Research · methodology
How we measure what happens in the store — before we publish a single number.
The Conversation Benchmark is a recurring report on how in-person retail sales conversations actually behave: how they open, how needs are discovered, what gets offered, and why sales close or don’t. Before publishing findings, we publish the method. This page describes how the data is captured, protected, aggregated and tested — so every number that follows can be judged on its merits.
Two commitments define the whole approach: consent first, and a census instead of a sample. Everything else derives from those two.
Consent-first capture, by design
Cognifyze captures in-person sales interactions through in-store sensors, with consent and clear notice, in compliance with LGPD and GDPR. The system is built to measure the interaction, never to identify the individual: no consumer identification, no biometric matching, no linking of a conversation to a person.
What reaches analysis is the structure and content of the professional interaction — greeting, discovery, offer, objection handling, close — not the identity of who was in it. Privacy is not a feature toggle; it is the architecture.
A census, not a sample
Classic store measurement samples: a mystery shopper sees a handful of visits per month; a survey hears the fraction of customers who choose to answer. The benchmark is built on the opposite basis — every captured interaction in participating stores enters the aggregate. That removes the two biases sampling cannot escape: who gets observed, and who chooses to respond.
Census-level coverage is what makes conversation-level metrics meaningful: rates of needs discovery, offer frequency, objection patterns and close behaviour can be stated as measured frequencies, not extrapolations from a dozen observations.
Aggregation and anonymity in reporting
Published findings are aggregates. No published figure identifies a retailer, a store, a seller or a customer. Results are reported at the level of patterns — by segment, by interaction stage, by conversation behaviour — with minimum group sizes that make re-identification impracticable.
Participating brands see their own data; the public benchmark sees only the anonymous whole.
Statistical treatment
Where we publish effect claims, they come from same-store comparisons — the same stores measured before and after — so seasonal and location effects don’t masquerade as results. Published effects are tested for significance; our reference deployment measured same-store conversion moving from 51.5% to 79.5% (+28pp) at p<0.001, with 383% ROI and payback in 1.4 months.
Findings that do not clear significance are reported as observations, or not at all. The benchmark’s value depends on that discipline.
What the benchmark will publish
The recurring report will cover the anatomy of in-person retail conversations: how often needs discovery actually happens, what distinguishes closing conversations from losing ones, how objections are handled, and how these patterns differ by segment. First edition: findings drawn from the aggregated base described above.
The methodology is published ahead of the numbers deliberately — so the numbers, when they arrive, inherit its credibility rather than asking for trust.
The survey measures what customers say. We measure what happened.
If you run physical stores and want the interaction layer measured — with consent, without identifying anyone — the same platform behind the benchmark is available to your operation.
Book an executive diagnostic.