The Detection Floor Check
Send session counts and store-day sales history. Get back a one-page answer on whether any read is possible at your sites — and if not, exactly what would have to change.
Most measurement vendors will take the engagement and produce a number. Some questions cannot be answered at any price, and it is cheaper for both of us to establish that in a week than in a proposal cycle.
A charger effect is small. Store sales are noisy. Whether the first can be seen through the second is arithmetic, and it can be worked out before any money changes hands.
The question is the minimum detectable effect: the smallest change in inside sales your data could distinguish from ordinary day-to-day variation. For 80% power at a 5% two-sided significance level it approximates to:
Residual SD is not the raw swing in daily sales. It is what is left after day-of-week, seasonality and holidays are removed — the variation a model cannot already explain. 2.8 is the constant that falls out of the power and significance levels above. Days per arm is how much history you have on each side of the install or the change.
A store doing $6,000 a day inside, with 12% residual variation, with 180 days either side of the install:
0.12 × 2.8 ÷ √180 ≈ 2.5% of inside sales ≈ $150 a day
A plausible charger effect at that site is in the tens of dollars a day. So that read is unavailable at one store, at any price, and no amount of method sophistication changes it. That is a real answer and you should have it for free.
The $1.85 per session is a Simulated figure from our worked read. Substituting your own is the first thing a real engagement does.
- Narrow the outcome. Read the categories a charging customer actually buys — dispensed beverage and foodservice — in the hours sessions actually happen, rather than the whole store all day. That is roughly a quarter of the sales base carrying most of the effect. The numerator improves about fourfold; the noise on the smaller base grows only about twofold. Net, roughly twice the power, at zero cost. The cheapest lever available and almost nobody uses it.
- Pool sites. The floor falls as the square root of the number of sites. Four sites is a factor of two. Ninety sites is a factor of about 9.5. This is why a network read is available where a single-site read is not.
- Use the broken plugs. Compare the same store on days the plugs worked against days they did not. The store is its own control, so nothing about the store — format, catchment, staffing, remodel — can explain the difference. It needs uptime data and costs nothing else. The sharpest instrument on this list.
- Give it a bigger shock. A price-to-zero window, or any bounded demand shock with a known on date and a known off date. A treatment that turns on and then off can be told apart from a season; one that only ever turns on cannot. See the worked read →
- Count site-days, not sites. A live network across two years is thousands of store-days, not a handful of stores. Power comes from the days as much as from the locations.
And before any of it, we set the noise floor. We run the same analysis on stores with no chargers, on randomly chosen dates, to see what the method returns when nothing happened. If it returns an effect there, the effect it returns anywhere else is not to be trusted. That is not a power lever — it is the check that has to pass before we report anything at all.
- Unit counts by category, by store, by day, two years back
- A narrowed category or SKU set, agreed with you — not every item you sell
- Site attributes: format, location type, layout, remodel dates
- Charger go-live dates for every charging site
- Session and uptime exports where you hold them
- Optional: fuel volume in gallons, never in dollars
- No revenue, prices, or transaction dollar amounts
- No personal data of any kind
- No individual receipts
- No loyalty, card, or payment data
- No cameras, no app data, no tracking
Nothing on the left lets us reconstruct your financial performance. We report unit changes; you apply your own margins.
- Whether a read is available at your sites — yes or no
- The smallest effect your data could detect, in dollars a day
- Which of the five levers closes the gap, if any of them do
- What we would need that you have not sent
- If the answer is no, that answer in writing, with the arithmetic behind it
It costs us almost nothing to run. If it comes back no, you have got the genuinely useful thing out of it for free: the knowledge that nobody else can read it either, and what would have to change before anyone could.
Helpful to include: number of charger-equipped sites, approximate inside sales per site per day, whether you hold session and uptime exports, and when the chargers went live.