ProcessConsumer Packaged Goods
Trade promotion post-event analysis: measuring what a promotion really earned
Post-event analysis answers one question: what did a promotion add that would not have happened anyway, and what did that cost once every allowance and deduction was counted? This page sets out a working method, from assembling shipments, sell-out and trade spend to estimating a baseline, splitting lift into its parts and turning results into guidance on mechanic, depth and frequency per retailer.
On this page
- TPM, TPO and post-event analysis: where each one fits
- The planning loop that post-event analysis closes
- Data to assemble for every promotion event
- Baseline methods and how each one fails
- Decomposing lift into the parts that matter
- Trade cost terms in a fully loaded return calculation
- When post-event results are not decision-grade
- Turning post-event results into guidance for the next plan
- A snack brand's buy-one-get-one at a grocery chain
- Questions and answers
TPM, TPO and post-event analysis: where each one fits
The three terms are often used interchangeably but describe different jobs. Post-event analysis is the measurement step between the other two.
| Aspect | Trade promotion management (TPM) | Post-event analysis | Trade promotion optimization (TPO) |
|---|---|---|---|
| Question answered | What did we agree, with whom, and what do we owe? | What did each past event add, and at what cost? | Which calendar should we propose next? |
| Timing | Before and during the event, through settlement | After the event, and again once deductions settle | Before the annual or quarterly negotiation |
| Main output | Approved plans, accruals and claims matched to agreements | Incremental volume, profit and return per event | Recommended mechanics, depths, frequencies and timings |
| Main failure | Spend booked without a link to the event it funded | Baselines that misstate what would have sold anyway | Optimizing on response curves never properly measured |
The planning loop that post-event analysis closes
- Plan the calendar
Account teams agree mechanic, depth, timing and funding with each retailer.
- Run the event
The promotion runs in store and online, with display, feature and price confirmed.
- Settle trade spend
Allowances, scan-backs and fees are matched to the event, including late deductions.
- Measure the event
Baseline, lift decomposition and fully loaded return, with a confidence label.
- Update guidance
Results adjust the rules of thumb and response curves for the next proposal.
Data to assemble for every promotion event
Gather these per event and retailer before modeling. A missing item does not stop the analysis, but lowers the confidence of the result.
Baseline methods and how each one fails
The baseline is the estimate of what would have sold without the promotion. Every lift figure inherits its errors.
| Method | How it works | Where it fails |
|---|---|---|
| Pre-period average | Average sales over the weeks just before the event. | Breaks with seasonality or trend, or when an earlier promotion falls in the pre-period. |
| Same period last year | Matching weeks of the previous year, adjusted for growth. | Breaks when distribution, pricing or competitors changed, or last year's weeks were promoted too. |
| Time-series model with causal factors | A regression or forecasting model fitted on unpromoted weeks, with price, seasonality, holidays and distribution as inputs. | Struggles with heavily promoted products that leave few clean weeks to learn from. |
| Machine learning across the portfolio | Gradient-boosted trees or similar models learn baseline patterns across many products' promotion histories. | Can absorb promotion effects into the baseline if promoted weeks are handled carelessly, and is harder to explain. |
| Control stores or synthetic control | Compares stores that ran the event with similar stores, or a weighted blend of stores, that did not. | Needs store-level data and events that did not run everywhere, which national promotions rarely allow. |
Teams often combine methods, for example a model-based baseline checked against a control comparison where one exists.
Decomposing lift into the parts that matter
Gross lift is sales above baseline during the event. Several adjustments come before any of it is incremental.
Measure gross lift in consumption
Compare event sell-out with the baseline for the same SKUs and retailer. Use consumption, not shipments, because retailers often buy ahead at the promoted cost.
Net off the post-promotion dip
Shoppers who stocked up buy less afterwards. A dip below baseline in the following weeks is volume borrowed from the future, not created.
Subtract cannibalization within your range
Check whether sister SKUs, other pack sizes or your other brands at the same retailer lost volume during the event, and offset that loss.
Add halo only where evidence supports it
Count lift in complementary or unpromoted lines only when it recurs across events. Note any simultaneous competitor promotion or category-wide growth, which distort the reading.
Reconcile with shipments
Compare the consumption result with retailer orders. Large gaps signal forward buying or diverted stock, which raise the real cost of the event.
Trade cost terms in a fully loaded return calculation
Return compares incremental gross profit with the full trade cost of the event. The cost side is where most analyses fall short.
- Off-invoice allowance
- A per-unit discount on orders shipped in the promotion window, including units bought ahead and later sold at full price.
- Scan-back
- A payment per unit sold to shoppers during the event, claimed afterwards from POS records.
- Bill-back
- An allowance claimed after purchase, usually against proof of performance, instead of on the invoice.
- Fixed fees
- Lump sums for circular features, retail media or secondary displays, paid regardless of volume.
- Deductions
- Amounts a retailer withholds against claimed allowances; they may settle weeks later and some prove invalid.
- Incremental gross profit
- Incremental units multiplied by margin at the promoted net price after cost of goods, not at list price.
When post-event results are not decision-grade
Label each result so planners know how much weight it can bear.
Overlapping events
Early signalPromotions on related SKUs ran at the same retailer in the same weeks.
MitigationMeasure the combined effect, or mark each result as indicative.
Out-of-stocks during the event
Early signalSell-out flattens mid-event while other demand signals stay high.
MitigationFlag the event as supply-constrained and exclude it when setting depth or frequency.
Thin history
Early signalA new SKU, a new retailer, or a mechanic run only once or twice.
MitigationBorrow strength from similar products and retailers, and state wider uncertainty.
Unsettled trade spend
Early signalLarge open deductions or disputed claims.
MitigationPublish a provisional return and re-run it when settlement closes.
Unverified execution
Early signalNo evidence that the display or feature ran as agreed.
MitigationKeep verified and unverified events apart when estimating response to support.
Turning post-event results into guidance for the next plan
- If
Gross lift is strong, but most of it disappears after pull-forward and cannibalization.
ThenReduce frequency or depth for that mechanic at that retailer, and test a shallower discount.
The promotion mostly moves existing demand in time or between your own products.
- If
A mechanic works at one retailer and fails at another.
ThenKeep guidance retailer-specific rather than averaging across customers.
Shopper missions, circulars and price perception differ between banners.
- If
Results are only indicative, because of overlaps, stock-outs or thin history.
ThenPass them on as directional notes, and design the next event to be measured cleanly.
One clean event teaches more than several noisy ones.
- If
Several seasons of decision-grade results exist for a category.
ThenBuild response curves by mechanic, depth and retailer for optimization models.
Optimization finally has measured history to learn from; custom AI model development covers how such models are trained and evaluated.
A snack brand's buy-one-get-one at a grocery chain
Questions and answers
How long after a promotion should post-event analysis be run?
Run a first read once sell-out for the event and the following weeks has arrived, so the post-promotion dip is visible, and treat it as provisional. Re-run the return calculation when the main deductions and claims have settled, which can take much longer. Planners usually need both: an early read for the next negotiation and a final figure for the annual trade spend review.
Can post-event analysis work with shipment data only?
Only partly. Shipments mix shopper demand with the retailer's forward buying and stock policy, so you can estimate an event's effect on shipments but cannot separate shopper pull-forward from retailer stock building. Where a customer shares no sell-out, label results as shipment-based and avoid using them to set discount depth.
What is the difference between promotional lift and incremental volume?
Lift usually means sales above baseline for the promoted item during the event. Incremental volume is what remains after subtracting the post-promotion dip and cannibalization of your own products, and adding any proven halo. Only incremental volume belongs in a return calculation; reporting gross lift as incremental is a common reason promotions look better on paper than they performed.
Is machine learning necessary for promotion analysis?
Not always. Simple baselines can work for stable products with few promotions. Machine learning helps when a portfolio has many SKUs, frequent overlapping events and several causal factors, because it learns baseline patterns across products. It also adds explanation work: account teams need to see why a baseline sits where it does before they accept the result.