Product and assortment
Did the new product create additional demand, or mostly shift purchases from something customers already bought?
Anything that changes what you sell: adding an item, removing one, bundling, or reshaping a menu or range.
Decisions this covers
- Product launch
- Product removal
- New bundle
- Menu change
- Assortment expansion
What RolloutGrade can conclude
- What the product group did against what it would likely have done anyway
- How much gross contribution the change added or removed
- How much of the new demand looks displaced from existing lines
- Whether the result survives removing any single location
What it cannot conclude without stronger data
- Whether unmatched demand came from genuinely new visits rather than simply not matching a decline — that needs customer-level purchase histories
- Whether an early effect persists once novelty fades — that needs a longer follow-up window
- Profit, as opposed to gross contribution — that needs labour, rent and overhead
What this needs from you
- Analytical unit
- Product group, per location, per trading day
- Typical data inputs
- Item-level transaction or order-line export
- Unit costs for the products in scope
- The change date and the locations affected
- Comparable untreated locations, when they exist
Outputs available for this change
Only metrics deterministic code already computes appear here. Anything RolloutGrade cannot produce today is listed further down as a gap, not as a feature.
From your transaction export alone
- Gross sales. What rang up, before discounts and refunds are taken off.
- Net revenue after discounts and refunds. Gross sales with discounts and refunds removed.
- Units, receipts and average ticket. How much was sold, on how many receipts, at what average size.
- Historical period comparison. This period measured against a comparable earlier period.
- Evidence grade. How much weight the result can carry, with the reasons stated.
- Scenario sensitivity range. The result recalculated under a fixed high and low scenario. A robustness range, not a confidence interval.
- Study readiness assessment. Whether the data and comparison on hand can carry the decision you want to make.
- Sample size and minimum detectable effect. How large an effect this much data could actually detect.
- Ongoing monitoring checks. Scheduled re-reads that tell you when a decision stops holding.
Available when the data and design allow it
Gross contribution after direct product costs. Net revenue minus the direct cost of the goods sold. Not profit — it carries no rent, labour or overhead.
Requires — A unit cost for every product in scope
Price, volume and mix decomposition. Whether a revenue move came from charging more, selling more, or selling a different mix.
Requires — Item-level rows with quantity and revenue in both periods
Expected performance without the change. Our estimate of what the period would have looked like if nothing had changed.
Requires — Either untreated comparison locations covering the same dates, or a comparable earlier period
Control-adjusted lift. The movement left over once the comparison group's movement over the same dates is subtracted.
Requires — Untreated comparison locations covering both the baseline and the change period
Estimated incremental gross contribution. The gross contribution we estimate would not have existed without the change. Not profit.
Requires — Untreated comparison locations covering both periods; A unit cost for every product in scope
Estimated displacement. How much of the new demand looks like it moved off something the customer already bought.
Requires — Item-level rows covering the products that could have lost volume; A credible counterfactual
Leave-one-location-out sensitivity. The result recalculated with each changed location removed in turn.
Requires — Two or more locations that received the change
Pre-change trend check. Whether the changed and comparison groups already moved together before the change.
Requires — A comparison group; Baseline history before the change date
Execution Fidelity. Whether the change was actually delivered everywhere it was supposed to be.
Requires — The intended rollout recorded for each changed location — which stores were meant to carry the change, over which dates
Locked success-rule evaluation. Whether the result cleared the success bar you locked before seeing it.
Requires — A success threshold locked, with an explicit denominator, before any result for the study was first shown — the lock applies on every surface and refuses once a result has been seen
Customer and repeat behaviour. Whether the same customers came back, and how often.
Requires — A stable anonymous customer identifier on a meaningful share of transactions; Enough history for a repeat window to close
Survey and review context. Why customers say they behaved as they did. Explanatory only — it never moves the economics or the decision.
Requires — Survey responses or a review export
Rollout projection. What the same change might be worth across further locations, with assumptions stated.
Requires — An incremental contribution estimate; A list of candidate locations
Not available yet (7) — what RolloutGrade does not compute for this change
- Decision Isolation. How much of the measured movement can be separated from everything else that changed at the same time. No deterministic calculation produces this today, so it is not offered.
- Decision Stability. Whether the recommended action holds up under reasonable alternative assumptions. No deterministic calculation produces this today, so it is not offered.
- Persistence or novelty decay. Whether an early effect held up or faded once the novelty wore off. No deterministic calculation produces this today, so it is not offered.
- Break-even or reversal threshold. The computed point at which the result would flip the recommended action. No deterministic calculation produces this today, so it is not offered.
- Value of more testing. What another two weeks of data would be worth against the cost of waiting. No deterministic calculation produces this today, so it is not offered.
- New demand versus displaced demand. Whether unmatched demand came from extra visits and larger baskets rather than simply not matching a decline. No deterministic calculation produces this today, so it is not offered.
- Location or geography rollout ranking. Which locations to change next, ordered by expected value. No deterministic calculation produces this today, so it is not offered.
The decision you receive
- Roll out
- Modify
- Do not roll out
- Continue testing
Each decision arrives with the economic impact behind it, the strength of the evidence, a stated condition that would reverse it, and what to check at 30, 60 and 90 days.
Evidence limitations to expect
- Displacement is an estimate built from matching declines, not an observed customer switch.
- Without untreated comparison locations the result stays an observed or historical comparison.