Home · Tooling

The minimum lab stack

There is a vendor-shaped fantasy of the "store of the future lab": ceiling lidar, shelf cameras, emotion analytics, a dashboard wall. It photographs beautifully and answers questions nobody asked. Meanwhile the four tools below, none exotic, cover the overwhelming majority of in-store experiments actually worth running. Start here; add sensors only when a specific question demands them.

1. POS data you can query yourself

The cash register is the best sensor in the building — it records the outcome you actually care about, time-stamped, item-level, free. The catch is access: in many chains, getting daily store-by-item data means begging a BI team with a six-week backlog. A testing practice needs its own tap into transactions, even a nightly CSV. If you can't pull baskets by store and day without a meeting, fix that before buying any hardware whatsoever.

2. A door counter — the humble denominator

Sales without traffic is a numerator without context: you can't tell conversion from footfall, a better store from a busier one. A basic beam or camera counter at the entrance turns transactions into conversion and makes matched-store comparisons vastly cleaner. Counter data is imperfect — staff walk through it, deliveries inflate mornings — but consistent imperfection cancels out in a controlled comparison. Perfection is not required; consistency is.

3. Screens you control remotely

Controllable displays are the experiment platform of physical retail: content variants, dayparting, per-store assignment, instant rollback — the things that make a test cheap to run and easy to reverse. The requirement to check is operational, not cinematic: can you schedule variants per store and per daypart from a desk, and can you prove what played where (a proof-of-play log is your treatment record — auditors and skeptical CFOs both like it). Any mainstream signage platform clears this bar; the differences that matter for a lab are reliability and how little time the scheduling takes you.

4. A boring, trusted analysis sheet

Not a BI suite — a spreadsheet (or a notebook, for the statistically inclined) that holds the pre-registered design: stores, assignment, window, primary metric, threshold. Results get computed the same way every time, by a method agreed before launch. The tool's job is to make the analysis so transparent that nobody can quietly relitigate it after seeing the numbers. Fancy dashboards invite exploration; exploration after the fact is how pilots get "rescued".

What waits until a question demands it

Shelf cameras and dwell sensors earn their place when the question is genuinely about attention paths — a layout experiment, a fixture-position test. Anonymised wifi or device counting helps when catchment overlap is the worry. Eye-tracking and emotion analytics almost never pay for themselves outside research projects. The pattern: instruments follow questions. A lab that buys sensors first and invents questions second is a showroom.

The whole stack above costs less than one regretted chain-wide rollout — which is, conveniently, exactly what it exists to prevent. Start with the pilot skeleton →