Research quality standards: Total quality, because someone will ask how you know.

Because of who does the work and how much of it there is. Every project has three senior specialists minimum, each with ten years or more, who moderate, code and analyse themselves rather than briefing someone cheaper. We buy the strongest panel for your audience rather than the cheapest one that can fill the quota, run a second cleaning pass on data the panel has already cleaned, and review everything twice before it reaches you. Every one of those is a cost we could remove, and each one is a place where research can go wrong without anyone noticing. On a decision worth six or seven figures, the research is the cheapest part of the risk.

The same three seniors who designed it. There’s no account manager sitting between you and the people doing the research, so nothing gets relayed, softened or lost on the way. You have three people to call and all three know the study, which in practice is faster than one coordinator who has to go and ask. Where a project needs a technical skill beyond the three, a data scientist for a complex model for instance, we bring in that specialist for that piece and tell you who they are. What never happens is the research being handed down to someone more junior so a senior can move on to the next pitch.

We don’t disappear. The evidence stays open to your team for as long as you need it, we stay reachable while the plan is executed, and when the market answers back we help you read what it actually said, including the parts that argue with us. What we can’t do is make an organisation move, and the scope document above says so in writing. Which priority wins the budget, who builds it and how fast, whether a decision survives the third meeting about it: that part is yours. We can put you on evidence that holds, argue for what we believe it means, and stay while you act on it.

By what has to be proved about the respondent, and we buy the strongest panel for that audience rather than the cheapest one that can fill the quota. General consumer work goes to first-party permissioned panels, because they own the relationship with the panelist and can tell us how that person was recruited. Healthcare goes to credential-verified panels, where identity, qualification and specialty are checked in three factors under ISO 20252. Experts, executives and hard B2B go to expert networks, which verify identity and run KYC but don’t verify professional credentials, so we use them where a credential isn’t what’s at stake. What we avoid wherever the audience allows it are marketplaces and exchanges, which resell inventory from thousands of panels they don’t own, so the chain of custody ends at their door. Where none of the above reach the person, we recruit by hand, and on the hardest projects the sample ends up mixed and you’re told which respondents came from where.

Yes, and probably more than many firms would admit. The line we hold is between scaffolding and structure: scaffolding helps you build, then it comes down, and it holds up nothing in the finished thing. So AI transcribes, and a person corrects the transcript before a word is coded. It translates, and someone who speaks the language checks it. It’s a sparring partner while we frame a study, arguing with our own hypotheses and pushing us towards angles we hadn’t tried. It fixes our grammar, sharpens our English, and tells us when a chart is doing a poor job of showing what we found. It searches thousands of pages of transcript in seconds, so a senior spends the afternoon thinking instead of looking. On a large study it can sketch a first map of the material or read our analysis again looking for contradictions, and a senior checks each proposal against the source before anything counts. What it never does is structural: it doesn’t run an interview, invent a respondent or a data point, decide what a finding means or write a conclusion. Most platforms ship far more than that, switched on, and we only turn on what has passed our own test against material coded by hand. One place where the line gets concrete: automated sentiment classification is reliable in around fifty languages and automated emotion classification only in English, so outside English we code emotion by hand and say so. Every use of AI in your project is named in your report, which is still rare enough in this industry to be worth saying out loud.

On participants, there are three layers of checks. At the panel, government ID validation and third-party identity tools. During fieldwork, more than 175 behaviours tracked passively per session, with fraudulent and disengaged respondents removed and replaced live. After fieldwork, our own pass against your specific questionnaire: straightlining, completion under half the median time, open-ends that answer a different question, and contradictions between paired questions. On findings, we rely on triangulation and review. A pattern that appears in one lens is a hypothesis, in two it’s evidence, and in all three it’s solid enough to build a decision on. Then a second senior reads the whole thing cold, hunting for weak claims and anything that doesn’t hold. Where the lenses disagree, we say so rather than force a story, because the disagreement is often the most interesting part.

Less than the logo implies. Every certificate carries a Statement of Applicability listing which annexes are in scope. A sample provider is typically certified for access panels, online self-completion and data processing, with telephone and face to face interviewing explicitly excluded. So we read that document for every partner, every time, and plan the fieldwork around what’s genuinely covered rather than around the badge on a homepage. It seemed only fair to publish our own, which is the scope document further up this page.

For quantitative work, four hundred is our floor for a readable total and around two thousand is where we stop, because above that the margin of error barely moves. The real driver is subgroup analysis: if six segments have to be read separately, the total is set by the smallest one, not by the headline. For qualitative, ten to fifty sessions depending on how many distinct audiences, markets and languages are in scope. A design that needs more than fifty usually needs to become two studies.

Yes, in two ways, and neither of them involves cutting checks. The first is free: narrow the question. A study that has to answer one decision moves faster than one carrying three, and we’ll tell you which of the three is really urgent. The second costs money: on a genuine deadline we can bring in more senior specialists and run stages in parallel that would normally run in sequence, which buys weeks and raises the price, because parallel work needs more people and more coordination, not fewer checks. We quote that openly rather than absorbing it into a worse study without telling you. Either way you get a fixed date at proposal stage and we meet it. Three weeks is the fastest a complete study can honestly run, brief to report, and that assumes an audience the panel reaches quickly and no surprises in the screener. Large multi-market programmes across three lenses run to about four months, and the stage that moves least is recruitment, because a niche B2B or regulated healthcare target takes as long to reach as it takes. Speed you’ve paid for is fine, but speed taken out of the evidence is the expensive kind, because the decision gets taken twice.

Yes. Every interview in video or audio, every transcript, every source, every report and dataset, in the Client Hub, with access levels set by role. Access is indefinite by default and can be shaped by contract when you need something specific. This matters more than it sounds: a report is one reading of the evidence, and the evidence should outlive the person who presented it.

Yes, and it’s common on trackers, where comparability with previous waves matters more than the quality of any single new one. We’ll audit the source the same way we audit our own: ESOMAR 37 answers, certificate scope, verification method, and what the supplier will actually commit to in writing. If that audit turns up something you should know, we tell you, even when the supplier is already yours and the news is inconvenient.

It becomes our problem. If a panel underdelivers on a quota, a platform breaks or a partner misses a date, we solve it, re-field where needed, and absorb the coordination. You have one team to call, whatever went wrong and whoever caused it, and you get told. A quota met by relaxing a screening criterion isn’t a quota met, and you’ll hear about that trade while you can still decide on it.