Articles / Industry Use Cases / AI Business Value
If AI makes one task faster, has productivity improved?
Not necessarily. Productivity improves only when the complete workflow uses less effort or produces more reliable output after verification, correction and handoffs. A fast AI draft can save minutes at one step while creating more checking, coordination or after-hours work elsewhere.
That distinction matters because organizations often measure the visible shortcut—the first draft, summary or recommendation—rather than the work required to make it safe and usable.
A workforce report puts the hidden work on the record
In September 2026, the Australian Resources and Energy Employer Association published Artificial Intelligence: A Resources and Energy Industry Workforce Report. It draws on structured interviews conducted from October 2025 to May 2026 with 33 participants from 23 member organizations.
The report describes adoption as established but uneven. Many organizations remain in early experimentation, while data readiness, governance and workforce capability constrain scale. It also says verification and oversight time should be made explicit, and warns that AI can intensify work even when it automates part of a task.
This is qualitative evidence, not a prevalence estimate. The participant group is small and industry-specific, so it cannot tell us how often workload intensification occurs. It is still useful because it identifies a measurement blind spot: perceived speed at one step is not the same as a better end-to-end process.
Local speed can move work downstream
An eight-month ethnography at one 200-person U.S. technology company offers a second signal. UC Berkeley Haas researchers reported on February 18, 2026 that AI widened some employees’ task scope, encouraged more parallel work and eroded natural pauses. Their study was still in progress, so it should not be generalized to every workplace. But it helps explain how time saved by a tool can be immediately filled with more tasks.
For a business case, the practical question is not “How much faster was the AI step?” It is “What happened to total effort, turnaround and output quality for comparable completed cases?” This extends the value logic in Why AI Pilots Scale Activity but Not P&L Value: activity is not an outcome unless it survives the rest of the workflow.
A 20-case example
Consider a fictional audit of 20 maintenance work packs. Before AI, drafting takes 120 minutes, verification 40 minutes, and correction plus handoff 40 minutes: 200 minutes in total.
With AI, drafting falls to 40 minutes. But verification rises to 90 minutes because reviewers must check sources and equipment details, while correction and handoff rise to 60 minutes. The workflow now takes 190 minutes.
The drafting step is 80 minutes faster, a 66.7% reduction. The complete workflow saves only 10 minutes, or 5%. That is a real improvement in this teaching example, but far smaller than the local speed figure—and it says nothing yet about acceptance quality, safety or employee strain.
Measure completed work, not generated output
For the next 20 comparable cases, record five items: time spent drafting, time spent verifying and correcting, handoff or approval time, whether the output passes first review, and whether it is reopened later. Add after-hours work if the change could shift effort outside normal reporting.
Compare those cases with 20 recent cases of the same type. Keep workload difficulty, reviewer role and required evidence as similar as practical. The goal is not to prove a universal AI effect; it is to find where this workflow gains or loses value.
Worker involvement belongs in the design, not only in the rollout message. The Business Council of Australia’s June 2, 2026 responsible-AI guide recommends testing and co-design with workers, capability building and continued human accountability. This is employer guidance rather than measured outcome evidence, but it gives managers a concrete way to surface verification and handoff work before scaling.
Three takeaways
- Task speed is not workflow productivity. Include verification, rework, approvals and handoffs.
- More output can conceal more work. Check first-pass acceptance, reopened cases and after-hours effort alongside time saved.
- Comparable cases beat impressions. A bounded before-and-after audit is more useful than asking whether the tool feels fast.
Next action: audit the next 20 comparable cases and report total minutes per completed case, first-pass acceptance and reopened cases. Use the same evidence fields in a related pilot evaluation checklist.
Limitation: time and quality measures do not capture every effect. Safety, learning, morale and long-term skill changes may require interviews and observation, and a 20-case audit may be too small for rare failures.
Answer to the opening question: AI has improved productivity only when the complete, comparable workflow—not merely the AI-assisted step—uses less effort or delivers better outcomes without unacceptable new risks.
Sources
- Australian Resources and Energy Employer Association, Artificial Intelligence: A Resources and Energy Industry Workforce Report, September 2026. Interviews conducted October 2025–May 2026.
- UC Berkeley Haas News, “AI promised to free up workers’ time. Researchers found the opposite”, February 18, 2026.
- Business Council of Australia, responsible AI adoption guide release, June 2, 2026.
Leave a Reply