Interactive tool · About 3 minutes

How ready is your AI pilot to scale?

Most pilots stall for organisational reasons, not technical ones. Answer twelve questions to see where yours stands in each phase, the gaps to close first and the right level of human oversight for its outputs.

  1. Before the pilot

    Decide what the AI will do in production and what that will cost.

    • Job definition
    • Use case
    • Data path
    • Cost model
  2. During the pilot

    Prove production value under production conditions.

    • Commercial measures
    • Oversight tiers
    • Real conditions
  3. Into production

    Give it an owner, a dated plan and a review cycle.

    • Single owner
    • Transition plan
    • Ongoing review
Part 1

Readiness check

Answer for the pilot as it stands today. Partly is a fair answer when a decision is made but not written down.

0 of 12 answered

Before the pilot

Decide what the AI will do in production and what that will cost.

  1. Can you state the user, task, output and a measurable quality threshold in one paragraph?
  2. Does the data already exist, does the output feed a workflow people use today, and will value show within 90 days?
  3. Is every production input traced to a source, a maintainer and a refresh schedule?
  4. Do you know what it will cost to run for a year at production volume, and what return justifies that?

During the pilot

Prove production value under production conditions.

  1. Are you measuring adoption and the commercial outcome, not just accuracy?
  2. Does each type of output have an agreed level of human review, with the reason written down?
  3. Is the pilot running with real users, live data and the awkward cases included?
  4. Have you agreed what result would make you stop?

Into production

Give it an owner, a dated plan and a review cycle.

  1. Is one named person accountable for the system's cost, quality and business result?
  2. Does each go-live decision have a date and an owner?
  3. Are support, a fallback and the switch-off of the old process planned?
  4. Is there a fixed cycle for reviewing quality, oversight and cost after launch?
Answer 12 more to see your result.
Part 2

Oversight tier finder

Pick one output your AI produces, such as a summary, a letter or a decision, and answer four questions about it. Repeat for each type of output.

  1. Is the output a judgement a qualified person is accountable for, such as credit, legal or clinical?
    If so, the AI should advise rather than decide.
  2. Does the output reach a customer, move money or affect someone's rights?
    External effects need approval before release.
  3. Could a regulator, auditor or court ask how it was produced?
    You will need a record of who approved what.
  4. Can a mistake be spotted and reversed before it causes harm?
    Reversible, internal outputs can often run automatically.

Answer all four questions to see the recommended tier. Until then, explore the ladder below.

Tier 3 of 4Approved before release

A named reviewer approves each output before it takes effect. The record shows who approved what, on which evidence.

Typical outputs
  • Client-facing letters
  • Payments or decisions above a limit
  • Anything a regulator may ask about
How review works

Every case, with turnaround tracked as a service level.

The guide

How to run an AI pilot that actually makes it to production

The reasoning behind every question, with examples, the full oversight model and the decisions to make before you scale.

Read the guide