Gartner's October 26 workforce research found that only 21% of time freed by AI was redirected to more innovative work. The finding comes from a second-quarter 2026 survey of 1,865 employees.
The result exposes a measurement gap. A company may know that an AI use case reduced the time spent on a task without knowing how much capacity became usable, where it went, or what the reassigned work produced.
The AIDE Institute recommends starting with a named unit of work: one alert, claim, review, ticket, interview, report, or other repeatable item that can be measured before and after the workflow changes.
What is the unit of work?
Time saved is evidence about a task. It can show that a queue moved faster, a review took fewer hours, or a recurring step required less manual work.
Those observations become more useful when the unit is explicit. “The process is faster” is difficult to audit. “Review time per application fell during the same measurement window, with the same review standard” identifies what was counted.
The record also needs a baseline, population, measurement window, quality or control threshold, and the human work required before and after the change. Otherwise, a time-saving claim can hide changes in volume, complexity, review effort, or error risk.
Gartner's public release says organizations should treat gained capacity as an asset and redesign work so employees can use it. Its 21% result describes how surveyed employees' freed time was redirected. It does not establish that AI caused innovation to rise or fall, and the release does not provide the full questionnaire or sampling frame.
How does the Block case separate the measures?
The AIDE AI Use Case Catalog surfaced a public Block security case and its original provider source. OpenAI reports that Block's security team uses Daybreak for work that includes false-positive triage and security investigations.
The provider page reports roughly 600 analyst-hours per quarter handed back from false-positive triage. It separately reports approximately 1,100 analyst-hours of investigation completed per quarter.
These figures describe different parts of the workflow. Throughput describes work processed. Hours handed back describe capacity returned. Investigation hours describe work completed. The source says the team now spends more time on deeper investigations and responsible automation, but it does not establish a one-to-one conversion between every triage hour saved and every investigation hour completed. The figures are provider-reported and have not been independently audited.
OpenAI says people validate results, the system includes human takeover hooks, and work can be routed to human review. A unit-of-work measure should include that review and control effort.
Where does the use-case catalogue fit?
The AIDE Institute built the AIDE AI Use Case Catalog as a structured record of publicly verifiable AI use cases across the S&P 500. The public beta description covers company, sector, business function, internal and customer-facing uses, production status, recorded results where public, and source links.
That structure keeps the workflow, evidence state, and source trail attached to the use case. Catalogue inclusion does not prove that a result is accurate, causal, or ready for external publication. The original source still governs the claim.
The unit of work adds the measurement anchor. The use-case record can connect that unit to deployment state, task change, capacity returned, capacity destination, downstream result, and source evidence.
What should each productivity use-case record answer?
The AIDE Institute recommends six questions.
| Record field | Question |
|---|---|
| Use case and unit of work | Which workflow changed, and what repeatable item is being counted? |
| Baseline and task effect | What happened to time, throughput, quality, or review effort for the same unit, population, and window? |
| Capacity returned | How much usable time became available, for whom, and over what period? |
| Capacity destination | Which work received the released capacity? |
| Downstream result | What changed because that work was completed? |
| Source and evidence state | Who reported the result, what does the source support, and what remains unaudited or uncertain? |
The fields support different decisions. Task evidence can support continued testing. Capacity and destination evidence can support a workflow or role change. Downstream evidence can support expansion or a financial claim. The source record shows how much confidence each decision can carry.
Time saved may contribute to AI ROI, but it is not a complete return calculation. A financial claim still needs an agreed baseline, implementation and oversight costs, a measurement period, and a result the organization values.
Gartner identifies the gap between time freed and time redirected to more innovative work. A stable unit of work makes the first change measurable. The use-case record then shows whether that task improvement became usable capacity, reached a named destination, and produced a result worth scaling.
Sources
- Gartner, “77% of Board Members Expect AI to Deliver Productivity and Cost Savings, Yet Just 3% of Organizations Exceed Cost Savings Goals Via Productivity,” October 26, 2026
- OpenAI, “How Block is building AI-powered security with OpenAI technology,” October 14, 2026
- AIDE Institute, “Request to join the AI Use Case Catalog beta”
