The lexicon

The share of a market's total all commodity volume flowing through stores that sell a given product. A 75% ACV distribution means the stores carrying the product account for 75 cents of every retail dollar in that market.

All commodity volume: the total dollar sales of a store or retailer across every product it sells. ACV is the weighting base for distribution measures, so placement in a high-volume store counts for more than placement in a small one.

An approach to data analysis where an AI agent plans and runs multi-step work end to end, choosing the method, running the queries, checking the results, and writing the findings. It replaces waiting on an analyst with asking and acting.

The class of errors made by summing or averaging syndicated measures in ways their definitions do not allow, such as averaging percent ACV across markets or summing weekly velocity into a monthly figure. Queries that fall into the trap run without error and return plausible wrong numbers.

An AI system that plans and executes multi-step work toward a goal, rather than responding to a single prompt. In CPG analytics, an agent decomposes a question, chooses measures, runs queries against syndicated data, checks the results, and assembles the answer.

The automatic detection, validation, and loading of a new data file the moment it lands in a connected folder, drive, or warehouse. Auto-load replaces the manual ritual of downloading, checking, and uploading each period's data drop.

The automatic re-running of every report and workbook built on a dataset when a new period of that data loads. With auto-refresh, new numbers, charts, and narrative appear without anyone rebuilding anything.

A brand whose analysis, reporting, and routine commercial decisions run on agents, with humans on strategy and taste.

An estimate of the volume a product was expected to sell in a period. Each provider models base from the item's own selling history using its own algorithm, so two providers can report different base for the same item in the same week.

A brand's compounding institutional knowledge, held by an AI system that learns from every question, answer, and decision.

Everything an analytics system has learned about a specific brand: its products, categories, retailers, competitive set, and the way the team talks about the business. Brand context is what lets an agent answer questions the way an experienced team member would.

The structured layer of data, semantics, memory, and rules that lets an AI reason about a specific brand instead of brands in general.

The accumulated record of what an AI system has learned about a brand across every question, correction, and period of data, retained so it never starts from zero. Brand memory is what makes an agent more useful in month twelve than in month one.

A living style guide captured from a brand's website and decks, covering colors, type, logo treatment, and voice, and applied automatically to generated materials. In Sous, brand styles keep every chart and slide looking like the brand made it.

An independent sales agency that represents multiple brands to retailers, handling selling, retail coverage, and often category review preparation in exchange for a commission. Emerging brands commonly use brokers to reach retailers they cannot cover with their own team.

The average amount a buying household spends on a product over a period, in dollars or units. Together with penetration, buy rate decomposes sales into how many households buy and how much each one buys.

The nested structure syndicated providers use to organize products, typically running from department down through category, subcategory, and brand to individual UPC. The level a number is calculated at changes what math is legal, so the hierarchy matters as much as the measure.

A retailer's periodic evaluation of a product category to decide what gets added, dropped, or repositioned on the shelf. Category reviews are where distribution is won and lost, and preparing the data for them dominates many brand team calendars.

The product pattern of asking questions about a dataset in plain language and getting answers back in conversation. Chatting with data only becomes trustworthy when each answer is verified against rules for how the underlying measures actually work.

The set of grocery, mass, drug, military, dollar and club retailers. It is the store universe behind the broad syndicated markets MULO and xAOC.

A store or retailer where a product could be selling but currently is not. A void can also be brand-level: a store where the brand is not selling at all. Void analysis compares where an item sells well against where it is absent to build the case for new distribution.

A velocity measure that divides dollar sales by total distribution points, showing how much revenue each point of distribution produces. It lets items and brands with different levels of distribution be compared fairly.

The average price per unit charged when a product is not on promotion, also called non-promoted price. It is the reference point against which promoted prices and discount depth are measured.

Checking every AI-generated query against deterministic rules that run the same way every time, before the answer ships. Fixed-rule verification makes accuracy a property of the system instead of depending on the model having a good day.

The percentage of households that bought a product, brand, or category at least once during a period. A 5% penetration means 5 in 100 households bought it at all. Penetration is a rate, not a count of buying households.

Sales above the modeled base in a period, calculated as total sales minus base sales. Incremental volume is commonly attributed to promotion, but anything that lifts sales above expectation can produce it.

The cycle by which an agent's output is corrected, confirmed, or refined, making every future answer sharper. Each pass through the loop leaves the system knowing the brand a little better than the last.

In syndicated data, a defined set of stores whose sales are combined into one reportable number, such as a channel, a retailer, or a geography. Market definitions differ across providers, so the same brand can hold a different share in two reports that both look national.

The in-store support behind a promotion, recorded as a set of causal conditions: no merchandising, TPR only, feature only, display only, feature and display, and special pack only, plus an any-promotion roll-up across the promoted conditions. The itemized conditions are mutually exclusive only at the level of one item, in one chain, in one week; above that they overlap.

Multi-outlet: Circana's standard total-US market view, combining grocery, drug, mass, club, dollar and military channels into one number. Two variants are routinely confused. MULO+ is an expanded universe, launched January 2024, which added 11 retailers and major e-commerce and increased coverage by more than 15%. MULO+C is the variant that adds convenience stores.

The set of retailers focused on natural and organic products, from co-ops and independents to large regional chains, tracked by SPINS as its own market. It is where many emerging brands prove themselves before expanding into conventional retail. It excludes Whole Foods.

A situation where a product is authorized and expected on the shelf but unavailable for purchase. Out-of-stocks depress velocity and can look like fading demand in the data when the real problem is supply.

Purchase data collected from recruited households who record what they buy (historically with in-home barcode scanners, now overwhelmingly through receipt capture) and projected to the US population. The main US panel is the National Consumer Panel, a joint venture between NielsenIQ and Circana, which SPINS also leverages; Numerator runs a separate, independent receipt panel. Panel measures shopper behavior and POS measures product performance.

A provider-defined block of weeks that syndicated data is reported in. Providers publish a range of rollups: 4, 12, 24 and 52 weeks are common, as are 13 and 26. Periods follow the provider's calendar rather than calendar months, which is why last period and last month rarely cover the same weeks.

A diagram specifying exactly where each product sits on a retail shelf, including position and number of facings. Planograms are how category decisions become physical shelf reality.

Sales data captured at the register when a product scans, showing what actually sold, where, when, and at what price. POS data is the backbone of syndicated retail measurement.

Dollar sales divided by unit sales, giving the average price shoppers actually paid across promoted and non-promoted purchases. Because it is a weighted average, it moves when the promotion mix changes even if no shelf price did. Also called average retail price (ARP). See everyday price.

Products sold under a retailer's own brand name rather than a manufacturer's. Private label competes in nearly every category and usually sets the price floor national brands are measured against.

The average price per unit paid when a product sells under any merchandising condition, such as a price cut, feature, display or special pack. Comparing promoted price to everyday price shows the discount shoppers actually received.

The percentage increase in sales during a promotion compared to the baseline expected without it. A 150% lift means the product sold two and a half times its normal rate while promoted.

A written set of rules encoding how a syndicated data provider's measures may be calculated and aggregated, enforced on every query. Sous maintains separate rulebooks for SPINS, Circana, and Nielsen because the legal math differs across providers.

The average number of times a buying household purchases a product during a period. Frequency shows whether growth comes from shoppers buying more often or simply from more shoppers buying.

An AI system that translates a plain-language question directly into a single database query. On syndicated data, raw question-to-query systems fail quietly because they skip the planning and verification steps that make an answer trustworthy.

The scheduled rearrangement of a retail shelf to a new planogram, when items are added, dropped, and repositioned. Resets typically follow category reviews and are when distribution changes actually take effect in stores.

A provider's revision of previously published data, usually after the measured store universe or methodology changes. Historical numbers shift after a restatement, so trends pulled before and after it may not match.

A category review that rebuilds itself each time new syndicated data lands, refreshing every number, chart, and written takeaway without manual work. The analysis is built once and stays current period after period.

A shared dictionary between data and its users that defines what every measure, hierarchy, and time period means, the same way everywhere. A semantic layer lets an AI system reason about measures instead of guessing from column names.

An analytics error that produces a plausible-looking wrong number instead of an error message. Silent failures are the dominant risk when AI queries syndicated data, because a wrong velocity or share figure arrives formatted and charted exactly like a right one.

Stock keeping unit: a distinct item a brand or retailer tracks, defined by attributes like flavor, size, and pack. A brand's SKU count is how many unique items it fields.

A fee retailers charge brands for placing a new item on the shelf. Slotting works like rent for shelf space, paid before the first unit sells.

Presentation decks generated directly from a workbook, with every number checked against the underlying data before the deck ships. Sous slides turn a finished analysis into a buyer-ready presentation without copy and paste.

Third-party retail sales data pooled from many retailers, cleaned, projected to represent a full market, and sold to the industry as a shared yardstick. SPINS, Circana, and Nielsen are the three major syndicated providers in US CPG.

Total distribution points: the sum of % ACV distribution across all of a brand's items, capturing how widely and how deeply the brand is distributed. A brand with five items at 40% ACV each has 200 TDPs.

The recurring arrival of a new period of syndicated data, typically every 1 to 4 weeks, and the rebuild ritual that traditionally follows it. When a brand says the drop handles itself, the file is detected, validated, loaded, and every report rebuilt without human effort.

Temporary price reduction: a shelf price cut, conventionally at least 5% off the regular price. Hold one long enough and it stops being temporary: in Circana and Nielsen data a TPR sustained for roughly 7 weeks is treated as the new regular price.

The money a brand pays retailers for promotions, discounts, features, displays, and shelf placement. Trade spend is typically a CPG brand's largest cost after making the product itself, which is why measuring what it earns back matters so much.

A velocity measure that divides unit sales by the number of stores selling and the number of weeks, showing the rate of sale in an average store. See velocity.

Universal product code: the barcode number that identifies a specific product at the register. UPC is the finest level of detail in scan data, and every syndicated product hierarchy rolls up from it.

How fast a product sells where it is actually available, also called productivity. Velocity is a family of measures rather than a single metric: dollars per point of % ACV distribution ($/SPP), dollars per TDP, and units or dollars per store per week all measure it against different denominators.

A living analytics document that combines queries, charts, and written narrative in one place, and can be re-run as new data arrives. In Sous, the workbook is the core artifact an agent builds and keeps current.

Extended all outlet combined: Nielsen's broadest standard market, covering grocery, drug, mass, club, dollar and military channels. Nielsen also publishes a version including convenience.