What this AI adoption audit measures
The AI Demonstration Gap is the gap between the AI capability an organisation declares and the capability its people can demonstrate in real work. This audit makes that gap visible one workflow at a time.
It does not ask whether the organisation has an AI strategy, a data platform or enough licences. Those questions belong in a readiness review. It asks whether a named task improves, whether the operator captures enough benefit to return, and whether the claim is backed by repeated, accepted work.
| Assessment | Unit of analysis | Main question | Typical output |
|---|---|---|---|
| AI-readiness assessment | Organisation or department | Are the conditions for AI in place? | Readiness profile across strategy, data, skills, infrastructure and governance |
| AI Demonstration Gap Audit | One bounded workflow | Can the team demonstrate changed work that repeats? | Quadrant diagnosis, evidence stage and next operating decision |
Use both when they help. Do not let a readiness score stand in for demonstrated adoption.
The two scoring axes
The scorecard uses two 1-to-5 working scales. They are prompts for an evidence review, not validated benchmarks.
Task value from AI
| Score | Working interpretation | Evidence to look for |
|---|---|---|
| 1 | No improvement, or the task becomes worse | Lower quality, slower cycle, higher risk or more cost |
| 2 | Weak, inconsistent or unclear value | A promising output with no stable advantage |
| 3 | Useful in some cases | A repeatable improvement appears, but conditions or comparison remain incomplete |
| 4 | Clear improvement | Better accepted work, cycle time, rework, risk, cost or decision quality |
| 5 | Strong measured improvement | The result survives comparison, review and the relevant control boundary |
User surplus
| Score | Working interpretation | Evidence to look for |
|---|---|---|
| 1 | The workflow makes the operator's job worse | More review, less autonomy, more anxiety or no usable gain |
| 2 | The burden exceeds the benefit | Use requires pressure, hidden coaching or repeated workarounds |
| 3 | Useful, but friction remains | The operator returns in some conditions and can name what still gets in the way |
| 4 | Clear personal benefit | The operator returns because the workflow saves effort or improves the work |
| 5 | Strong voluntary repeat use | People return without prompting and would miss the method if it disappeared |
For this worksheet, scores of 3 to 5 map to the high side of each axis and scores of 1 to 2 map to the low side. That threshold is a working heuristic. Keep the observations, comparison and limitations beside the number.
How to run an AI Demonstration Gap audit
- Name one bounded workflow. Choose a repeated task with a clear trigger, operator, output and reviewer.
- Score task value. Rate whether AI improves the task itself, using observed quality, speed, cost, risk or decision evidence.
- Score user surplus. Rate whether the person doing the work captures enough benefit to return without being pushed.
- Record the evidence stage. Separate a declaration or one-off test from repeated use and a measured outcome.
- Act on the diagnosis. Scale, redesign, test further or leave the workflow alone according to the quadrant and evidence gap.
The interactive tool assigns a provisional quadrant when the evidence is only declared or tested once. A provisional result is a hypothesis to test, not an adoption claim.
The four workflow diagnoses
| Diagnosis | Task value | User surplus | Operating response |
|---|---|---|---|
| Compounding Adoption | High | High | Protect the controls, save the method and test independent reuse |
| The Willingness Gap | High | Low | Redesign review burden, incentives, autonomy or workflow friction |
| AI Theatre | Low | High | Stop counting activity as value and run a real task comparison |
| Correctly Left Alone | Low | Low | Leave the task alone and spend attention elsewhere |
The quadrant tells you the likely failure mode. The evidence stage tells you how seriously to take the diagnosis.
The evidence stages
- Declared only: a strategy, licence, training session or claimed use case exists, but no bounded task has been shown.
- Tested once: someone produced an example, pilot output or demonstration, but repeated operating behaviour is not established.
- Repeated in real work: the workflow has run more than once in its intended setting and produced work that met the acceptance standard.
- Measured outcome: repeated use is connected to an operating measure, comparison and control record with limitations stated.
Do not average these stages into the two axis scores. A workflow can look valuable in a demonstration and still have no evidence that it travels.
What to do after the audit
Take the smallest defensible next decision.
- Scale only when accepted use repeats, the task improves and the gain remains inside the control boundary.
- Redesign when the task has value but the operator carries too much friction, review burden or personal risk.
- Test further when the proposed value is plausible but the comparison, sample or evidence stage is too weak.
- Stop or leave alone when the task does not improve or the control burden erases the gain.
For a fuller measurement contract, use the guide to measure AI adoption in investment and advisory teams.
Frequently asked questions
What is an AI adoption audit?
An AI adoption audit checks whether AI has changed a defined workflow in repeated, acceptable and measurable ways. It separates access and experimentation from demonstrated capability in real work.
What is the AI Demonstration Gap?
James Kerr's AI Demonstration Gap is the gap between the AI capability an organisation declares and the capability its people can demonstrate in real work.
How is this different from an AI-readiness assessment?
A readiness assessment usually examines broad conditions such as strategy, data, infrastructure, skills and governance. This audit starts with a specific workflow and tests task value, user surplus and the strength of the operating evidence.
Is the result a validated maturity score?
No. The 1-to-5 inputs and working threshold are a practical diagnostic inside James Kerr's framework, not an externally validated benchmark or certification.
Does the audit store company data?
No. The interactive audit runs in the browser. Entries are not sent to James Kerr or CURN; users can download their own CSV or print the page locally.
Method and sources
This audit operationalises James Kerr's AI Demonstration Gap and the workflow-level evidence stack in How to measure AI adoption in investment and advisory teams.
The OECD SME AI Readiness Tool is a useful example of a broader readiness assessment. The NIST AI Risk Management Framework supports the continuing discipline of measuring and managing AI across its lifecycle. Neither source validates this scorecard's thresholds or quadrants.
The worksheet is deliberately not a certification, maturity model or customer-proof claim. Its purpose is to make the next evidence gap and operating decision visible.