The Aggregation Trap: Syndicated Math that Breaks Silently
Many syndicated measures are non-additive: you cannot average % ACV across markets, sum weekly velocity into a monthly figure, or combine the "only" and "any" merchandising conditions and get a correct answer. The spreadsheet will not warn you, because the arithmetic executes cleanly. Only the measure definitions say it is wrong. This chapter is the reference for which math breaks, why, and what to compute instead. The framing that follows is Sous interpretation built on the provider-verified definitions in chapter 4; the definitions themselves are the providers' own [1][2].
The worked example
Two markets. In Market A you hold 80% ACV; the market's total ACV is $8 billion. In Market B you hold 20%; total ACV is $2 billion. What is your combined % ACV?
Averaging the two gives 50%. That is wrong, and the error is 18 points.
Done properly: covered ACV in A is 80% × $8B = $6.4B. In B it is 20% × $2B = $0.4B. Combined covered ACV is $6.8B against combined total ACV of $10B, so combined % ACV is 68%.
Understate your distribution by that much and you pitch expansion the buyer thinks you already have. Overstate it and you claim shelf presence her analyst can disprove from the same feed.
The time version breaks the same way. 4 weeks at 40, 42, 44 and 46% ACV do not add to 172 (a share cannot exceed 100), and the period figure is not their average either. The provider computes the period figure on the period's ACV base. Pull the period-level number; do not build it from weeks.
The four ways syndicated math breaks
| Failure type | A query that runs fine | What to do instead | Where it bites |
|---|---|---|---|
| Non-additive shares | Averaging % ACV across markets | Recompute on the combined ACV base, or pull the combined market | The national story slide |
| Period-locked measures | Summing weekly velocity into a monthly figure | Use the provider's period-level measure; the denominator moved every week | Trend charts and QBRs |
| Overlapping categories | Summing the "only" merchandising conditions and adding an "any" condition | Pick one framing; the conditions are exclusive only at item-chain-week | Promo recaps |
| Modeled decompositions | Rolling base and incremental up across markets | Re-pull the decomposition at the level you need | Trade reviews |
The common thread: every measure with a base (a share of ACV, a per-TDP rate, a modeled baseline) is only valid at the level where the provider computed it.
Valid but meaningless
A separate category deserves naming, because treating it as an error sends people looking for a bug that is not there. You can sum promotional volume to a total. The arithmetic is fine. It just does not tell you anything, because promotions are executed retailer by retailer and a channel-level average blends events that had nothing to do with each other.
Similarly, you can sum incremental sales across markets. The number is not illegal; it is muddled, because each piece was modeled against a different baseline. The right instinct is not "this is forbidden" but "what would this number mean, and would I defend it?"
Why experienced analysts do not make these mistakes
Nothing in the extract prevents these queries. What prevents them at good data teams is a habit of interrogating every number: what base is this share sitting on? Did the denominator change across the cells I am about to combine? Was this figure computed at this level, or am I about to manufacture it from parts?
That is also why the trap catches capable people who are new to this data specifically. The math skills transfer from finance or e-commerce analytics; the measure definitions do not. Percentages that cannot be averaged and rates that cannot be summed are unusual enough that general spreadsheet instinct works against you.
The judgment is learnable, and it is also encodable. In Sous, these rules live in each provider's rulebook: SPINS, Circana, and Nielsen each get their own, and every query is checked against the applicable one before an answer ships, so an illegal aggregation is refused rather than silently computed. How that verification layer fits the broader system is the subject of how an agent works with syndicated data, and the accuracy stakes are quantified in how accurate are AI agents on syndicated data.
One habit covers most of the risk either way: whenever you combine syndicated numbers across markets, periods, or product levels, stop and ask whether the provider already publishes the combined figure. It usually does, and its version is the defensible one.
Common questions
Can I average % ACV across markets? No. % ACV is a share of each market's ACV and markets differ in size. Combine on the dollar base, or pull the combined market from the provider.
Why is my % ACV over 100? Someone added shares across markets or geographies. Item-level % ACV is capped at 100 by definition.
Can I sum weekly velocity into a monthly number? No. Velocity's denominator changes week to week, so weekly rates do not add. Use the provider's period-level figure.
Which measures can I safely sum? Counts and totals: dollars, units, EQ units. Everything with a denominator needs recomputing on the combined base.