AI Maturity Framework: Assess Your Team’s Next Step

An AI maturity framework is a structured way to assess whether a team can use AI reliably, with clear ownership, repeatable work, and appropriate review. The worksheet below assesses one workflow at a time, then turns the most important gap into a change you can check.

A shared record such as fenbs can preserve the assessment, the improvement people choose, and the lessons from the next trial. This is an independent team worksheet, not a vendor maturity model, certification, or prediction of business performance.

What should you assess before assigning a score?

Choose a workflow with a recognizable beginning and end. “Our team uses AI” is too broad. “An assistant drafts a client update from approved task status, and the account lead checks it before sending” gives you something observable. Include the steps before and after generation, since mistakes often appear at those boundaries.

Agree on the purpose, the people affected, and the consequence of failure. A rough brainstorming list and a customer-facing recommendation should not share an acceptance test. Record the approved tools and information sources, plus the activity you want to improve. Your baseline might be review effort, recurring correction types, or the ability of another teammate to repeat the process.

The NIST AI Risk Management Framework↗ (opens in a new tab) is a voluntary framework for managing AI risks. It is useful background, rather than the scoring method for this worksheet. Our NIST framework guide explains that separate structure, while the AI governance library covers the surrounding responsibility questions.

Which dimensions belong in a practical worksheet?

Assess the conditions that make the work usable without a private explanation from its original operator. Use the same dimensions at the next review so you can compare evidence. You can add a dimension for a specific risk, but keep the starting set small enough to discuss with the people who actually do the work.

For the continuity dimension, distinguish an archive from a usable starting point. A long chat transcript may contain the answer somewhere while still leaving the next person uncertain. The guide to AI memory for teams explains what belongs in a shared work record.

How can you rate progress without false precision?

Use a short descriptive scale and require an example for each rating. The scale below is a suggested local convention, not an external standard. “Unknown” is a valid answer when nobody has checked the evidence. Do not silently turn it into a positive rating.

• Unclear: the team cannot identify an agreed process or evidence for this dimension.
• Person-dependent: the process works when a particular person remembers the necessary steps.
• Repeatable: another person has followed the documented steps and produced a reviewed result.
• Maintained: the team checks the process after relevant changes and updates it from observed failures.

Keep the dimensions separate instead of averaging them into a flattering total. Strong documentation does not cancel excessive access. Good outputs do not make missing ownership disappear. Identify which gap could invalidate the workflow and address it before using a general score to justify expansion.

Ask two people to rate the same example independently, then compare their reasons. Disagreement often reveals an unstated assumption or missing evidence. The NIST AI RMF Playbook↗ (opens in a new tab) similarly offers suggestions to select for your circumstances, rather than a mandatory checklist to complete in full.

Keep the discussion about the workflow rather than grading colleagues. A person-dependent process often reflects missing time, access, or instructions. Ask what would let another teammate perform the step, then use that answer to define support. The assessment should reveal a repairable condition, not create an incentive to hide weak results.

What does a completed assessment look like?

Say a three-person design studio uses AI to draft project updates. Its drafts read well, but only the account lead knows which client promises are approved. Another teammate can run the prompt yet cannot confidently review the result. This hypothetical assessment describes that dependency rather than claiming the whole studio has reached a universal maturity level.

• Workflow: draft a client update using current task status; a person reviews before sending.
• Ownership: repeatable. The account lead and alternate are named in the instructions.
• Inputs: person-dependent. Approved promises are scattered across private conversations.
• Verification: person-dependent. Review depends on what the account lead remembers.
• Improvement decision: record approved commitments and require the reviewer to compare the draft against them.
• Success check: the alternate reviews a new draft, finds an intentionally incorrect commitment, and explains the correction using the approved source.

A single successful check would support the specific change, not prove that all future drafts are safe. Keep the scope of the conclusion beside the result. In fenbs, the improvement decision and the resulting lesson can stay with the project for the next assistant to read.

How do you choose and record the next improvement?

Choose a change small enough to complete and important enough to affect the next run. Name its owner, the work it covers, the evidence required, and what happens if the check fails. “Improve governance” is an aspiration. “Confirm that only the reviewer can send the draft” is a checkable task.

Workflow and purpose:
People affected:
Approved tools and inputs:
Dimension and rating:
Evidence examined:
Unknowns or disagreements:
Improvement chosen by:
Owner and next action:
Success check:
Review trigger:
Lesson after the next trial:

Review again when the tool, input, purpose, or possible consequence changes. Keep the previous assessment so you can explain why a rating moved. If your next question is how to introduce AI across the team, use the separate AI adoption guide; this worksheet stays focused on assessing the work you can observe.

How to do this with fenbs

1. Create a free account and add a project for the workflow you are assessing.
2. Add an assessment task using the worksheet. Record the ratings, the evidence reviewed, and what remains unknown in its note and plan.
3. Record the improvement a person chooses in Decisions and rules, then add the follow-up task and its success check.
4. After the trial, save the test result, add the lesson learned, and update Where we left off with the next review step.

fenbs does not calculate a maturity score or independently verify your evidence. Use a specialist assessment or technical testing tool when that is the job you need. For this worksheet, the useful result is a specific improvement and a record the next reviewer can understand.