Insights
How do you know whether employees can work effectively with AI?
Updated September 2026
You know by observing their decisions and the business results those decisions move, not by counting logins or course completions. Usage tells you people have AI. Capability tells you they use it well: they catch its errors, override it when needed, and turn the time it saves into better work. Measure at five levels, from usage up to business results, and remeasure on a schedule so you can prove movement.
Usage is up. Results are not keeping pace.
- 74% of frontline employees are now regular AI users, yet 66% get limited or no guidance on what to do with the time it saves (BCG, 2026).
- 80% of respondents say AI has improved their personal productivity, but only 37% report any enterprise-level EBIT impact (McKinsey, 2026).
- AI high performers are twice as likely as others to have defined processes to measure the impact of their AI initiatives (McKinsey, 2026).
The gap between individual use and organizational results is a measurement problem before it is a technology problem.
Five levels of evidence
| Level | Question it answers | What to measure |
|---|---|---|
| 1. Usage | Are people using AI? | Active users and frequency. Necessary, never sufficient. |
| 2. Stated readiness | How do people say they approach AI? | A short behavioral profile such as n.print: a directional baseline |
| 3. Observed decisions | How do they actually decide with AI? | Error catch rate, override accuracy, and decision quality in realistic scenarios |
| 4. Team behavior | Does it hold up across the team? | Team composition gaps and how aligned decisions are under pressure |
| 5. Business results | Did the work outcome move? | One business metric per workflow, set before AI arrives: cycle time, cost per case, error rate, or revenue per rep |
Most organizations measure level 1 and assume level 5. The levels in between explain why results do or do not follow.
Metrics that mislead
- Seats provisioned and prompts sent measure access, not ability.
- Training completions measure attendance, not judgment.
- Time saved without a plan for that time measures leakage.
Metrics that matter
- Error catch rate: the share of AI errors people catch before they reach a customer or record.
- Override accuracy: how often people are right when they overrule the AI.
- Decision alignment: how consistently a team reads the same AI-supported decision.
- Reinvested time: hours saved that were redirected to named, higher-value work.
- The business metric: the one number each AI-supported workflow was meant to move.
What our field data shows
Self-reported confidence tells you where people think they are, not what they can do. In our field data, action confidence (3.49 out of 5) trails mindset (4.11) for every PACE type, and colleagues agree on their organization's AI phase only about half the time. Both gaps are reasons to measure observed decisions, not opinions. These figures are what people report, which is why capability is best judged in real decisions, not surveys alone.
Source: Neurocollective n.print field data, 464 respondents from 182 email domains, as of September 2026. Self-reported. No single organization exceeds about 30% of responses.
Where this fits in the discipline
AI Adoption Engineering, developed by Dr. Lisa Palmer and Neurocollective, treats measurement as a system, not a survey. Its diagnostics follow the Bold AI Leadership Model: n.print measures individual mindset, n.team shows how individuals work together, n.score measures organizational readiness across the Five AI Success Pillars, and n.action shows momentum on the AI Performance Flywheel. n.sim gives people realistic decisions to practice. Together they compose one readiness picture, and remeasurement proves movement.
Related answers: Self-reported vs. observed AI readiness · How do organizations create measurable business value from AI? · What skills do people need to supervise, validate, and override AI?
Get a baseline you can remeasure.
n.diagnose packages four connected diagnostics and one executive readout into a single fixed-scope engagement.
Questions leaders ask
How should companies measure AI adoption?
Measure five levels: usage, stated readiness, observed decisions, team behavior, and business results. Usage alone shows access, not capability.
What is a good metric for AI capability?
Error catch rate and override accuracy in realistic scenarios. Both show whether people can judge AI output, which is what turns AI use into results.
How often should AI capability be remeasured?
Every 60 to 90 days during an active program, so movement can be tied to specific changes in training, decision rights, or team composition.
Dr. Lisa Palmer
Founder of Neurocollective. Her doctorate produced the IRB-approved research behind AI Adoption Engineering, validated by external reviewers, and brought to life in Show AI—Don't Tell It: Build Buy-In with Visual Storytelling (Wiley, 2026). drlisa.ai