AI readiness checklist: a worksheet for intent, processes, data, approvals and measurement

Fifteen questions across five areas. Mark each one yes, partly or no with your leadership team in one sitting. The examples show what a yes looks like in a company of about 30 people.

Published 28 September 2026 by Attacca

This worksheet is for a conversation, not a score to publish. It works best when two or three people mark it separately and then compare, because the disagreements are the useful part. If you want a scored result with a recommendation, the AI readiness assessment does that in a few minutes. This page is the slower version you can print and argue over.

Count yes as 2, partly as 1 and no as 0. Nothing on this page is sent anywhere; the choices stay in your browser.

Intent

Whether the business knows what it wants AI to change, and what it will not accept.

  1. Can leadership name one business outcome AI should move this year?

    What yes looks like: The owner says “cut proposal turnaround from five days to two” rather than “use more AI”.

  2. Are the things AI must never do written down?

    What yes looks like: A one-page note says no client-facing message goes out unapproved and no pricing changes without a partner.

  3. Is one named person accountable for AI decisions?

    What yes looks like: The operations lead owns the list of AI experiments and can stop one without a meeting.

Processes

Whether the work you want to change is repeatable enough to describe.

  1. Are your three most frequent workflows written down step by step?

    What yes looks like: Client onboarding, weekly reporting and invoicing each have a one-page process map someone updated this quarter.

  2. Do you know where work gets dropped between people?

    What yes looks like: You can say that meeting follow-ups are the leak, and roughly how many a month go missing.

  3. Do you know how often each candidate workflow runs?

    What yes looks like: Weekly reports: 18 a week. New-client setups: two a month. The numbers come from the task tool, not a guess.

Data

Whether the information an agent would need exists, in one place, in a version people trust.

  1. Does each kind of document have one home?

    What yes looks like: Briefs live in one folder per client; nobody keeps a private copy on a laptop.

  2. Are the facts you repeat to clients written down once, with a source?

    What yes looks like: Each client has a short facts file: contacts, what you sell them, the numbers in every report, dated.

  3. Can you grant read-only access to that information and revoke it in one step?

    What yes looks like: The shared drive sits under a company account, and access for a tool is a single permission you control.

Approvals

Whether a person checks AI output where it matters, and whether that check is designed or accidental.

  1. Is it decided which AI outputs a person must approve before they leave the building?

    What yes looks like: Drafts to clients and changes to the task board need a named approver; internal summaries do not.

  2. Have you ranked workflows by what a mistake would cost?

    What yes looks like: A wrong internal note is an annoyance; a wrong invoice or a wrong claim to a client is not. The list says which is which.

  3. Would you know afterwards who approved an AI-assisted output?

    What yes looks like: Every approved draft carries the approver’s name and the date, in the same place as the work.

Measurement

Whether you would know if it worked.

  1. Do you have a baseline for the workflow you want to change?

    What yes looks like: Before starting, you counted it: of 13 actions agreed in last month’s client reviews, 3 reached the task board.

  2. Is there a test you could run on AI output with known answers?

    What yes looks like: Ten questions whose answers you already know, checked against the source file each time the setup changes.

  3. Is there a date to review results and decide whether to continue?

    What yes looks like: A 30-day review is on the calendar, with the baseline, the new count and the caveats on one page.

Reading your total

The maximum is 30. The total matters less than where the zeros are, but as a rough guide:

  • 24 to 30. You are ready to pick a first workflow and run it with a baseline. Our guide to choosing your first AI workflow is the next step.
  • 14 to 23. The usual picture. Fix the lowest area first; for most small companies that is data, and the Google Drive knowledge base guide covers it.
  • Under 14. Start with intent. Tools bought before anyone agrees what they are for tend to be abandoned, which one agency we worked with learned across three tools in a row.

Two patterns worth noticing

A yes on intent with a no on measurement means the company knows what it wants but will not be able to tell whether it got it. Write the baseline before anything else. A yes on data with a no on approvals is the riskier combination: the information is ready for an agent, and nobody has decided where a person must check its work. Chapter 4 of our book, Trust as Architecture, explains how to match the amount of checking to what a mistake would cost.

When you do not need this

If you already run one AI-assisted workflow with a baseline, a named approver and a review date, you do not need this worksheet; you have answered it in practice. The same goes for a team of two or three people who talk every day: the questions still apply, but you can answer them over lunch.