tool security checklist · Writing an internal AI usage policy · The costs not on the price page

Funding — Korean SME AI support programmes — how to find them and be ready

Frequently asked questions

Do we have to go through all four stages?

You can compress a stage, but reordering them usually fails — buying a tool before assessing is the most common failure path. An assessment can be done in a day, so compress rather than skip.

How long does it take?

It depends on size and scope. Keep the run-up to the pilot short, though: the longer it takes, the sooner organisational attention runs out. Two to three weeks for assessment and prioritisation and four to eight for the pilot is a common shape.

Where do we need outside help?

Most of the assessment can be done internally. Outside help becomes necessary when a judgement needs technical grounding, and when priorities collide between teams.

What do we need for executive approval?

Current cost, target numbers, and the break-even point. "Worth it eventually" does not pass; "in profit from month seven" usually does.

Start at stage 1

Begin with the assessment

The AX readiness assessment produces your four axis scores and your automation priorities. Free — and if you need the formal version, tell us your situation.

Check readiness