Also on request: marketing and campaign spend, measured against what would have happened anyway. No published study yet, so it is quoted after a scoping call.
How much volume does a rise cost?Estimate how volume responds to price, product by product. The available price variation and its causes determine what the data can support.
Check the price variation firstProducts with too little independent price variation need further evidence before a price scenario is useful.
Sales fell 10% the month the price went up. Was that the price?Sales may also reflect promotions, seasonality or changes in demand.
Adjust for recorded influencesInclude available promotion and calendar information. Compare model specifications and explain what local demand changes could still bias the estimate.
So which products can take a rise?The elasticity alone does not say. It depends on what each product's margin can absorb.
Compare the estimate with the break-even lossAt constant unit cost, the break-even volume loss is x ÷ (margin + x), where x is the price increase and margin is the starting gross-margin rate. Compare each product’s modelled loss with this limit.
A product-level assessment: candidates for price tests, products to hold and cases needing more evidence. Compare modelled volume losses with break-even losses, then test small price changes with agreed review and stopping rules.
Cross-price estimates where the evidence supports them, with substitution monitored during the test.
Estimated volume response by product, with an interval showing sampling uncertainty under the fitted model.
The volume the margin can afford to lose, computed first.
Test, hold or gather more evidence, with the reason next to each product.
A small move in matched stores, a named owner, an end date.
A named owner reviews results against agreed thresholds and applies the stopping rule.
Use the test results to update the estimates before deciding on further changes.
An elasticity is not a constant: it moves with competition and with the category, so it is measured again rather than assumed to hold.
Did the promotion work? Sales jumped that week.Volume is the easy number to see, and the wrong one to judge on.
Judged on marginMost of what sells that week would have sold anyway, and all of it carries the discount. What the week earns has to beat what a normal week would have earned.
What about the week after?It belongs to the promotion as much as the promotion week does.
The quieter week after, counted tooSales were weaker after many promotions. Counting the following week worsens the estimated return; the data does not establish why sales fell.
Keep it or drop it?That is one decision. The size of the discount is another, and often the one that matters.
The discount depth, decided as wellThe same product can pay at 10% off and lose money at 30%.
A verdict per promotion: what each dollar of discount bought back, the discount sizes that pay and the ones that lose money, and how the lifts differ by placement. Then the shortlist: which discounts and placements to stop or test first, and what each is worth to check against your own costs and supplier funding.
Two in three lost money in the published study, at assumed costs.
Price or mix, which moved margin?The answer depends on the order the effects are taken in, so the order has to be fixed and written down.
One fixed order: volume, then mix, then price, then costEach step is measured with the earlier ones already moved to this year and the later ones still at last year's, so the same two periods always give the same split.
Why do two bridges of the same numbers disagree?Each formula puts the overlap between price and volume on a different line. Both can still add up to the difference, so the two are not comparable until the convention is stated.
One rule for which year goes whereVolume and mix at last year's prices, price on this year's units, written on the page and applied the same way every period, so this quarter's bridge can be set beside last quarter's.
Where do launches and delistings go?A launch has no last year, a delisting has no this year. Pushed into mix, they distort every other product's share.
On their own linesTheir revenue sits apart, so mix compares only the products that existed in both years.
The bridges, a presentation explaining the results, and a workbook holding each product's contribution. A refreshable tool and training can be added to the scope.
One-off analysis, recurring review, or a tool your team is trained to refresh.
Is last year times growth good enough?Only when last year's week was the same kind of week, and the calendar drifts all year.
The same kind of weekLast year's week 12 carried a promotion, this year's does not. Compare to the week that matches, and the budget stops inheriting the drift.
The year came in close. Is the budget fine?The total can hide a lot: stock, staffing and cash are planned by the week.
Checked week by weekIn the published study the rule was 10% out on the weekly total while the year looked fine.
Do we need a forecasting system?Not before the miss has been measured, method by method, on your own weeks.
The miss, measured firstEvery method rolled forward week by week. Most of the gain came from comparing to the right week, with no model.
The corrected rule your team can run in a spreadsheet, the ranked comparison behind it, a range around every number measured at the horizon you use it for, and what the history can and cannot explain, in writing.
In the published study the corrected rule took the miss from 10.1% to 4.16%, on promotion matching and fitted calendar adjustments, with no new data.
A free readiness check first, in writing. Then, if your data can support the study, one fee and about ten working days from export to decision. One export is enough, and no system access. The five steps →
The same list for every study, agreed before anything starts.
The meeting deck and a long-form report on your data, every figure named to the cell it came from.
Every figure in the study, built on your raw rows and nothing pasted, with the working written beside it so your team can follow each step and argue with it. Where a number is meant to be reworked with your own inputs, that sheet carries live formulas.
The order of the calculation, the conventions chosen, the mapping of your P&L lines, the limits, and the code that reproduces it. Six months later nobody has to call me to understand a number.
The study: a fixed scope and one fee, put in writing after the free readiness check. The quarterly follow-up: the same workbook refreshed each quarter, with a short note on what moved and why, for a company with no analyst to run it. Both are priced on scope.
Each study lists the data it needs. If yours is not there yet, these come first, and you do not pay twice for the same data work when the study runs.
A monthly P&L by store, category and channel, against last year and budget. Paste next month's export and it recomputes.
A budget by month with seasonality built in, plugged into the same reporting. Every month shows plan against actual.
Start with the question. If your export cannot support the study you had in mind, you will know on the first call, and in writing within two days.