01
Map the task types
Separate repeatable research, drafting, analysis, and judgement-heavy work instead of treating the team as one use case.
Research-informed case study · Knowledge work
Better work at the edge of what the tool can do.
A practical model for using AI in research, drafting, and analysis while routing ambiguous work to experienced people for review.
Starting point
Knowledge teams need speed, but not every task is a good fit for the same tool. The difficult part is knowing where assistance helps and where it creates risk.
Work delivered
A guided model that separates tasks by fit, uses AI for research and drafting, keeps ambiguity with a human reviewer, and measures quality instead of assuming it.
Useful result
A controlled consulting experiment reported more than 25% faster completion on tasks within the tool's capability frontier, while performance fell on tasks outside it. The boundary matters as much as the speed.
The workflow
Start with the work, then define the boundary.
Useful knowledge-work adoption is a decision system: it identifies suitable tasks, gives people a reliable review path, and makes exceptions visible before they become expensive rework.
What changed
25%+
faster on suitable tasks in one controlled experiment
02
Set the review boundary
Define which outputs need expert checking, source validation, or a second opinion before use.
03
Test a useful workflow
Run a focused task with real inputs, realistic time pressure, and a clear quality measure.
04
Route the exceptions
Keep unusual, high-impact, or ambiguous work visible and deliberately on the human path.
The practical difference
Speed comes from knowing what not to automate.
A team gains more from a clear boundary than from a vague instruction to use AI everywhere. The workflow should show where the tool is strong, where it needs checking, and where it should not be used.
01
Better task fit
People use assistance where it matches the shape of the work.
02
Visible exceptions
Ambiguous or high-stakes outputs have a defined route to review.
03
Useful measurement
Time saved is considered alongside accuracy, sources, and rework.
What to carry forward
A useful next test is specific.
Use the pattern as a starting point. Choose one workflow, one owner, and one measure that will tell you whether the change is worth keeping.
- 01 Classify tasks by capability fit before selecting a platform.
- 02 Give reviewers a short, repeatable way to check sources and assumptions.
- 03 Test both suitable and unsuitable work so the boundary is understood.
- 04 Treat exception handling as part of the design, not as an edge case.
Evidence and limits
Useful context, clearly labelled.
The 25%+ figure is drawn from a controlled study of 758 consultants summarised by Harvard Business School's Generative AI and Productivity research. The result applies to the study's task mix and conditions, not every knowledge-work process.
Read the HBS research summaryHave a similar workflow?
Let’s work out what is worth testing first.
Bring the repeated work, uncertain handoff, or daily bottleneck. We will help you decide what to improve and how to keep a person accountable for the result.