Home / Benchmark Methodology
Benchmark Methodology

How Replio benchmarks, and what it will never share.

Replio tells an operator how their stores perform against comparable restaurants. This page documents exactly where that comparison data comes from, how cohorts are built, and the gates that protect every operator's data. Written for the person whose job is to say no.

Two data layers

A public reference panel, plus an anonymized network.

Layer 1 · Reference panel

Comparable restaurants, public data only

A curated panel of restaurants that are not Replio customers, tracked from their public Google listings. This is the national baseline behind "how does my store compare?"

  • Aggregate numbers only: star rating, cumulative review count, and owner response rate
  • Rating and review count refreshed daily; response rate sampled from each store's most recent public reviews on a rotating cadence
  • No review text and no reviewer identity are ever stored
Layer 2 · Customer network

Anonymized cohorts from real operations

Nightly snapshots of each customer location's own metrics: rating distribution, 90-day review velocity, response rate, and median response time.

  • Surfaced to other customers only as anonymized cohort aggregates that pass every gate below
  • Zero review text and zero guest or reviewer information in the benchmark tables, by schema design
  • Demo, test, and duplicate records are excluded before any math runs
Cohorts

Stores are compared to stores like them.

Every location carries cohort dimensions: service model (QSR, fast casual, casual dining, bar and tavern, fine dining), cuisine, drive-thru, metro and state, chain or independent, and volume band. A benchmark always names its cohort, so "top quartile" means top quartile among genuinely comparable stores, not a grab bag.

The privacy gates

Five gates every benchmark must pass.

Minimum cohort size

No network comparison is shown unless the cohort contains at least 5 locations drawn from at least 3 separate accounts. Below the floor, the comparison falls back to a broader cohort or the public reference panel.

Dominance suppression

If any single account contributes more than 40 percent of a cohort's locations, that cohort is suppressed. A large operator's numbers can never be reverse-engineered by the rest of the cohort.

Fallback hierarchy

When a cohort fails a gate, the comparison widens in a fixed order: brand and metro, then category and region, then category national, then all QSR national. It widens rather than weakens: gates apply at every level.

Bands, not raw numbers

Network benchmarks render as quartile bands or rank-of-N, never as another operator's raw value. There is no view in Replio that displays a specific competitor's number.

Real data only, counted once

Demo and test accounts are excluded by data classification. Multiple records that point to one physical store are deduplicated by canonical place, so no store is ever double-counted in a cohort.

Data integrity

Definitions and durability.

Append-only snapshots
Metrics are captured nightly and never rewritten. If a formula changes, new snapshots carry a new calculation version, and history stays exactly as it was measured. Trends are real observations, not backfilled estimates.
Response rate, customer layer
Share of the location's reviews from the trailing 90 days that received a response, measured per platform with an all-platform roll-up.
Response rate, reference panel
Share of the store's most recent public Google reviews (up to 40) that carry an owner reply. A fixed recent-sample definition keeps stores of very different volumes comparable.
Review velocity
New reviews over trailing 7-, 30-, and 90-day windows for customer locations; day-over-day change in cumulative public review count for panel stores.

What never happens

  • One operator's identifiable numbers are never shown to another operator, in any view, at any tier.
  • Review content never leaves the account that owns it. The benchmark layer contains no review text at all.
  • No guest or reviewer personal information enters the benchmark layer. The schema has no columns for it.
  • Benchmarks never name another operator, account, or franchise group.
  • Replio does not sell data. The benchmark exists to coach your stores, not to broker information.

Questions from your security or data team?

This page is written to be handed to a security review. If your team needs more depth on schema, access controls, or data handling, the founder will get on the call.

Talk to the founder