What benchmarks can and cannot do
Cohort analysis groups scanners by when they first scanned, such as the week of first scan, then follows each group over time. It reveals whether newer scanners behave differently from earlier ones, which a single total hides.Benchmarks for cohorts are tempting because they promise a shortcut: one number that tells you whether you are doing well. In practice, a benchmark is only as useful as its definition, and definitions vary widely. One source may count every scan, another may count only unique scanners, and a third may measure a different placement type entirely. Two benchmarks that disagree are not a puzzle to solve; they are a signal that the underlying measurements differ. Treat any benchmark as a rough reference, never as a verdict.
Why benchmarks for cohorts vary
Benchmarks vary for reasons that have nothing to do with quality. Placement changes everything: a code at a checkout behaves differently from a code on a poster. Audience matters too: existing customers scan for different reasons than strangers do. Offer strength, season, device mix, and time of day all shift the number. Even the definition shifts it: changing the denominator or the time window changes the result without changing the campaign at all. Cohorts make change visible. If each new cohort retains worse than the last, something in the experience is getting worse.This variation is why copying a published benchmark as a target is risky. You may be comparing your menu code against someone else's checkout code, or your first month against their twelfth. The gap you see may be a measurement gap, not a performance gap, and chasing it can lead you to fix things that were never broken.
How to build your own baseline
Your own baseline is the benchmark that matters most, because it shares your definition, your placements, and your audience. Build it in four steps. First, write down the definition: Cohort retention = members of a cohort who return within a window divided by the cohort size, multiplied by 100. Second, measure it for a fixed period on a stable placement and record the raw counts, the date range, and the method. Third, repeat the measurement for two or three more comparable periods. Fourth, use the range you observed as your baseline, not a single point, because a range respects normal variation.Illustrative example numbers, not real-world data: a January cohort of 200 scanners has 60 return within 30 days, giving 60 / 200 x 100 = 30% cohort retention. A February cohort of 200 scanners with 40 returning gives 20%, so February scanners retained less well in this example.
Once the baseline exists, judge new results against it. A result inside the baseline range is normal variation. A result outside the range for two consecutive periods is a signal worth investigating. This approach is slower than copying a published number, but it is honest, and it improves as your own history grows.