The compost promotion that might not be a success
A regional garden-centre chain ran a three-week '20% off all potting compost' promotion in all eight of its stores in March. Compost sales rose 34% versus February. The marketing lead wants to declare the promotion a success and repeat it every quarter. You have weekly sales by store and product for January to April, the promotion calendar, and local weather records - March was unusually warm and dry. No stores were held back as a control. Sales figures are monthly totals in currency, not units, and you do not have prior-year data in the same system because the till software changed last November. Competitor pricing is unknown.
1. The marketing lead wants to repeat the promotion quarterly on the strength of the 34% rise. What can and cannot be concluded from the evidence described, and why?
Read the worked answer
The 34% is a before-and-after comparison against February, the weakest month in a garden centre's year, so it bundles at least three effects together: spring seasonality, an unusually warm dry March (planting weather), and the 20% price cut.
Nothing in the data shows what would have sold without the promotion. No stores were held back, and there is no prior-year series in the same system to estimate a normal March uplift. That counterfactual is the entire question, so the defensible conclusion is that the result is consistent with a promotion effect but is not evidence of one.
Two further cautions:
- The 34% is in currency, not units. With 20% off, unit volume must have risen about 68% (1.34 ÷ 0.80) just for revenue to be up 34%.
- Gross margin will have moved far less favourably than revenue, because every unit sold earned less.
Recommending a quarterly repeat on this basis treats a seasonal coincidence as a causal result.
2. Using only the data you have, what analysis gives the best available estimate of the promotion's incremental effect?
Read the worked answer
Build a difference-in-differences estimate rather than a simple before/after.
- Choose a comparison group that shares the confounders but not the promotion - non-promoted categories in the same eight stores (this absorbs store-level seasonality and the warm weather), or the same weeks in the previous year if any comparable extract exists.
- Establish the pre-promotion trend: plot weekly compost sales against the comparison group for January and February and check the two moved together. If they did not, the comparison is not usable and you should say so.
- Expected sales = the comparison group's growth rate applied to compost's pre-promotion level. Incremental = actual − expected.
- Report a range and split it by store; a promotion that works in two large stores and fails in six is a different decision from a chain-wide lift.
State the assumption explicitly: the comparison group must not be affected by the same promotion (exclude anything else discounted in those weeks).
3. What single metric should the chain use to decide whether to repeat the promotion, and what data would that require?
Read the worked answer
Incremental gross margin, not sales value.
Incremental gross margin = (extra units sold × margin per unit after the discount) − promotion costs (markdown, signage, extra staffing) − cannibalisation and pull-forward effects.
Why not sales value: the discount guarantees each unit earns less, so revenue can rise while profit falls. Sales value is also inflated by customers who would have bought at full price anyway - those units are not incremental.
Data you would need to collect or confirm:
- Cost price per unit, so margin after the 20% cut can be computed.
- The baseline unit volume from the difference-in-differences estimate above, not February's actual.
- Post-promotion weeks, to detect pull-forward: if April sales dip below trend, some March 'lift' was borrowed, not created.
- Whether full-price compost or adjacent full-price lines (plants, containers) fell during the promotion.
A promotion can be worth repeating on margin even when the revenue headline understates it, and vice versa - which is exactly why the decision rule must be profit, not turnover.