Kantar is the largest marketing data and analytics business in the world, and in 2026 it is also a company in the middle of being taken apart. Koji is an AI-native customer research platform that runs moderated voice and text interviews with your own customers and returns coded themes, charts and quotes in hours. They are not the same kind of purchase, and the honest comparison matters more than the marketing on either side.
Short answer: choose Kantar when you need syndicated measurement, nationally representative panels, audience currency data, or a globally comparable brand tracker across dozens of markets, and you have both the budget and the project runway to buy it. Choose Koji when you need to understand why a number moved, and you need that understanding this week rather than next quarter. A large number of insights teams now run both, because they solve genuinely different problems.
What Kantar actually is in 2026
Kantar is not a software product. It is a full-service insights organisation built through decades of acquisition, including Millward Brown and Taylor Nelson Sofres, and it sells research as a service delivered by people.
The scale is real and worth stating plainly:
- Kantar operates in more than 90 markets and reports serving more than half of the Fortune 500.
- Headcount data from Revelio Labs puts the Kantar Group at roughly 29,000 employees as of September 2025. Kantar itself has cited a figure above 25,000.
- Independent company trackers put group revenue in the region of 4 billion US dollars a year.
That scale buys you things no software subscription can replicate: continuous panels, audience measurement accepted as currency by media buyers, and normative databases built from decades of comparable studies. If your question is "how does our brand equity compare to the category norm in 14 countries", Kantar is a serious answer and Koji is not.
The ownership situation you should know about before signing a multi-year contract
Kantar is owned by Bain Capital, which holds roughly 60 percent, and WPP, which holds the remaining 40 percent. Bain acquired its stake in a 2019 transaction that valued Kantar at approximately 4.0 billion US dollars.
In 2025 and into 2026, both owners have been reported to be exploring an exit. Sky News reported that they were seeking a sale of the Worldpanel division for 5 billion pounds or more, around 6.5 billion US dollars, while the Financial Times reported that Kantar was set to be broken up and sold outright. Kantar had acquired Numerator in 2021 and merged it with Worldpanel in January 2025, creating a combined business of roughly 5,800 employees now operating under the Numerator brand.
None of this makes Kantar a bad supplier. It does make a specific procurement point: a brand tracker is a multi-year commitment to methodological continuity, and methodological continuity is exactly what tends to wobble when a business is carved into pieces and sold. Ask directly which legal entity will hold your contract in three years, and whether the panel behind your tracker sits inside the division being sold.
What Koji actually is
Koji is an AI-native research platform. You write a brief, Koji generates an interview guide, and an AI moderator conducts the interviews itself, by voice or by text, in the participant's own language. It probes. When a respondent says something unexpected, the moderator asks the follow-up that nobody wrote into the discussion guide.
The parts that matter in a comparison with an agency:
- AI-moderated voice interviews. Real conversations at panel scale, running in parallel, at any hour, with no moderator calendar to negotiate.
- Six structured question types. Interviews are not only open-ended talk. Koji supports
open_ended,scale,single_choice,multiple_choice,rankingandyes_no, so one study returns an NPS distribution and the explanation behind it. See the structured questions guide for how the quantitative and qualitative sides combine. - Automatic thematic analysis. Every open-ended answer is coded with descriptive and in-vivo labels grounded in the participant's verbatim words, then clustered into a canonical codebook across interviews. The method is the same one described in our complete guide to thematic analysis, executed automatically.
- One-click reports. Charts for the structured questions, themes and quotes for the open ones, with every claim traceable back to the transcript that produced it.
- No moderator bias. The AI asks every participant the same core questions in the same way. It does not get tired at interview 30, and it does not lead a witness because it already believes it knows the answer.
Koji vs Kantar at a glance
| Dimension | Kantar | Koji |
|---|---|---|
| Model | Full-service agency plus the Kantar Marketplace automated platform | Self-serve AI-native platform |
| Typical time to insight | Weeks to months for custom work; Marketplace markets itself as hours for automated products | Hours to days, including analysis |
| Who runs the interview | Human moderators and agency researchers | AI moderator, voice or text, running in parallel |
| Who you talk to | Kantar panels and recruited samples | Your own customers, or a sample you bring |
| Analysis | Delivered by the agency team | Automatic thematic coding plus structured charts |
| Entry price | Enterprise contracting; third-party estimates put custom engagements in the tens to hundreds of thousands | 29 euro or 79 euro per month, 10 free credits to start |
| Strongest at | Syndicated measurement, audience currency, global brand trackers, category norms | Explaining why a metric moved, discovery, churn diagnosis, message testing |
| Weakest at | Speed, iteration cost, small questions | Nationally representative quant at scale, audience currency, syndicated norms |
The gap that actually costs teams money: measurement is not diagnosis
This is the distinction that decides most Kantar-versus-alternative debates, and it is rarely stated clearly.
Kantar is exceptional at measurement. A tracker tells you that unaided awareness fell 4 points in Spain, that your ad scored below the category norm, that penetration slipped in a shopper panel. Measurement is a number attached to a population, and buying it from a firm with 90 markets and decades of norms is a defensible decision.
But measurement never explains itself. The number tells you something changed. It does not tell you which of the six plausible causes is the real one. And the moment you want the cause, you are commissioning a new piece of qualitative work, which restarts the clock: brief, proposal, sign-off, recruitment, fieldwork, analysis, debrief.
Independent industry reporting continues to describe traditional agency study cycles running four to six weeks, with third-party estimates of Kantar engagements citing 4 to 12 week project timelines and annual contracts from 50,000 to 500,000 US dollars or more. Treat those as directional rather than official, because Kantar does not publish a rate card. The direction is what matters: the diagnosis loop is measured in weeks, and the decision that needed the diagnosis is usually measured in days.
So the real cost of the measurement-diagnosis gap is not the invoice. It is the decisions that get made without the diagnosis, because waiting six weeks was not an option and somebody in the room had a confident opinion.
Koji is built for that specific loop. The tracker still tells you awareness fell in Spain. Koji tells you why, from 40 conversations with actual Spanish customers, before the end of the week.
Where Kantar is still the right call
Being honest here is more useful than pretending otherwise.
- Audience measurement and currency data. If media money is traded against a number, you buy that number from the organisation the market accepts.
- Nationally representative quant with published methodology. When a regulator, a board or a court needs the sampling frame defended, syndicated infrastructure is worth what it costs.
- Global comparability across many markets at once. A 20-market tracker with consistent norms is genuinely hard, and Kantar has spent decades building the comparability.
- Category norms and historical benchmarks. You cannot generate a decade of comparable prior studies with any platform, at any price. Norms are an asset that only accumulates.
- You need someone accountable. An agency absorbs methodological risk and puts a named human on the hook. That transfer of responsibility is a real product, and some organisations are right to buy it.
For a wider view of the firm landscape, see our roundup of the best market research companies in 2026.
Where Koji wins
Time to why
The comparison that matters is not fieldwork speed. Automated platforms, Kantar Marketplace included, have already compressed time to field. The weeks that hurt live in the second latency: time to understanding. A study that fields in two days and returns an analysed debrief four weeks later has not solved your problem. Koji collapses fielding and analysis into the same window because the coding happens as the interviews land.
Cost per question asked
An agency engagement makes every question expensive, so teams over-load the brief and ask everything at once. A Koji study costs credits: text interviews cost 1 credit, voice interviews cost 3, and a report refresh costs 5. That changes behaviour. You run a tight study, read it, and run the sharper follow-up study the next day, which is how research actually produces understanding.
The follow-up nobody planned
A survey commits 100 percent of its questions before a single answer arrives. An AI-moderated interview commits roughly the guide and generates the rest live. The difference between the two is discussed in more detail in our surveys vs interviews guide and in the practical breakdown of qualitative vs quantitative research.
Talking to your customers, not a panel
Kantar samples populations. Your churn problem lives with the 41 accounts that left last quarter, and no panel contains them. Koji runs with your own list. That distinction drives most of the value in win-loss work, onboarding diagnosis and pricing objections, and it is why voice or text mode selection matters more than sample size for these questions.
What each one costs
Kantar does not publish pricing. Engagements are scoped and quoted, Kantar Marketplace offers self-serve, guided self-serve and full-service tiers, and the number depends on markets, sample and method.
Koji publishes everything:
- Insights: 29 euro per month, including 29 credits.
- Interviews: 79 euro per month, including 79 credits.
- Annual billing at 290 euro and 790 euro, which is two months free.
- Extra credits at 1 euro each, with volume packs at 50 credits for 50 euro, 100 for 95 euro and 250 for 225 euro.
- Enterprise is custom.
- 10 free credits on signup, no card required.
One detail worth calling out because no agency contract works this way: Koji applies a quality gate, and only conversations scoring 3 or above consume a credit. You do not pay for a junk interview.
Which should you choose?
Buy Kantar if your question is about a population, needs to be defended externally, must be comparable across many markets, or has to be benchmarked against category norms you cannot build yourself.
Buy Koji if your question is about your customers, needs an answer this week, and is really a "why" question wearing a "what" costume. If you are sizing a market or planning a launch, the sequencing guides for go-to-market research and TAM SAM SOM for product researchers show where each kind of evidence belongs.
Run both if you are a large brand. The tracker measures. Koji explains. The combination is far stronger than either one arguing it can do the other's job.
Frequently Asked Questions
Is Koji a replacement for Kantar?
Not entirely, and any vendor claiming otherwise is selling. Koji replaces the custom qualitative and diagnostic half of an agency relationship: churn analysis, message testing, discovery, pricing objections, win-loss. It does not replace syndicated audience measurement, currency data or nationally representative quant with a defended sampling frame. Teams that need both usually keep both and stop paying agency rates for the fast diagnostic work.
How much does Kantar cost compared to Koji?
Kantar does not publish a rate card, and any figure you see online is a third-party estimate. Independent estimates commonly cite annual contracts from 50,000 to 500,000 US dollars with 4 to 12 week timelines for custom work. Koji publishes its pricing: 29 euro or 79 euro per month, extra credits at 1 euro, and 10 free credits to start. The structural difference is that Koji prices per conversation rather than per project, so a follow-up study costs credits rather than a new statement of work.
Is Kantar being sold, and does that affect my contract?
Both owners have been publicly reported to be exploring exits. Bain Capital holds around 60 percent and WPP around 40 percent, and reporting in 2025 described a possible sale of the Worldpanel division at 5 billion pounds or more as well as a wider break-up. Nothing here has been confirmed as final. The practical implication for a buyer is to ask which entity holds your contract, whether your panel sits in a division being divested, and what happens to methodological continuity if it changes hands mid-tracker.
Can Koji do quantitative research, or is it only interviews?
Koji runs both in a single study. Alongside open-ended conversation it supports five structured question types, scale, single_choice, multiple_choice, ranking and yes_no, which aggregate into charts exactly as a survey would. The difference is that the AI moderator can probe a structured answer immediately, so a 3 out of 10 satisfaction score arrives with the reason attached rather than as an anonymous data point.
How fast can Koji actually return results?
Interviews run in parallel rather than sequentially, so fielding is limited by how quickly participants respond rather than by moderator availability. Thematic coding happens as interviews complete, which means the report is largely built by the time fielding closes. For most studies the practical answer is same day to a few days, against the four to six week cycle that industry reporting still attributes to traditional agency projects.
Does an AI moderator produce lower quality interviews than a human?
It produces different quality. A skilled human moderator is still better at reading a room, handling a hostile participant and improvising a genuinely creative line of questioning. An AI moderator is better at consistency: it asks every participant the same core questions the same way, it does not fatigue at interview 30, and it does not unconsciously lead participants toward the hypothesis the client is hoping for. For diagnostic work at volume, the consistency usually matters more than the artistry.
Run the diagnostic half yourself
If you already buy measurement, the expensive gap is the explanation. Koji closes it in hours instead of weeks, with your own customers, and with every theme traceable to the words that produced it.
Start free with 10 credits. No card, no scoping call, no statement of work. Write a brief, launch a study, and read coded themes today. If it does not answer your question better than a six-week project would, you have lost nothing.
Related reading: Koji vs YouGov (2026) covers what a 32.4 million person standing panel can and cannot diagnose when a tracked metric moves.
