Experience Research Platforms

Two capabilities appear in Forrester's current evaluation of this market that, taken together, describe something the research profession has not finished arguing about.

The first is the AI moderator: an agent that interviews human participants at scale, in real time, adapting its questions based on what people say.

The second is synthetic responses, which Forrester records in its assessment of Qualtrics as a way to accelerate reach to a target audience.

One automates the interviewer. The other automates the respondent. A market that exists to collect qualitative data from humans is being sold tooling that can, in principle, remove the humans from both chairs.

Neither capability is illegitimate and both solve real problems. But the distance between them is where the interesting question in this category sits.

What an experience research platform is

Forrester defines experience research platforms as platforms companies use to collect qualitative data to weigh alongside quantitative data when making decisions about products or services.

The operative word is alongside. This category exists because numbers describe what happened and not why, and product decisions made on behavioural data alone are decisions made without knowing the reasoning behind the behaviour.

Practically, these platforms handle participant recruitment, remote research sessions, usability studies, live interviews, diary studies, research communities, and the analysis of what comes out. Forrester's framing is that they help organisations reach insight quickly, understand customers in depth, reduce risk in product and service decisions, and scale research beyond what a small internal team can run.

That last point is the commercial engine. Most organisations have a handful of researchers and far more decisions than those researchers can inform. The platform's promise is that research stops being a bottleneck.

The field contracted hard

Forrester's coverage of this market shows an unusually steep narrowing.

The Q1 2023 Landscape mapped thirty nine vendors. The Q3 2025 Landscape mapped sixteen. The Q1 2026 Wave scored eight against thirty one criteria.

Thirty nine to sixteen in under three years is not gradual consolidation. It reflects a market where a large number of point tools, survey platforms, usability testing services, recruitment panels, and community managers, either merged, exited, or stopped being credible as standalone offerings once buyers started asking for one platform instead of five.

The vendor whose Leader position anchors the current Wave describes the same thing from the inside, characterising the previous state as a fractured landscape with data trapped in disparate tools.

That framing is self-serving and accurate. Research teams genuinely did run recruitment in one system, sessions in another, transcription in a third, and analysis in a spreadsheet, with the insight arriving too late to affect the decision it was meant to inform.

Inside The Forrester Wave: Experience Research Platforms, Q1 2026

The evaluation scored eight providers against thirty one criteria across current offering and strategy.

Discuss placed as one of only two Leaders and took the highest score in the current offering category, with maximum scores in fifteen criteria including innovation, roadmap, live interviews, and AI-powered research methods.

Qualtrics placed as a Strong Performer in its first appearance in this evaluation, earning top marks in the strategy category. Forrester recorded its vision as AI-powered research delivering instant contextual insight through agents managing the end-to-end research flow, alongside synthetic responses to accelerate reach to target audiences. Its qualitative capability arrived through a market research product launched in 2024, which brought research into the same platform as its existing customer and employee experience programmes.

Two Leaders in an eight-vendor field is a narrow top tier. Worth noting alongside this that Forrester has moved to a three-band Wave graphic, showing Leaders, Strong Performers, and Contenders rather than the previous four tiers, which changes how placements compare against older evaluations across every category.

The AI moderator is genuinely useful

It would be easy to treat automated interviewing as a corner-cutting exercise. The reported benefits argue otherwise, and they are specific.

Language is the first. An AI moderator can conduct interviews in multiple languages, which removes a constraint that has quietly shaped what gets researched. Organisations research the markets where they can afford moderators fluent in the language, which means smaller markets get less research and product decisions get made about them from data collected elsewhere.

Time zones are the second. Qualitative research is synchronous, and a human moderator in one region conducting depth interviews in another is either working nights or not doing it.

Scale is the third and most consequential. Depth interviews are expensive per participant, which caps sample sizes and pushes teams toward asking a few people a lot or many people a little. An agent that can run hundreds of adaptive conversations changes that trade-off.

There are real limitations underneath the enthusiasm. A skilled moderator notices hesitation, follows an unexpected thread, recognises when someone is performing rather than reporting, and builds enough rapport that a participant admits something unflattering. Whether an agent does any of that well is an empirical question the industry has not settled, and the answer probably varies enormously by research type. Usability testing, where the task structures the session, is a different proposition from exploratory work on why someone abandoned a category entirely.

Synthetic responses are a different claim

Synthetic responses mean model-generated answers standing in for human ones, and they belong in a separate conversation from AI moderation.

The case for them is practical. Reaching a specific target audience takes time and money, hard-to-recruit populations can take weeks, and a directional read in hours is genuinely valuable when the alternative is making the decision with no research at all. Used to pressure-test a discussion guide, prioritise which questions to ask real people, or sanity-check an early concept before committing to fieldwork, they are a reasonable tool.

The case against is epistemic and worth stating plainly. A model generating what a customer segment would probably say is producing a summary of patterns in its training data. Where those patterns hold, the output looks right, which is the problem. It will be most convincing precisely where it is least informative, because agreeing with the conventional view is what pattern-matching produces.

The reason to run qualitative research is to encounter the thing you did not anticipate. A synthetic respondent cannot supply that, because it has no access to the reality that would contradict the expectation. It can only recombine what has already been said.

There is a second-order risk that deserves more attention than it gets. Research findings inform decisions, decisions become published artefacts, and published artefacts become training data. A research practice that leans on synthetic responses is feeding its own prior assumptions back into the system that generates the next round.

None of this means the capability should not exist. It means the discipline around it has to be explicit: what synthetic data is used for, what it is never used for, and how it is labelled in the findings that reach a decision-maker. A slide that does not distinguish between what people said and what a model predicted people would say is not a research output.

Why small samples work, and when they stop working

There is a methodological point here that gets lost when qualitative research is discussed in the same breath as scale.

Qualitative research is not underpowered quantitative research. Interviewing eight people is not a failed attempt to survey eight hundred. The logic is different: you are looking for the existence of a pattern, a mental model, a workaround, an unmet need, rather than its prevalence. Once you have heard the same thing from enough people that new interviews stop producing new categories, you have what you came for.

That is why a well-run study with a small sample can be decisive, and it is also why scaling qualitative research is not straightforwardly better. More interviews produce diminishing returns once saturation is reached, and the value of the additional sessions is mostly reassurance.

Where scale genuinely helps is coverage: more segments, more markets, more contexts, each reaching saturation on its own terms. That is the argument for AI moderation and it is a good one. It is not an argument for treating qualitative volume as a proxy for confidence, and platforms that present large synthetic sample sizes invite exactly that confusion.

Where this leaves the research function

The contraction from thirty nine vendors to eight scored providers, and the arrival of platforms that own customer and employee experience data alongside research, point at where this is heading.

Research is being absorbed into the same infrastructure as everything else the organisation knows about its customers. Qualtrics' positioning is explicit about this, offering research that runs where CX and EX data already lives.

The upside is that findings stop being orphaned. A qualitative insight sitting in a slide deck influences nothing; the same insight attached to the segments and behavioural data it explains is actionable.

The risk is subtler. Research earns its value partly by being separate. A researcher whose job is to find out what is true, including when the answer is unwelcome, provides something that a system optimised for producing insight quickly does not. Speed and comfort tend to move together, and a research function embedded in the decision infrastructure is under pressure to deliver findings on the timetable of the decision rather than on the timetable of finding out.

That tension has always existed in corporate research. What is new is that the tooling now makes it possible to produce something that looks like research fast enough to satisfy any deadline, which removes the friction that used to force the conversation about whether the question was worth answering properly.

The platforms are not responsible for how they get used. But anyone building a research function around them should decide, before the first deadline, which parts of the process are negotiable and which are not.

Forrester does not endorse any vendor named here, and tier placement should not be read as a recommendation to buy.