Peer benchmarking is only useful, and only fair to participants, if no single business can be singled out from the comparison. k-anonymity is a formal way of guaranteeing this: a cohort (for example, "cleaning businesses with 50 to 200 staff in Queensland") is only shown as a benchmark once at least k real businesses sit inside it, where k is a minimum threshold set in advance.
If a cohort has fewer than k members, the benchmark is withheld rather than shown with a small, potentially identifying sample. This trades off some benchmark coverage (smaller or newer cohorts may not have a benchmark yet) for a firm privacy guarantee that no participant's individual figures can be isolated from the group number.
How it is typically enforced
- The threshold (k) is checked at write time, before a benchmark is computed or displayed, not just at read time.
- Cohorts are defined on shared, non-identifying attributes (industry, size band, region), never on any single business's name or account.
- Benchmarks refresh periodically (for example nightly) rather than in real time, which further reduces the chance of inferring a single business's change from a shift in the group figure.
Nuromi's peer benchmarking uses a k>=10 gate: a cohort must have at least 10 real tenants before any "how you compare" figure is computed or shown, checked at write time via a nightly job, so a benchmark only ever appears once it is genuinely anonymous.