Measure AI’s effect against a defined baseline at the task or workflow level—not by adoption rates alone. Track who uses the tool separately from what changes in output, time, quality, rework, customer or stakeholder results, and worker experience. When feasible, compare teams or rollout periods using randomized access or a phased introduction; keep monitoring after launch.
Start by defining what “better performance” means
Choose a specific task or workflow that actually uses the AI system, then state the expected improvement in observable terms. For example: “reduce minutes per completed case without lowering resolution quality” or “increase accepted drafts per week without increasing rework.” “AI adoption improved productivity” is too vague to test.
Pick measures that fit the work and the decision you need to make. NIST’s AI Risk Management Framework says measurement methods should reflect the AI system’s context, with metrics and conditions documented.
Separate exposure and usage from results
Record who was eligible for the tool, who received access, how often they used it, and for which tasks. Where possible, note whether AI output was accepted, edited, or discarded. These are measures of exposure and use—not proof of better performance. Low use may help explain a limited effect; high use can coexist with neutral or negative results.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minute#1 Best Overall
That distinction matters in practice: a randomized six-month experiment across 66 firms and 7,137 knowledge workers found that frequent users spent less time on email, but researchers detected no shift in task quantity or composition from individual access. The study shows why time saved and work delivered should be measured separately.
Build a balanced scorecard for the work
Pair speed or volume with quality and value. Select measures tied to the task’s intended benefit and risks; there is no universal metric set that fits every team.
| Dimension | Possible measures |
|---|---|
| Throughput and time | Completed tasks, cases resolved, accepted deliverables, or cycle time |
| Quality | Accuracy, first-pass acceptance, error rate, escalations, rework, or defect severity |
| Customer or stakeholder value | Satisfaction, issue resolution, adoption of a recommendation, or a relevant downstream outcome |
| Worker effects | Workload, worker experience, learning, retention, and how gains are distributed |
| Risk and oversight | Privacy, security, safety, fairness, reliability, and human review or override rates where relevant |
Set the definitions before comparing results. For example, specify what counts as a completed task, a quality failure, or an accepted deliverable. Keep the measurement method consistent before and after rollout, and record conditions that could affect results. NIST’s AI measurement and evaluation guidance emphasizes context-specific metrics, documented methods, benchmarks, uncertainty, and monitoring in production.
Choose a comparison that can support attribution
A simple before-and-after comparison can show that a metric changed, but not necessarily that AI caused the change. Workload, staffing, seasonality, process changes, or a different AI version may also explain it. Use the strongest practical comparison and document the benchmark, sample, time window, deployment conditions, and uncertainty.
Rank #3
| Approach | How it works | What to watch |
|---|---|---|
| Randomized access or timing | Randomly assign eligible workers or teams to receive access earlier or later, then compare outcomes. | Usually gives the clearest comparison when practical; track whether assigned users actually use the tool. |
| Phased rollout | Introduce the tool to groups at different times and compare groups not yet using it with those already using it. | Record changes in workload, staffing, process, and tool version across rollout periods. |
| Matched comparison | Compare the adopting team or task with a similar one that has not adopted the tool. | Differences between groups can complicate attribution; document how they were selected and what differs. |
| Before and after | Compare the same measures for a team or task before rollout and afterward. | Useful for tracking change, but other changes over time may account for it. |
Two studies illustrate credible designs without supplying a forecast for other organizations. A staggered rollout among 5,179 customer-support agents was used in the support-agent study; a randomized field experiment was used in the cross-industry knowledge-worker study. Their estimates are specific to the settings, tools, populations, and periods studied.
Examine results by task and worker group
A team average can hide meaningful differences. Where sample size and privacy permit, break results out by task type, experience, skill, or other relevant groups. This can reveal who benefits, who needs more training, whether certain tasks are poorly suited to the tool, or whether work has shifted elsewhere.
In a study of 5,179 customer-support agents, researchers reported a 14% average increase in issues resolved per hour, with a 34% increase for novice and lower-skilled agents and minimal impact for experienced and highly skilled agents. These are findings from one company’s tool and work setting, not a general productivity estimate. The paper was published as an NBER working paper in 2023, revised that year, and later published in the Quarterly Journal of Economics in 2025.
Teamwork results can differ from individual-task results, too. In a preregistered 2025 field experiment at Procter & Gamble involving 776 professionals working on product innovation challenges, individuals using AI matched the performance of teams without AI. That finding concerns a particular creative collaboration setting, not all team tasks. Read the study.
Recommended Free Tools
Best Value
Reassess performance after launch
A pilot may miss learning, adaptation, or work reorganization. NIST’s AI RMF Measure function states: “AI systems should be tested before their deployment and regularly while in operation.” It also calls for documented test sets and metrics, benchmarks, uncertainty measures, and production monitoring. NIST AI RMF Core — Measure.
Repeat the core measures in production and compare live behavior with the baseline and expectations. Track task mix, quality, usage, overrides, user feedback, and incidents. Decide in advance what results would lead you to adjust the workflow, add review, or roll back the tool.
Keep organizational findings in perspective, too. A Denmark study revised in March 2026 estimated no effects larger than 2% on earnings or recorded hours two years after ChatGPT’s launch, while documenting task reorganization and occupational transitions. Those aggregate labor-market measures do not rule out local task-level benefits or costs. See the study.
What the evidence can—and cannot—tell your team
The studies measure different outcomes at different levels: support cases per hour, email time and task mix, performance on innovation challenges, and labor-market earnings or recorded hours. Their findings do not establish one universal AI productivity effect. A local evaluation should therefore focus on the work your team is changing, the outcomes that matter for that work, and whether gains persist without unacceptable tradeoffs in quality, customer value, or worker experience.
Free tools Windows power users keep installed
One-click scans. No signup required.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
