A promotion can produce the biggest sales day of the quarter and still leave the business worse off.
Some offers create demand that would not otherwise have happened. Some persuade customers to buy sooner, buy a different product, or use a discount on an order they were already going to place. Most promotions contain a mixture of all four.
The useful analysis begins after the revenue screenshot. It asks what changed, who changed their behavior, what the change cost, and what happened once the offer disappeared.
What you’ll learn
- How to define the job of a promotion before reading its result.
- How to compare the event with a credible baseline.
- How to inspect incrementality, pull-forward, margin, customer mix, and product halo.
- How to turn the review into a better promotion rather than another debate about topline revenue.
First, decide what the promotion was meant to do.
A launch offer, a stock-clearing event, a new-customer incentive, and a loyalty reward have different jobs. Judge each against the intended behavior.
Before opening the dashboard, complete this sentence: “This promotion exists to get this customer to take this action within this period, while protecting this commercial boundary.”
For example: “Move existing customers with no purchase in six months into a second order during the next ten days, while keeping contribution per order above the agreed floor.” That is measurable. “Drive excitement and sales” is not.
Use a six-part promotion scorecard.
1. Build a baseline you can defend.
The cleanest answer comes from a randomized holdout: an eligible group that does not receive the offer. Many brands cannot run one for every campaign. In that case, use the strongest comparison available and state its weakness.
A matched prior period can work if weekday mix, season, stock, media pressure, and competitive context are reasonably similar. A forecast based on recent run rate can help, provided it was made before the result was known. Last year’s event can add context, but price, audience size, and brand maturity may have changed.
Record the expected orders, revenue, contribution, and customer mix without the promotion. Add the assumptions behind the estimate. A baseline rewritten after a strong result is just a story that flatters the outcome.
2. Estimate the demand created above baseline.
Start with observed lift: promotion-period orders or contribution minus the baseline expectation. Call it observed lift, not proven incrementality, unless a credible holdout supports the stronger claim.
Break the lift down by new and returning customers, product, market, and channel. An event may add new customers while merely shifting the timing of loyal-customer orders.
3. Look for pull-forward after the offer.
Extend the measurement window beyond the promotion. If sales fall below baseline immediately afterward, part of the event may have borrowed demand from the future. That does not automatically make the promotion bad. Pull-forward can be useful when cash, inventory, or launch momentum matters. It should be visible in the economics.
For products with a longer replenishment cycle, a one-week post-period may be too short. Match the follow-up window to how customers normally buy.
4. Rebuild the result in contribution, not revenue.
Include product margin, discount cost, shipping subsidy, returns, payment fees, fulfilment, and incremental media where the data supports it. Avoid false precision if some costs arrive later. Label the estimate and update it.
A promotion can reduce contribution on each order yet still create valuable total contribution through real incremental volume. It can also create record revenue while destroying contribution because existing customers received an unnecessary discount.
5. Inspect who and what entered the business.
New-customer count is not enough. Compare first product, order value, full-price versus discount behavior, return rate, market, and early repeat signals. The goal is not to declare a cohort good after a few days. It is to see whether the offer selected for the customers the business intended to attract.
Also inspect product mix. A discount on a hero product may introduce customers to a higher-margin companion product. Or it may cannibalize a full-price bundle. The basket tells you which.
6. Read the operational consequence.
Did the event create stockouts, service contacts, fulfilment pressure, return risk, or cash strain? These are part of performance. A promotion that overwhelms the operating system can damage the next month even when the campaign dashboard looks healthy.
That language does not weaken the analysis. It keeps the team from making a certainty claim the design cannot support.
A worked example: strong event, weak new-customer engine.
Suppose a homeware brand runs a four-day 20% promotion. Orders and revenue are well above the recent run rate. The initial summary calls it the best acquisition campaign of the season.
The scorecard changes the interpretation:
- Most of the order lift came from existing customers who had purchased recently.
- The week after the event fell below the expected run rate, suggesting some pull-forward.
- New-customer orders increased, but their first baskets were smaller and concentrated in already discounted products.
- Contribution rose in total, though much less than revenue.
- A complementary full-price accessory sold unusually well, creating a useful product halo.
The promotion was not a failure. It created cash and total contribution, rewarded existing demand, and revealed a strong attachment product. It was also not the new-customer breakthrough the headline implied.
The next test could keep the accessory bundle, narrow the broad discount, and reserve the strongest incentive for a genuinely new or lapsed customer segment. The learning is more valuable than declaring the event either brilliant or disastrous.
Make the next offer answer one question.
Promotion calendars encourage repetition. A better system treats each meaningful offer as an experiment with a commercial job.
- Name the audience and behavior. Who should do what differently?
- Choose the minimum incentive. What is enough to change that behavior?
- Protect the boundary. Contribution, stock, service capacity, and brand rules belong in the brief.
- Preserve a comparison. Use a holdout, segment, geography, product, or timing difference where practical.
- Measure through the after-period. Include pull-forward and early customer-quality signals.
- Record the learning. State what the next event will repeat, change, or stop.
The purpose of a promotion review is not to defend the calendar. It is to improve the next commercial decision.
Frequently asked questions
Do we need a holdout for every promotion?
No, but a valid holdout produces a stronger incrementality answer. When one is not practical, use the best available baseline and label the limitations.
How long should we watch for pull-forward?
Use the normal buying or replenishment cycle as a guide. A frequent-purchase product may show the effect quickly; a durable product may require a much longer view.
Is a promotion bad if margin per order falls?
Not necessarily. The event may create enough incremental orders or valuable customers to improve total economics. Judge the full contribution and downstream behavior, not one margin rate in isolation.
What if the promotion was intended to clear stock?
Then inventory released, cash recovered, and disposal or holding costs avoided belong in the scorecard. New-customer quality may be secondary, though brand and service consequences still matter.
A sale is an event. A useful promotion is a controlled attempt to change customer behavior at an acceptable cost.