Take the cohort of customers present at the start of the period, measure their revenue at the end including expansion, contraction and churn, and divide by their starting revenue — with no new logos in either side. State the cohort window and whether it is gross or net of downgrades every time you publish it.
NRR is one of the most commonly mis-compared metrics because panels differ so much: monthly versus annual cohorts, ARR versus MRR basis, and whether partial-period churn is annualised. A comparison that fails the panel or time-basis check is measuring definitions, not retention.
Every claim above comes from work LeanScale published. These are the sources.