AI in practice
Measure AI work from start to finish.
Measure preparation, checking and rework alongside drafting before deciding whether an AI workflow gives the team useful capacity.
An AI tool produces a draft in seconds. Preparing its inputs takes ten minutes, and checking the answer takes another ten. In this hypothetical example, the generation time is impressive, but it tells the operations leader very little about whether the whole task has improved. Before calling the workflow a success, follow the work from the original request to an accepted result.
Choose the same finish line
Define what counts as complete for both the current process and the AI-assisted route. For a routine report, that might mean the figures have been checked, the commentary approved and the finished report placed in the agreed system. A first draft and a report ready for use are different outputs.
Keep the quality requirements the same. If the faster route produces more missing information, incorrect statements or follow-up questions, record those alongside the time. Agree which errors make an output unusable before looking at the results.
Count the work around the tool
Use a small task log to capture the main stages. Include work done by someone else: moving a check from the author to a manager does not remove it from the process. Record setup and training separately from recurring work so neither disappears from the comparison.
- Preparation: finding, cleaning and arranging the source information.
- Production: drafting or processing, including repeated attempts.
- Review: checking facts, judgement and suitability for use.
- Completion: corrections, transfer to the destination and follow-up rework.
Separate effort from waiting
Record active staff time separately from elapsed time. A report can require little hands-on effort yet sit in an approval queue until tomorrow. Improving one measure does not automatically improve the other.
Compare similar tasks using the same completion criteria. Note differences in complexity, source quality and staff familiarity. Include an awkward case, and retain failed attempts in the record. A handful of observations can identify where to investigate next; it cannot establish a dependable saving across every task.
In the hypothetical report example, an AI draft might shorten writing while creating extra checking because the source figures are difficult to trace. The useful next change could be clearer source references or a simpler input template. Buying more licences would not resolve that specific problem.
Decide what the released capacity is for
If the comparison shows a repeatable reduction in effort at acceptable quality, identify where that capacity can be used. It might support a growing backlog, more customer follow-up or a task the team currently postpones. Small fragments of time spread across a day may not combine into a usable block.
Do not describe released staff time as a cash saving unless an actual cost changes. Keep observed results, estimates and hoped-for benefits separate. Recheck the workflow after normal users have adopted it, especially when the source material or review arrangements change.
Rethink’s AI review and enablement can help define a practical baseline and test one scoped workflow. Ongoing adoption support can then help the team improve the complete process and check whether the intended value is appearing.