Everything here is a proposal to test, not a finding.
Conditions held equal in both groups
These are not extra treatment features.
- Purpose: the recipient knows why the collection matters to a task they already perform.
- Coverage: the collection states which interviews and dates it includes, because an answer over selected notes is not a picture of all customers.
- Interpretation: the team separates reported customer evidence, a human interpretation and a decision. A citation permits checking, but it does not guarantee that the transcription or the reasoning is correct.
- Control: contributors choose the material and the audience, and review the underlying source where needed, not only the visible summary.
- Setup support: both groups receive the same instructions and support for repeating the task. Whether recipients can repeat it without the champion's help is assessed after facilitation stops, not held equal.
Facilitator time is kept close to equal, so extra attention is not mistaken for product value. Record who attended the source meetings, because a non-attendee's reuse is stronger evidence than the author rereading their own note.
Measures
| Measure | Why collect it | Qualification |
|---|---|---|
| Eligible conversations included | Shows whether useful source context reaches the collection | Count only conversations appropriate for the agreed purpose and audience, and never reward indiscriminate capture |
| Non-author reuse across review cycles | Tests the team-habit hypothesis directly | Use consented observation or a short task log, since reuse may happen outside Granola |
| Review, correction and recap time | Shows whether recipient gains survive contributor effort | Self-reported minutes are estimates, not audited savings |
| Access confusion or unwanted disclosure | Checks whether adoption is eroding contributor trust | Investigate the cause and stop inappropriate sharing, and never widen access to improve the metric |
| Paid workspace expansion | Tests the commercial consequence separately | Measure paid membership and retained use, because shared-link recipients are not seats |
Design notes
- Randomise at team or workspace level where feasible, because members of one review group affect one another.
- Begin with qualitative observation to identify failure points. Then size a controlled test from baseline rates and a detectable effect worth having.
- Predeclare the observation period from the review cadence, and assess whether the behaviour continues after facilitation stops.
- Two successful reviews are an early signal, not proof of a durable habit or long-term retention.
- Allow an inconclusive result. A null result under small samples is not evidence that the treatment adds nothing.