CPG Promo Analysis with AI: the Complete Guide
CPG promo analysis (also called trade promotion analysis) is the process of measuring what a retail promotion actually earned: isolating the promoted weeks in syndicated point-of-sale data, separating baseline sales from incremental sales, netting out volume that would have sold anyway, and comparing the incremental margin to the trade dollars spent. Its output is a per-event, per-retailer read of lift and return that decides which promotions to run again.
- Trade promotion is often the second-largest line on a CPG P&L after cost of goods, and in POI's 2026 State of the Industry survey nearly 68% of manufacturers said they spend more than 15% of revenue on it.
- The last large public benchmark, Nielsen's 2015 Trade Promotion Landscape Analysis of 76 million event weeks, found that 59% of promotion events across 7 countries did not break even, and 71% in the US did not.
- Promo analysis has four layers, each answering one question: which weeks were promoted (the event), how much more sold than expected (the lift), what that volume earned after cost (the return), and where the volume came from (the source).
- Syndicated data sees only three retailer tactics: a temporary price reduction, a feature, and a display. Coupons, bonus packs, retail media, and anything else the manufacturer does outside the store are invisible to it.
- Lift is not return. A promotion can lift volume and still lose money once trade cost, the cost of goods on the extra units, and the share of the discount the retailer kept are counted.
- An AI agent earns trust on promo work by refusing the shortcuts an analyst learns to refuse: averaging lift across retailers, summing incremental across markets, and computing a return when the cost file is missing.
Chapters
The six merchandising conditions at a glance
| Condition | What it is | How the data detects it | What it is for in analysis | Common misread |
|---|---|---|---|---|
| TPR only | A temporary shelf price cut with no feature or display | Promoted price more than 5% below regular price; reverts to regular after 7 weeks (Nielsen) or 6 (Circana) | Reading price response on its own | Treating a permanent price change as a promotion |
| Feature only | The item appears in the retailer's circular, insert, or mailer | Retailer print and coupon activity coded by the provider; manufacturer coupons excluded | Reading the value of retailer advertising | Counting a manufacturer FSI as a feature |
| Display only | The item sits in a secondary, shoppable location | Store audits record location and price; Nielsen counts temporary displays only, Circana includes permanent | Reading the value of secondary placement | Assuming a display always carries a price cut |
| Feature and display | Circular support and a secondary display in the same store-week | Both conditions present; almost always the lowest promoted price | Reading the ceiling on lift for an item | Adding it to Feature only and Display only and double counting |
| Any promo | One or more of the three tactics active | The union of all merchandising conditions | The default cut for total promoted volume and subsidized volume | Summing the 'only' conditions and adding Any promo on top |
| No promo | Regular shelf price, no support | No merchandising flag on the store-week | The observed non-promoted volume and price | Using it as the baseline; base is modeled, non-promo is observed |
Frequently asked questions
Promo analysis, also called trade promotion analysis, is the measurement of what a retail promotion earned rather than what it sold. It isolates the promoted weeks in SPINS, Circana, or Nielsen data, separates baseline from incremental sales, nets out volume that would have sold anyway, and compares the incremental margin to the trade dollars the brand spent. The result is a per-event, per-retailer read of lift and return.
Take the provider's modeled baseline for the promoted weeks, subtract it from actual sales to get incremental, and divide incremental by base. If a week sold $100 on promotion and the baseline was $80, incremental is $20 and lift is 25% by the incremental-over-base convention. NielsenIQ's dictionary describes the same event as a 20% lift, the share of promoted sales that was incremental. Always state which convention you are using, and compute it per retailer and per event.
Follow the framework Nielsen used in its 2015 benchmark: incremental sales equals total sales minus baseline; incremental cost equals direct trade expense plus the cost of goods on the incremental units; trade return equals incremental sales minus incremental cost; trade efficiency equals trade return divided by dollars invested. Convert retail dollars to manufacturer margin before comparing to cost. Above $1 of efficiency the event paid back.
The most recent large public benchmark is Nielsen's 2015 Trade Promotion Landscape Analysis, covering 331 categories and 76 million event weeks across the US, Canada, and 5 European markets. It found 59% of promotion events did not break even globally and 71% did not in the US. No comparable public benchmark has been published since, so treat the figure as dated.
Per POI's 2026 State of the Industry survey, nearly 68% of consumer goods manufacturers allocate more than 15% of annual revenue to trade promotion, with many between 16% and 23% and a significant segment above 27%. POI's 2023 survey put the historical range at 11% to 27% of revenue. These are member self-reports, not audited figures.
Off-invoice is a discount on the units the retailer buys, paid whether or not the shopper sees a lower price. Bill-back is paid after the event on proof that the retailer performed. Scan-back is paid per unit scanned at the promoted price, so the money follows the shopper. Off-invoice carries the most forward-buying risk for the brand; scan-back carries the least.
Pass-through is the share of a manufacturer's trade discount that reaches the shopper as a lower shelf price. In the largest published study, covering more than 1,000 stores in over 30 states, a 10% manufacturer price cut reached the shopper as a 4.1% cut on average, and retailers passed through about 69 cents of each wholesale discount dollar. The variance across retailers was so large that the averages are of little use for any one account.
The drop in sales after a promotion ends because shoppers bought earlier or stocked up during the event. Household studies expect it; store data rarely shows it plainly. One study of store scanner data in two categories estimated the dip at 4% to 25% of the promotion's current-period sales effect. Dips are stronger for high-priced, frequently promoted, mature, high-share items.
Because lift counts volume and ROI counts money. The usual causes are a deep discount that the brand funded, a high share of subsidized volume (sales that would have happened anyway), low pass-through (the retailer kept part of the discount), and the cost of goods on every incremental unit. A promotion can lift volume 60% and still return less than $1 per dollar invested.
Yes, if it enforces the rules a good analyst enforces by habit: one promo condition per query, decompositions pulled at the level being reported rather than rolled up, ratios recomputed rather than averaged, base plus incremental checked against total, retailer-level reads, and an explicit refusal to compute a return when the trade cost input is missing. An agent that skips those rules produces confident wrong numbers faster.