YouGov is a panel company that sells continuously collected data. Koji is an AI-native research platform that runs moderated interviews with your own customers and returns coded themes the same day. The difference is not speed or price. It is whether the question was decided before you arrived.
Short answer: choose YouGov when you need nationally representative measurement, brand tracking against a category norm, or public-opinion data across many markets with a methodology that will survive outside scrutiny. Choose Koji when the people who hold your answer are your own customers, prospects and churned accounts, and when the question is why rather than how many. Most serious insights functions run both.
What YouGov reported for FY2025
YouGov is listed in London, so its numbers are published rather than estimated. For the year ended 31 July 2025:
- Group revenue of 388.9 million pounds, up 16 percent as reported but only 1 percent on an underlying basis, with most of the reported growth coming from the full-year contribution of the acquired CPS business, rebranded YouGov Shopper.
- Adjusted operating profit of 60.7 million pounds, up 22 percent, at a 16 percent margin against 15 percent the prior year.
- Data Products: revenue of 83.9 million pounds, flat as reported and up 1 percent underlying, with adjusted operating profit of 25.8 million pounds at a 31 percent margin.
- Research: revenue of 176.9 million pounds, flat as reported and up 1 percent underlying, with adjusted operating profit of 19.5 million pounds at an 11 percent margin.
- Shopper: revenue of 128.1 million pounds, with adjusted operating profit of 27.2 million pounds at a 21 percent margin.
- 32.4 million registered panellists, up 12 percent from roughly 29 million, across 63 markets.
- Stephan Shakespeare returned to the chief executive role in February 2025.
That is a well-run business with a genuinely rare asset. It is also a margin map, and margin maps are the most honest document any supplier produces.
The margin map tells you which product you will be offered
Look at the two research-side divisions side by side.
| Division | FY2025 revenue | Adjusted operating margin | What you are buying |
|---|---|---|---|
| Data Products | 83.9m pounds | 31 percent | Access to data already collected from the panel |
| Research | 176.9m pounds | 11 percent | A custom study designed around your question |
| Shopper | 128.1m pounds | 21 percent | Purchase behaviour data from the shopper panel |
Selling you data that already exists is roughly three times as profitable as going out and asking your question. That is not a criticism of anyone's sales practice. It is arithmetic, and it follows directly from the delivery model: a subscription product is written once and sold many times, while a custom study consumes researcher hours in fixed proportion to its size. We made the general version of this argument with published figures in Koji vs Ipsos, where the same coupling shows up as a 12 percent operating margin.
What matters for you is the gradient it creates. When you bring a question to any panel-first supplier, the commercial gradient runs toward "here is what we already measure that is close to what you asked" and away from "let us design a study around what you actually need to know." Often the nearest existing measurement is genuinely the right answer, delivered faster and cheaper than a custom study could manage. The risk is the case where the closest available proxy quietly becomes the answer, and nobody in the room notices the substitution because the data is real, the sample is large and the chart is excellent.
A panel is a fixed asset, and fixed assets shape products
Maintaining 32.4 million registered panellists across 63 markets costs money continuously, whether or not anyone commissions a study this week. The rational product on top of that asset is a subscription to data collected on a standing schedule: brand and reputation tracking, audience profiling, syndicated attitudinal measurement.
This produces a specific and underappreciated property. A survey commits every question before the first answer arrives, which is the classic limitation covered in our surveys versus interviews guide. Standing panel data goes one step further: the questions were finalised before you had your problem.
That is exactly what you want when the job is comparability over time. A tracker is only useful because the wording never changes; the moment you improve a question you break the series. It is exactly what you do not want when the job is diagnosis. Tracking will tell you, credibly and in 63 markets, that consideration among 25 to 34 year olds fell four points this quarter. It cannot tell you which of six plausible causes is real, because nobody was there to ask a follow-up.
Scale is a representativeness asset, not a relevance asset
32.4 million panellists buys something real: the ability to fill a nationally representative quota, in many countries, quickly, with weighting that a regulator or a journalist will accept.
It does not buy relevance to your specific commercial question, and the gap is widest exactly where the money is:
- Your churned accounts are not in the panel. Neither are your trial users who never activated, or the buyer who evaluated you and signed with a competitor last month.
- In B2B the incidence problem is decisive. If your buyer is a procurement lead at a mid-market logistics firm, the share of a general consumer panel that qualifies rounds to nothing, and screening cost rises accordingly. The mechanics of why low incidence makes sample expensive are covered in our breakdown of survey sample cost, CPI and incidence rate.
- Panel populations are self-selected and repeatedly surveyed. That is a manageable, well-documented condition rather than a scandal, and the effects are explored in professional respondents and panel conditioning and in our guide to sampling bias.
The honest summary: a panel is the right instrument when you want to know what a population thinks, and the wrong one when you want to know what your customers think.
What Koji does instead
You write a brief. Koji drafts an interview guide. An AI moderator conducts the interviews itself, by voice or by text, in the participant's own language, and probes when someone says something the guide did not anticipate.
- AI-moderated voice and text interviews running in parallel, at any hour, with no scheduling. Twenty conversations can happen simultaneously at nine in the evening.
- Six structured question types inside one study:
open_ended,scale,single_choice,multiple_choice,rankingandyes_no. This is the part that matters against a tracker: a single Koji study returns a satisfaction distribution and the reasons behind every score, from the same conversation. The structured questions guide covers how the quantitative and qualitative halves aggregate together. - Automatic thematic analysis with descriptive and in-vivo codes grounded in participants' own words, then clustered into one canonical codebook across the whole study, as described in the thematic analysis guide.
- One-click reports where every theme traces back to the transcript line that produced it.
- No moderator bias and no moderator fatigue. Interview 40 is conducted exactly like interview 1, which is what makes comparison across accounts clean.
- Customizable AI consultants that carry your context between studies, so the second study starts where the first finished. Our research repository guide covers why that compounding matters more than any single report.
Which one for which job
| The job to be done | Better fit | Why |
|---|---|---|
| Nationally representative opinion measurement | YouGov | Panel scale, weighting, published methodology, external credibility |
| Brand tracking with comparable history | YouGov | A tracker is valuable because the wording never changes |
| Public-opinion and political polling | YouGov | Specialist practice with a long public record |
| Category norms across many markets | YouGov | 63 markets and years of accumulated series |
| Diagnosing why a tracked metric moved | Koji | Follow-up questions are the whole instrument |
| Understanding why customers churned | Koji | Your own list, not a panel; results in days |
| Win-loss and evaluation interviews | Koji | Your CRM contacts, no scheduling, no recruitment premium |
| Message and positioning testing before launch | Koji | Cheap enough to iterate several rounds instead of buying one verdict |
| Continuous discovery for a product team | Koji | Per-conversation pricing makes weekly research economically sane |
The two are complements far more often than substitutes. The pattern that works: use tracking to find out that something moved, and interviews to find out why. Buying only the first leaves you with a very well measured mystery. This is the same division of labour we describe for analyst reports and for software review sites: secondary sources define the shape of a market, and primary work explains your position in it.
What each costs
YouGov does not publish a rate card. Data Products are sold as subscriptions priced by market coverage, category and seat count; custom Research is quoted per project. Its disclosed 83.9 million pounds of Data Products revenue is a subscription base, which tells you the commercial shape without telling you your number.
Koji publishes everything:
- Insights: 29 euro per month, including 29 credits.
- Interviews: 79 euro per month, including 79 credits.
- Annual plans at 290 and 790 euro, which is two months free.
- Credits: a text interview costs 1, a voice interview 3, a report refresh 5. Extra credits are 1 euro each, with packs at 50 for 50 euro, 100 for 95 euro and 250 for 225 euro.
- A quality gate: only conversations scoring 3 or above consume a credit. A junk interview is free.
- 10 free credits on signup, no card required.
Forty voice interviews with your own customers is 120 credits. The point is not that it undercuts a panel subscription, which is a different product. The point is that it is cheap enough to be iterative: run a study, read it, realise you asked the wrong question, and run the better one tomorrow. That loop is where understanding actually comes from, and it is the thing a fixed-question instrument cannot offer at any price.
How to decide
Ask one question: is the person who holds my answer in a panel, or on my customer list?
If your answer lives in a population, buy the population. YouGov is very good at populations and no platform is going to change that.
If your answer lives with people who have an invoice, a renewal date, a support history or a competitor's contract in their hands, then no panel contains them, and the study you need is a conversation. Do it yourself, in days, with the people who actually pay you. Start with the practical framing in our customer validation guide or the jobs-to-be-done interview method.
Related reading: for the suppliers behind standing panel and retail data, see NielsenIQ vs Circana vs Numerator, and for the buy-versus-ask decision itself, syndicated data vs custom research.
Frequently Asked Questions
Is Koji an alternative to YouGov?
For part of the portfolio, yes. Koji replaces the custom qualitative work that sits alongside tracking: churn diagnosis, win-loss, message testing, concept feedback, pricing objections and discovery with your own customers. It does not replace nationally representative measurement, political polling, or brand trackers with years of comparable history, all of which depend on panel infrastructure and consistent wording that a conversational instrument does not provide. Most teams that adopt Koji keep a panel relationship for the measurement half.
What did YouGov report for FY2025?
For the year ended 31 July 2025, YouGov reported group revenue of 388.9 million pounds, up 16 percent as reported but only 1 percent underlying, with adjusted operating profit of 60.7 million pounds at a 16 percent margin. Data Products revenue was 83.9 million pounds at a 31 percent margin, Research was 176.9 million pounds at an 11 percent margin, and Shopper was 128.1 million pounds at a 21 percent margin. The panel reached 32.4 million registered panellists, up 12 percent, across 63 markets.
Why does the margin difference between divisions matter to a buyer?
Because it tells you the commercial gradient inside the supplier. Data Products earn roughly 31 percent while custom Research earns roughly 11 percent, which means selling access to already-collected data is materially more profitable than designing a study around your question. Often the existing measurement genuinely is the best answer, faster and cheaper than a custom study. The risk is the case where the nearest available proxy quietly substitutes for the question you actually asked, and the quality of the data makes the substitution hard to notice.
Can a panel answer why a tracked metric moved?
Not directly. A tracker is valuable precisely because its wording is fixed, which is what makes quarters comparable, and that same property means no follow-up question was ever asked. You can cut the existing data by segment and generate hypotheses, but distinguishing between several plausible causes requires new fieldwork with the ability to probe. That is a qualitative job, and running it in days rather than weeks is the difference between a diagnosis that reaches the decision and one that arrives after it.
Does Koji work for B2B audiences?
Yes, and the advantage is largest there. B2B respondents are hard to schedule and expensive to recruit from panels because incidence is low, which is what drives cost per interview up. Koji interviews your own contacts, so recruitment is outreach rather than screening, and an AI moderator that runs at nine in the evening removes the single biggest barrier to senior participation. The consistency also helps: every account gets the same core questions, so comparison across them is clean.
How does analysis compare with a research supplier's team?
Every open-ended answer is coded during the study with descriptive and in-vivo labels grounded in the participant's own words, then near-duplicate codes are clustered into one canonical codebook across all interviews, while structured questions aggregate into charts in parallel. The output is a report where each theme links back to the transcript that produced it, so a skeptical stakeholder can verify any claim in seconds instead of trusting a summary slide. The practical difference is timing: the analysis is largely complete when fielding closes rather than several weeks later.
Answer the questions your tracker cannot
A tracker tells you the number moved. It will never tell you why, because the follow-up question was not in the questionnaire and there was nobody there to ask it.
Start free with 10 credits. Write a brief, launch a study with your own customers, and read coded themes today. No card, no scoping call, no subscription negotiation.
