Three companies dominate the market for knowing what people bought: NIQ (NielsenIQ), Circana and Numerator. If you work in consumer goods, retail, beverage, beauty or any category that moves through a physical or digital shelf, you are almost certainly paying one of them, and quite possibly two.
Short answer: NIQ is the broadest retail measurement footprint and the reference currency for share; Circana is the strongest across durables, discretionary and foodservice alongside CPG; Numerator is the strongest household and receipt panel, which is the only one of the three that reliably ties purchases to a person. All three tell you what was bought with great precision. None of them tells you why, and none of them contains the people who did not buy. That second gap is the expensive one, because it is where growth actually lives.
This guide compares the three on what they measure, who owns them, what they cost to run, and where each one runs out of road.
The 2026 landscape at a glance
| NIQ (NielsenIQ) | Circana | Numerator | |
|---|---|---|---|
| Core unit of analysis | Item x store x week | Item x store x week | Household x trip x receipt |
| Origin | Nielsen Global Connect, carved out of Nielsen in 2021 | IRI + The NPD Group merger | Founded as InfoScout, renamed 2018 |
| Combined with | GfK (completed 11 July 2023) | NPD and IRI unified under one brand in March 2023 | Kantar's Worldpanel business, 2025 |
| Ownership 2026 | Public (NYSE), IPO July 2025; Advent International legacy backer | Hellman & Friedman, Vestar, New Mountain Capital and management | Bain Capital |
| Stated reach | 90 countries, ~82% of world population | POS from 290 million households globally | More than 1 million households, 50+ countries |
| Strongest in | Global CPG share and distribution | Durables, tech, toys, fashion, foodservice, CPG | US shopper behaviour, cross-retailer basket view |
| Weakest at | Explaining a number it reports | Explaining a number it reports | Explaining a number it reports |
That last row is not a joke, and it is the reason this article exists.
NIQ: the share currency
NIQ is the largest of the three and the only public one. It became the direct parent of the Advent-created holding entities in July 2025 and generated $985.1 million in net proceeds from its IPO.
The FY2025 numbers, from the company's own results release, give a useful sense of scale:
- Revenue of $4,198.4 million, up 5.7% year over year and 5.7% on an organic constant-currency basis
- Adjusted EBITDA of $916.5 million at a 21.8% margin, up 320 basis points
- Annualized Intelligence Subscription revenue of $2,877.1 million, growing 6.6%
- 105% net dollar retention and 98% gross dollar retention
- Coverage of 90 countries and roughly 82% of the world's population, spanning more than $7.4 trillion in global consumer spend
- A data engine capturing 4 trillion data records per week
For 2026 the company guided to 5.0% to 5.3% organic constant-currency growth and a 23.5% to 23.8% adjusted EBITDA margin.
Read those retention figures carefully, because they say something a feature comparison will not: 98% gross dollar retention means almost nobody cancels. Retail measurement is infrastructure. Once your share numbers, your trade terms and your retailer conversations are denominated in a supplier's data, switching means renegotiating the meaning of the word "share" across your whole commercial organisation. That is a strength of the product and a fact about your leverage as a buyer.
The GfK combination, completed 11 July 2023, extended NIQ from fast-moving consumer goods into tech and durables, which is precisely the ground Circana holds.
Circana: the widest category coverage
Circana is what happened when IRI and The NPD Group merged, completing in August 2022 and adopting the Circana name in March 2023. Hellman & Friedman had acquired NPD in 2021; the combined business is held by Hellman & Friedman, Vestar, New Mountain Capital and management.
Circana's distinguishing feature is breadth of category rather than depth in any single one. It compiles point-of-sale transactions from 290 million households globally across a $2.9 trillion consumer and retail market, organised into eight segments: food and beverage, fashion, home, technology, beauty and health, alcohol and tobacco, entertainment, and auto parts.
If your portfolio crosses the CPG/durables line, or if you sell something that Nielsen historically never tracked well (toys, books, apparel, footwear, video games, restaurant traffic), Circana is usually the shorter path to a defensible number.
Numerator: the only one that knows who
NIQ and Circana are, at their core, store-centric. The atomic record is a product in a store in a week. You can slice it a thousand ways and it still never contains a person.
Numerator inverts that. It is shopper-centric: the atomic record is a receipt submitted by an identified, profiled household. Numerator's own materials describe proprietary zero-party purchase and survey data from more than one million households, operating in more than 50 countries, headquartered in Chicago under CEO Eric Belcher.
The ownership history is worth getting right because it is widely reported wrong. Kantar announced a definitive agreement to acquire Numerator from Vista Equity Partners on 19 April 2021, when Numerator had 2020 revenues of $176 million and core shopper panel revenue growing 57%. Numerator today states that it was "acquired from Vista Equity Partners in 2021 in a transaction led by Bain Capital", that it is owned by Bain Capital, and explicitly that it is not owned or operated by Kantar. In 2025 Numerator took on Kantar's Worldpanel business and rebranded it Worldpanel by Numerator.
Because Numerator observes the same household across retailers and categories, it can answer questions the store-centric datasets structurally cannot: who else does this buyer shop, what did they buy instead, what is the demographic profile of the people driving a share shift, are we gaining buyers or gaining volume from the buyers we already had.
That last question is the single most important question in consumer growth, and we will come back to it.
The measurement ladder
Here is the frame that makes the three-way comparison actually decidable. Data about customers sits on a ladder, and each rung has a different unit of analysis:
- Transaction. Item x store x week. NIQ and Circana core retail measurement. Enormous coverage, no people in it.
- Household. Basket x trip x household. Numerator, Worldpanel. Fewer observations, but now there is a persistent identity behind the purchase.
- Person. Respondent x occasion. Trackers, surveys, panels. Attitudes, awareness, stated consideration.
- Reason. Decision x cause. Interviews. Why this, why now, why not the other one.
Two things are true about this ladder, and together they explain most of the frustration insights teams feel.
First, coverage falls as you descend and cost per observation rises. A retail measurement subscription buys you effectively the whole market. An interview programme buys you forty conversations.
Second, decision value rises as you descend. Nobody ever changed a roadmap, a pack format, a claim or a price architecture because of a number. They changed it because of a reason.
So the ladder is priced upside down relative to decision value, and every insights function in the world has quietly organised itself around that fact: buy the top two rungs continuously, buy the bottom rung occasionally and only when a decision is big enough to justify the cost and the six-week wait.
For decades that was a rational allocation, because rung four genuinely required one trained human moderator per conversation, plus recruitment, plus scheduling, plus transcription, plus a coding pass. The cost was real. It is no longer.
What none of the three can tell you
This is the part vendor comparison pages leave out, and it applies equally to all three.
They measure buyers. Growth comes from non-buyers.
Bain & Company studied the buying habits of nearly 100,000 shoppers globally using Kantar Worldpanel data, alongside more than 600 brand growth projects over a decade. The finding, published by Guy Brusselmans, John Blasberg and Bruno Lannes, is blunt: the one thing all leading brands have in common is that they lead their categories in penetration, not loyalty. In their indexed comparison of a leading brand against the average of its top 20 competitors, penetration ran from roughly 2.4x to 13.1x, while purchase frequency and repurchase rate clustered between about 1.0x and 2.3x. Loyalty barely moves. Penetration is the swing variable.
This aligns with the Double Jeopardy law formalised by Ehrenberg, Goodhardt and Barwise in the Journal of Marketing (1990, 54(3), 82-91): smaller brands are punished twice, with both fewer buyers and slightly less loyal ones.
Now put those two facts next to your data subscription. Penetration is a statement about people who do not currently buy you. Your retail measurement data contains the shelf. Your household panel contains buyers. The population that determines whether you grow is, by construction, the population your data has the least to say about.
Bain also found the bucket leaks: it is not unusual for a large majority of a brand's buyers in one year not to buy it the next, and even top brands can experience churn of nearly 50%. Their estimate is that a brand doing everything right adds roughly 1% penetration a year, meaning a brand at 10% takes close to 15 years to reach 25% - while few companies plan beyond the next 12 months.
They report the change, not the cause. Every one of these datasets is excellent at telling you that units fell 6% in the Northeast in the last four weeks. None can tell you whether that is a pack-size problem, a shelf-position problem, a competitor's promotion, a price threshold you crossed, or a claim on the front of pack that stopped making sense. Those are the five possible actions, and the data is silent on all of them. For a structured way to move from a number to a cause, see our guide to root cause analysis for customer research and the 5 Whys technique.
They are a census of purchase, blind to rejection. The person who picked up your product, read the label and put it back is not in any of these datasets. Neither is the person who never entered the category. See survivorship bias in customer research for why this systematically flatters your conclusions.
Nobody publishes pricing. None of the three lists a price. Contracts are annual, negotiated, scoped by category, market and module, and typically renewed rather than re-bid, which is exactly what a 98% gross dollar retention rate looks like from the buyer's side.
Where Koji fits
Koji is not a replacement for retail measurement. If you need share, distribution or the number your retailer will accept in a line review, buy the number from NIQ or Circana.
Koji owns the rung the ladder was always too expensive to reach. It runs AI-moderated voice interviews at panel scale: the AI conducts each conversation, probes the interesting answer instead of moving to question four, and does it with hundreds of people simultaneously.
What that changes in practice:
- You can ask non-buyers. Recruit category buyers who do not buy you, lapsed buyers, and rejecters, and ask them directly what happened. This is the population your subscription structurally excludes and the one penetration growth depends on. Our guides on reaching hard-to-reach audiences and win-back interviews with lapsed customers cover the recruiting design.
- You get the number and the reason in one instrument. Koji supports six structured question types -
open_ended,scale,single_choice,multiple_choice,rankingandyes_no- inside the same conversation. The closed questions give you a chart you can put next to your syndicated data; the open ones, with AI follow-up probing, give you the causal account that makes the chart actionable. - No moderator bias. The AI asks every participant the same way. There is no interviewer who leads the witness in interview 30 because they already believe the answer.
- Automatic thematic analysis and one-click reports. Themes, quotes and segment cuts are generated, not hand-coded over three weeks.
- Hours, not quarters. From question to insight in a day is normal, which means a share anomaly can be diagnosed inside the same reporting cycle in which it appeared.
The honest positioning: NIQ, Circana and Numerator answer "what happened". Koji answers "why, and what would change it". You need both, and until recently the second one was rationed by cost. It no longer has to be.
For category-specific playbooks, see AI-powered customer research for CPG and consumer goods brands and AI customer research for retail.
How to choose
Choose NIQ if you need global CPG share as a governing currency, you operate across many markets, and you want one supplier whose numbers your retailers and your board already accept.
Choose Circana if your portfolio spans durables, discretionary, entertainment or foodservice as well as CPG, and you want one supplier who covers the whole thing rather than two who each cover half.
Choose Numerator if your hardest questions are about shoppers rather than shelves: source of volume, cross-retailer behaviour, buyer demographics behind a share move, and switching.
Add Koji in all three cases, because none of them will ever tell you why. Budget for it as diagnosis rather than measurement, and use it the moment a tracked number moves in a direction you cannot explain.
Frequently Asked Questions
Is NielsenIQ the same as Nielsen?
No. Nielsen split into two businesses. The media measurement business kept the Nielsen name; the consumer intelligence business was acquired by Advent International in 2021 and operates as NIQ, which listed publicly in July 2025 and reported FY2025 revenue of $4,198.4 million. Note also that "Nielsen" in a UX context usually refers to the unrelated Nielsen Norman Group.
Is Numerator owned by Kantar?
No. Kantar announced a definitive agreement to acquire Numerator from Vista Equity Partners on 19 April 2021, but Numerator states that the transaction was led by Bain Capital, that it is owned by Bain Capital, and that it is not owned or operated by Kantar. In 2025 Numerator took over Kantar's Worldpanel business and rebranded it Worldpanel by Numerator.
What does retail measurement data cost?
None of the three publishes list pricing. Contracts are annual, negotiated by category, market and module, and renewed rather than re-bid. NIQ reported 98% gross dollar retention for FY2025, which indicates how rarely these subscriptions are cancelled once a company's commercial reporting depends on them.
Can syndicated data tell me why my sales dropped?
No. Scanner and panel data record what was purchased, not why. They can localise a decline to a market, retailer, pack or period, which narrows the question, but the causal explanation has to come from asking people. That is a different instrument, and it is the one AI-moderated interviews make affordable at scale.
Do I still need a household panel if I have retail measurement?
Usually yes, because they answer different questions. Retail measurement tells you what moved off shelves; a household panel tells you which people moved it, whether a share gain came from new buyers or heavier existing buyers, and where the volume was sourced from. Since penetration rather than loyalty drives brand growth, the buyer-level view is often the more strategically important of the two.
How is Koji different from these three providers?
NIQ, Circana and Numerator sell observed purchase data on annual subscriptions. Koji runs AI-moderated voice interviews with people you choose, including non-buyers and lapsed buyers who appear in none of those datasets, and returns thematic analysis and reports within hours. It complements rather than replaces measurement: they supply the number, Koji supplies the reason.
Stop waiting a quarter to find out why
Your measurement subscription will tell you that something changed. It will not tell you what to do about it, and it will never introduce you to the people who chose someone else.
Koji runs AI-moderated interviews with buyers, lapsed buyers and non-buyers, analyses them automatically, and gives you a shareable report the same day. No moderator to book, no coding backlog, no research background required.
Start free with 10 credits and run your first study this week.
