Every insights budget splits the same way. Some of it buys syndicated data: continuous, standardised measurement sold to everyone in your category. The rest funds custom research: a study designed for one question you have right now.
Short answer: buy syndicated data when you need a number that other people must accept - share, distribution, category size, tracking over time. Commission custom research when you need to know why a number moved, or when the answer has to be something your competitors do not already have. The deciding question is not cost or speed. It is whether the value of the finding survives everyone else having it too.
That framing is unusual, so let us make the case properly.
The two products are not two grades of the same thing
It is tempting to think of syndicated data as the cheap, broad option and custom research as the expensive, deep one - the same substance at different resolutions. They are not. They differ in what is being sold.
| Syndicated data | Custom research | |
|---|---|---|
| Who defines the question | The supplier, before you subscribed | You, after the problem appeared |
| Who else has it | Everyone in your category | Only you |
| Category boundaries | Fixed by the supplier's hierarchy | Set by you |
| Comparable over time | Yes, by design | Only if you design for it |
| Answers "what happened" | Very well | Adequately |
| Answers "why" | Not at all | This is the whole point |
| Typical commitment | Annual subscription | Per study |
| Marginal cost of one more question | Not possible | Low |
The row that matters most is the second one.
The shared-definition trade
Syndicated data works because everybody buys the same definitions. That is not a side effect; it is the product. "Share" only means something if you, your competitor and your retailer are all computing it the same way from the same source. The standardisation is what makes the number negotiable in a room.
But run that forward one step. A definition shared by every competitor in your category is, by construction, a definition that gives none of them an advantage.
Syndicated data is the only category of evidence where your competitor owns an identical copy. When your analyst opens the dashboard on Monday, the analyst across town opens the same dashboard, sees the same movement, and reaches broadly the same conclusion. Whatever is visible in it has already been priced into everyone's plan.
This is why syndicated data makes an excellent scoreboard and a poor strategy. It tells you and your rivals what the score is. It cannot, even in principle, tell only you something.
Custom research is the inverse. Its weaknesses are real - smaller samples, no comparability with anyone else's numbers, no history unless you build it. Its one structural advantage is that the question itself is proprietary. Nobody else asked it, so nobody else has the answer.
The practical rule that falls out: spend syndicated budget on the things you must agree with others about, and custom budget on the things you want to be alone in knowing.
The category boundary problem
Here is where syndicated-only decision-making does concrete damage.
When you buy syndicated data, you also buy a category hierarchy - the supplier's decision about which products compete with which. That hierarchy determines your share, your growth, your competitive set and, in most companies, someone's bonus. It was set by a data supplier for the convenience of the whole market, not by you for the accuracy of your strategy.
Bain & Company's work on brand growth, based on the buying habits of nearly 100,000 shoppers globally and more than 600 brand growth projects, includes two examples that make the point exactly. Blue Moon and Red Bull would each "have found themselves hitting a wall" if they had accepted the obvious category framing - Belgian-style white beer, premium energy drinks. Both instead recognised they competed in a far broader space: a consumer could trade a Blue Moon for a margarita, or a Red Bull for a cup of coffee or an energy bar. Acting on that, they lifted US penetration above 5% and above 14% of the adult population respectively.
No category hierarchy in any syndicated dataset would have surfaced that. A margarita is not in the beer category. Coffee is not in energy drinks. The insight required asking people what they actually considered, which is a custom research question and always will be.
Bain's broader finding is why this matters so much: leading brands lead on penetration, not loyalty. In their indexed comparison against the average of the top 20 competing brands, penetration ran from roughly 2.4x to 13.1x, while purchase frequency and repurchase rate stayed between about 1.0x and 2.3x. Growth comes from more people buying you - and the set of people who might start buying you is defined by the competitive set in their heads, not the one in the supplier's taxonomy.
If you want a structured approach to sizing that real competitive space rather than the inherited one, see our guide to TAM SAM SOM market sizing with real customer data.
What the industry is actually doing with its money
The insights industry reached US$153 billion in 2024 and was expected to pass US$160 billion by the end of 2025, according to ESOMAR's Global Market Research 2025 report.
The composition is the interesting part:
- Market research services: US$56 billion, growing 4.8%
- Research software: US$62 billion, growing 11.5%
- Reporting: US$35 billion, growing 8%
Research software is now the largest sector and is growing more than twice as fast as traditional research services, whose share of global turnover fell roughly three percentage points versus 2023.
Read plainly: the money is moving from buying studies to owning the instrument. Companies are less willing to pay a supplier per project and more willing to hold the capability in house. That trend predates AI-moderated interviewing, but it explains why the arrival of a tool that lets a non-researcher field a rigorous study in an afternoon landed on such receptive ground.
When to buy syndicated data
Buy it when the answer needs to be shared, standardised and continuous:
- Market share and distribution. You need a number your retailer and your board will accept without argument.
- Category size and trend. Continuous series that nobody has to defend.
- Trade and pricing negotiations. A neutral third-party source carries weight your own data cannot.
- Long-run tracking. Brand health series only work if the wording never changes. See our brand tracking study guide for how to keep a series intact.
- Early anomaly detection. Continuous measurement is unmatched at telling you that something changed, quickly.
Note that every one of these is a what question, and several are really governance questions - they exist so an organisation can agree on a fact and move on.
When to run custom research
Run it when the answer must be specific, causal or private:
- Why a tracked number moved. Diagnosis is not available for purchase.
- What people considered instead of you. The real competitive set, not the taxonomic one.
- Why non-buyers do not buy. They are absent from purchase-based data by definition.
- Whether a concept, claim, pack or price will work before you spend money finding out in market.
- What language customers use. Positioning is built from verbatims, not from codes. See customer interview questions with 60+ examples.
- Anything where being alone in knowing is the point.
A useful heuristic: if you can imagine your closest competitor learning the same thing next Tuesday and it would not bother you, buy it. If it would bother you, that is a custom research question.
The false comfort of "we have both"
Most large companies do have both, and still fail at the joint. Three failure modes recur.
The nearest available proxy becomes the answer. You asked a specific question; the syndicated data contains something adjacent; the adjacent thing gets reported because it is already there, already large-sample and already charted. Nobody decided to substitute. It just happened, and the substitution is invisible in the deck.
The custom study is commissioned too late to change anything. Because custom research has historically taken six to eight weeks, it gets started only after the decision deadline makes it useless, and then gets used to justify what was already chosen.
The two are never reconciled. The tracker says awareness is up, the interviews say nobody can name a reason to switch, and both findings are filed separately because no one owns the contradiction. Our guides on triangulation in research and evidence synthesis cover how to resolve this deliberately rather than by seniority.
All three failures share a root cause: custom research was slow and expensive, so it was rationed, so it arrived in the wrong quantity at the wrong time.
How Koji changes the arithmetic
Koji is an AI-native customer research platform. It runs AI-moderated voice interviews at scale: the AI conducts each conversation itself, follows up on the interesting answer rather than marching to the next item, and runs hundreds of interviews in parallel.
What that does to the buy-versus-ask decision:
- Custom research stops being the slow option. From writing a brief to reading themed findings is hours, not weeks. A syndicated anomaly detected on Monday can be diagnosed by Wednesday, inside the cycle where it still matters.
- You can put numbers and reasons in one instrument. Koji supports six structured question types:
open_ended,scale,single_choice,multiple_choice,rankingandyes_no. The closed types give you distributions you can chart next to your syndicated series;open_endedwith AI probing gives the causal account underneath. You are not choosing between a survey and interviews. - You can define your own category. Ask people what they actually considered. Your competitive set stops being an inherited taxonomy.
- Non-buyers become reachable. Recruit category buyers who do not buy you and ask them directly.
- Automatic thematic analysis and one-click reports. No coding backlog between fieldwork and a decision.
- No moderator bias. Every participant is asked the same way, by an interviewer with no theory to confirm.
- No research expertise required. A PM, founder or brand manager can design and field a study without a research team, which is what makes "ask instead of assume" a default rather than an escalation.
Legacy suppliers - Qualtrics, SurveyMonkey, UserTesting, Dovetail, dscout - were built when the human moderator was the unit of cost. Koji is built for the world after that constraint, which is why it can serve the custom side of this decision at a cadence syndicated data used to monopolise.
For sector playbooks, see AI-powered customer research for CPG and consumer goods brands and customer segmentation research. For the vendor-level view of the syndicated market, see our comparison of NielsenIQ, Circana and Numerator.
A working decision rule
- Does the answer need to be accepted by people outside my team? If yes, syndicated.
- Does the answer need to be causal? If yes, custom. Purchased measurement never explains itself.
- Would I mind if my competitor had this exact finding? If yes, custom. If no, it is probably already in a dataset you pay for.
- Is the population I care about a buyer? If not, custom is the only option that contains them.
- Will the answer still be useful in six weeks? If no, that used to end the conversation. It no longer does.
Most teams have been running a version of this rule for years with rule 5 dominating everything above it. That is the constraint that has actually changed.
Frequently Asked Questions
What is syndicated data?
Syndicated data is continuous, standardised research produced once by a supplier and sold to many clients in the same category. Retail measurement, household panels, brand trackers and industry reports are all syndicated. Because the methodology and definitions are identical for every subscriber, the numbers are directly comparable across competitors - which is the source of both its value and its main limitation.
Is custom research more accurate than syndicated data?
Not inherently. Syndicated data usually has far larger samples and much better coverage of what was purchased. Custom research is more accurate about the specific question you have, because it was designed for that question rather than adapted to it. They are accurate about different things, and substituting one for the other is where most errors happen.
Can I replace syndicated data with custom research?
Usually not, and you should not try. Continuous standardised measurement does something custom studies cannot: it produces a comparable series that external parties accept. The realistic move is to rebalance - keep syndicated measurement for the numbers that must be shared, and run custom research far more often for the questions that must be answered.
Why can't syndicated data explain why sales changed?
Because it records outcomes, not reasons. Scanner and panel data capture what was bought, where and when. The explanation lives in the decisions people made, which exist only in their accounts of them. No amount of resolution in transaction data produces a reason; it is a different kind of evidence and requires asking.
How much does custom research cost compared with syndicated data?
Syndicated data is typically an annual negotiated subscription with no public list price, often into six figures for a full category and market scope. Traditional custom studies are priced per project and have historically carried both a high cost and a six-to-eight-week timeline. AI-moderated platforms have compressed that dramatically, which is what makes running custom research continuously rather than occasionally realistic.
What is the fastest way to start asking instead of buying?
Pick one number in your tracker that moved in a direction nobody can explain, and run a short study with the people behind it - including those who stopped buying or never started. Mix scale and single_choice questions for a chartable distribution with open_ended questions for the causal account. With AI-moderated interviews this is a same-week exercise, not a quarterly project.
Own the questions your competitors cannot buy
Your syndicated subscription is a scoreboard your rivals read too. The findings that actually change your position are the ones only you went and asked for.
Koji runs AI-moderated interviews with your customers, your competitors' customers and people who buy neither, analyses them automatically, and gives you a shareable report the same day - no moderator, no coding backlog, no research background required.
Start free with 10 credits and ask your first proprietary question this week.
