Did AI improve productivity, or just move the work? A measurement guide
A practical guide for workers and managers to measure whether AI truly boosts productivity or simply shifts work to review, correction, and escalation. Covers time, quality, rework, and human oversight.
Many organisations have adopted generative AI tools with the hope of saving time and boosting output. Yet early enthusiasm can mask a hidden cost: the work does not disappear, it simply moves. Tasks such as reviewing AI-generated drafts, correcting errors, escalating ambiguous outputs, and managing quality assurance can consume the time that was supposedly saved. This article provides a practical framework for workers and managers to measure whether AI is genuinely improving productivity or merely redistributing effort.
Why first-draft speed is misleading
The most common metric for AI productivity is the time taken to produce a first draft. A worker who can generate a report, email, or code snippet in seconds feels faster. But the full cycle includes reading, editing, fact-checking, and approving that output. If the AI produces content that requires heavy correction, the net time saving may be small or negative.
For example, a marketing assistant might generate a social media post in 30 seconds, but then spend 10 minutes correcting tone, verifying brand guidelines, and checking for factual accuracy. The total time may exceed writing from scratch. To get a true picture, measure the entire workflow, not just the generation phase.
Measuring review and correction time
To assess real productivity, track the time spent on each task stage separately. Use a simple log for one week:
- Generation time: minutes spent prompting and receiving AI output.
- Review time: minutes spent reading, checking, and editing the output.
- Correction time: minutes spent fixing errors, omissions, or formatting issues.
- Escalation time: minutes spent asking a colleague or manager for clarification or approval.
Compare these totals to the time taken to complete the same task manually. If review and correction exceed the manual baseline, AI is not improving productivity for that task.
Tracking rework and quality defects
Rework is a hidden productivity killer. An AI-generated document that passes initial review may still contain subtle errors that surface later, requiring a second round of edits. Track the number of revisions per deliverable and the frequency of quality issues flagged by stakeholders.
Use a simple quality scorecard:
- Accuracy: number of factual errors found after delivery.
- Completeness: percentage of required elements present.
- Consistency: adherence to style guides or standards.
- User satisfaction: feedback from recipients or approvers.
If AI-assisted work shows higher defect rates, the apparent speed gain is offset by downstream correction costs. The International Labour Organization notes in its 2025 update that most jobs will be transformed rather than made redundant, and that managing the transition through social dialogue is essential to enhance both working conditions and productivity. This underscores the need for transparent measurement.
Comparing results by task type
Not all tasks benefit equally from AI. To identify where AI truly helps, compare productivity metrics across different categories:
- Routine writing: internal memos, standard emails, status updates.
- Creative work: marketing copy, social media content, presentation design.
- Analytical tasks: data summaries, report generation, code debugging.
- Client-facing communication: proposals, responses, documentation.
For each category, record the total time from start to final approval, plus the number of corrections and escalations. A task type that consistently requires more than 20% correction time may be better done manually or with a different tool setup.
Human oversight as a necessary cost
The NIST AI Risk Management Framework Playbook emphasises the importance of governance, mapping, measuring, and managing AI systems. In practice, this means that human review is not a bug but a feature. Workers must be empowered to question AI outputs, and managers must budget time for oversight.
If your organisation treats review time as wasted time, it will underestimate the true cost of AI. Instead, build review into the standard workflow and measure it as a core productivity metric. A useful benchmark is the ratio of review time to generation time. A ratio above 3:1 suggests that the AI is creating more work than it saves.
A practical measurement checklist
Use this checklist to run your own productivity audit:
- Select three to five common tasks that involve AI assistance.
- For one week, log generation, review, correction, and escalation times for each task.
- Record the number of revisions and quality defects per deliverable.
- Compare total cycle time and defect rate to a manual baseline from the previous month.
- Discuss results with your team to identify tasks where AI adds value and tasks where it does not.
- Adjust workflows accordingly, and repeat the audit quarterly.
Limitations and next steps
This protocol relies on self-reported data, which can be subject to bias. For more rigorous analysis, consider using time-tracking software or having a manager validate a sample of logs. Also note that productivity is not only about speed: worker satisfaction, learning, and creativity matter. The ILO research highlights that one in four workers globally are in occupations with some degree of GenAI exposure, and that the transition should be managed through social dialogue. Involve your team in setting metrics and interpreting results.
Finally, remember that no single metric tells the whole story. Combine time data with quality scores and qualitative feedback to form a complete picture. By measuring what really happens after the AI generates its output, you can make informed decisions about where to invest in automation and where to rely on human skill.
Sources and further reading
Guidance and regulations can change; check the current primary sources before applying them to a consequential workplace decision.