AI changed the economics of research. Most research teams haven't caught up.

Two weeks ago I promised to close the agentic series, UXR for AI-native teams, with one last post on agents. Then I got into a discussion about democratization that I couldn't stop thinking about, and this post is what came out of it.

It's longer than my usual posts, and a bit meta. I've been circling these ideas for a long time and writing about them in pieces: how research teams accumulate instead of being designed, why more research is not turning into more research jobs, why hiring loops test the wrong things, and who decides what counts as evidence. This essay is where these thoughts and many more come together, around one question: what happens to research teams when research gets cheap?

The final part of the series is still coming. I promise. Again.

A PM can now run a study in days, and that raises a question research teams don't like

A product manager describes a research question to an AI tool. It writes a discussion guide. A platform recruits participants. An AI moderator runs the interviews, and a few days later she has a synthesized report with themes, quotes, and recommendations.

That used to take coordination, weeks, and a specialist. Now it takes days, sometimes hours.

That's a good change. It lowers the cost of understanding customers, gives research to teams that could never afford it, and gets questions answered that would otherwise have been skipped.

It also raises an uncomfortable question. If a PM can do all of that without a researcher, why does the company need a research team?

The profession's instinctive answer is that doing research and doing good research are different things. Researchers understand sampling, validity, bias, and interpretation, which is what separates evidence from a pile of interesting observations.

That's true. It also misses the point. An organization doesn't have to understand the value of methodological expertise to decide it no longer needs to pay for it.

If executives believe research is mostly interviews, surveys, prototype tests, and insight decks, AI makes it easy to imagine all of that happening somewhere else. At that point research stops being a question of method. It becomes a question of economics and organization: who does the work, who controls the tools and budgets, who decides what counts as evidence, and who gets paid to do research.

Those questions are related, and they don't have the same answers. I've written before that most research models are priced on the belief that engineering speed will plateau. This post is about the rest of the bill. I suspect AI is exposing problems research teams have had for years.

Research got cheaper to produce, but bad evidence costs as much as it ever did

Research has always cost a lot. Recruiting, interviewing, surveying, observing, transcribing, analyzing, and presenting all take time, people, and tools.

AI cuts the effort in most of those steps. It helps design studies, runs parts of data collection, processes qualitative data at scale, writes analysis code, draws charts, and makes old research easier to find.

Not every cost disappears. Recruiting the right people still matters, some populations are still hard to reach, and sensitive research still needs care. A polished interface does not come with methodological expertise built in. But the cost of producing most familiar research outputs is dropping.

When something gets cheaper, more people can afford it. Questions that never justified a formal study become worth asking, and teams that ran on assumptions can gather some evidence instead. I think that's good. I've never believed researchers should run every interview or supervise every usability test.

The complication is that producing evidence and knowing whether it can be trusted are two different costs, and AI only cut the first one.

Take a pricing study, as a hypothetical. AI can write the survey, send it, analyze the answers, run fairly advanced statistics, and build a convincing deck. None of that guarantees the sample matches the people who will pay, that the questions measure what they're meant to, that respondents understood the pricing scenario, or that the analysis supports the decision.

The report can look professional and still be a misleading basis for a decision worth millions.

Qualitative research has the same problem. AI can run hundreds of interviews and find recurring themes. If the participants came from the wrong population, the questions carried hidden assumptions, the analysis flattened meaningful differences, or the research question was badly framed, more interviews don't fix any of it. AI writes good questions now, and that was never the problem.

This doesn't mean AI can't help with method. It can. A novice working with AI sometimes produces better work than an experienced researcher running on habit. The point is that producing a research artifact and establishing that its conclusions hold are separate jobs.

That matters because research failures are hard to see. A broken feature shows symptoms. Bad research produces a plausible recommendation that can go unchallenged for months or years, and when the decision underperforms, the organization blames execution or the market.

The person reading the research often can't judge its quality. So the organization sees the cost of research clearly and the value of avoiding bad evidence barely at all. AI made research output cheaper. It didn't make the consequences of bad evidence cheaper, and organizations will be tempted to optimize for the first while underestimating the second.

Research tool companies are following a market much bigger than researchers

There's another economic force here, and I think the profession has paid too little attention to it. Research tool companies are changing who they sell to.

Many research platforms built their products and brands around professional researchers, who were their main users, advocates, and sometimes buyers. But researchers are a small share of the people who make product and business decisions. Product managers, designers, marketers, engineers, customer success teams, founders, and executives all need customer evidence, and many of them have budgets and tools of their own.

To a vendor, that's a much bigger market, and the product pages show it. Maze's AI moderator page says it's for everyone "from junior and senior researchers to designers, PMs, and beyond," including PMs who "need to run research but lack the confidence or bandwidth." Dovetail's page for product managers tells them to "build roadmaps on customer evidence, not opinion." Outset's page for cross-functional teams says it helps "product, marketing, and strategy teams get high-quality customer insight fast," without "a dedicated moderator or analyst."

These companies still serve researchers, and some of them stress method and guardrails. Selling to more people doesn't mean they've abandoned the profession. The commercial logic is simple: selling to a team of ten researchers is a smaller business than selling to hundreds of people who make product decisions.

None of that proves PM and design organizations have bigger research budgets than research teams do, or that vendors are trying to get rid of researchers. It shows that a product that depends less on specialists can sell to more people. There's nothing wrong with that. Tools shouldn't need years of training to operate.

But the interests aren't aligned. A vendor grows when more people use its product. A research team might grow too, if its expertise reaches further, or it might watch its work get spread across the organization. The vendor can grow while research headcount stays flat, and it can grow precisely because the company decides it can do more research without hiring more researchers.

That isn't a conspiracy. It's what happens when you grow a market by lowering the barrier to entry. It still shifts power. More research capability is bought and used by people outside research teams, and they bring their own expectations about speed, cost, and how much specialist involvement is needed. Over time, those expectations shape what the organization thinks research is, and that definition matters when budgets get decided.

More research does not mean more researchers

Product managers and designers were running interviews and usability tests long before generative AI. What's changing is how much they can do, how much help they get, and how fast they can produce something that looks like a finished study.

The surveys already show it. User Interviews surveyed 485 researchers for its 2025 State of Research Strategy report, and 71% said their organization has people who do research without being dedicated researchers. Great Question surveyed 301 UX researchers and research operations professionals in 34 countries for its 2025 democratization report, and 84% work at organizations that let non-researchers run studies, at least in a limited way.

Both are industry samples, not a census, and neither shows that AI caused the change. They do show how far research has already spread in the organizations they reached. I expect AI to spread it further.

The question is what that does to the demand for researchers. There are at least two plausible outcomes. In one, cheaper research creates more demand: teams ask more questions, find harder problems, and need specialists to handle them, so researchers spend less time on routine studies and more on complex ones. In the other, companies use AI and self-serve platforms to move work away from researchers, and research activity goes up while the number of researchers stays flat or drops.

Both are plausible, and both can happen in the same company at once. This is the split I wrote about as the K-shaped divide in research work.

So we should drop the comforting idea that more research means more research jobs. A company could go from fifty studies a year to five hundred without growing its research team. It could decide that three researchers overseeing a distributed program can replace ten researchers running studies themselves. Whether the research gets better or worse is a separate question, which is exactly the problem.

Demand for research and demand for researchers are different things. I've written before that demand for research is rising while research jobs are not. The profession has often treated the two as the same: if companies value research more, research teams must become more valuable. That only follows if the company believes the research team gives it something it can't get as well elsewhere. Which brings us to a problem that is older than AI.

Most research teams grew around whoever was hired first

Here's a thesis I think deserves more scrutiny: many research teams were never designed around what the business needed to know. They accumulated.

A company needed usability tests, so it hired someone who could run them. That person came from design, psychology, anthropology, marketing, computer science, or somewhere else, with or without formal research training. The company grew, the researcher became senior, then perhaps a manager or a director. Along the way they hired more researchers, set up processes, wrote career ladders, and shaped what the company thinks research is.

None of that required anyone to ask what the business needed from a full research capability. The function grew around the skills and habits of the people already there.

I'm not claiming every team grew this way, and I don't have systematic evidence that it's the dominant pattern. Some research organizations were built deliberately and have wide methodological range. But I think the pattern is common enough to deserve a hard look.

The risk it creates is specific. A team's methods end up reflecting its hiring history instead of the evidence the business needs.

Take a team built mostly around interviews, usability tests, and fairly simple qualitative synthesis. Those are legitimate methods, and done well they take expertise and produce a lot of value. Then the company starts asking questions about measurement, experiment design, causal inference, behavioral modeling, pricing, segmentation, or large-scale quantitative analysis. Does the team learn those methods? Does it work with specialists elsewhere? Or does it keep answering new questions with the methods it already knows?

The reverse happens too. A team made up mostly of quantitative researchers can lack the interpretive depth that ethnography, long-term fieldwork, participatory methods, or complex social practices require.

This isn't an argument that quantitative research is more sophisticated than qualitative research. That misunderstands both. Some qualitative methods take extraordinary methodological and theoretical depth, and some quantitative work is little more than descriptive statistics on a badly designed survey. The distinction that matters is breadth against narrowness, and people who know why a method fits against people who are simply used to it.

It isn't about credentials either. UX research was never one profession, and people come into it by many routes. Some build deep method knowledge through years of practice, mentorship, and study. Others spend years doing research without growing much past what they learned early on. Ten years of doing research is not the same as ten years of developing research expertise, and the difference matters a lot once those people start building and leading teams.

A research manager's experience becomes the team's strategy

People recognize expertise they understand. A manager whose experience sits in one method tradition can be very good at judging candidates from that tradition, and much worse at telling deep expertise from surface familiarity in methods they've never practiced.

A manager who knows little about quantitative research, for example, might judge candidates by how familiar they are with survey tools, dashboards, or statistical vocabulary. Running a survey is not the same as understanding sampling, measurement validity, inference, or the assumptions behind a model.

A candidate with much deeper expertise can be hard for that manager to evaluate, and hard to manage too, because the manager can't review the work or guide the person's growth. The same thing happens when quantitative leaders judge strong qualitative researchers. It's part of why UXR interviews often end up as auditions for a bar nobody wrote down.

Over time it reproduces itself. Leaders hire the skills they can recognize, those hires become the next senior researchers and managers, and their combined experience defines what research is supposed to do. The team becomes very efficient at a narrow set of work and poorly equipped for the rest, and because its practices are familiar and well regarded internally, nobody pushes it to change.

No manager needs to be an expert in every method. That person would be hard to find. But research leadership does need enough method literacy to see the edges of its own expertise, judge unfamiliar skills sensibly, and bring in specialists when needed. A leader doesn't have to build every model or run every ethnography. They do have to know what those methods contribute, when they fit, and how to tell credible expertise from a convincing presentation.

This matters most when the company expects research leaders to set standards for everyone else. If researchers are supposed to set methodological guardrails across the organization, someone has to ask whether the people setting them are qualified to. A senior research title doesn't make someone a methodological authority.

Then AI arrives. A team whose identity rests on repeatable research activities comes under pressure as those activities get easier to do elsewhere. If its leaders haven't broadened its skills, admitted its gaps, or given it a strategic role, adapting gets much harder. AI didn't create those weaknesses. It made them matter more.

A capable research team can still look replaceable

There's another part of this that gets underestimated. A team can have deep methodological expertise and still be organizationally exposed, because companies allocate money based on what they believe a function contributes, not on everything it can do.

Imagine a team with strong qualitative and quantitative researchers who can design complex studies, evaluate evidence, build measurement frameworks, and work on hard product questions. But the rest of the company mostly sees them through interview findings, usability reports, presentations, and the occasional survey. Over time, stakeholders come to think of the team as those deliverables.

Now AI produces similar deliverables through self-serve platforms. A product executive watches a PM generate interview findings in a day and starts asking why the research team needs several weeks for a project.

The comparison may be methodologically absurd. The two studies may answer different questions, with different samples, for decisions with very different risks. But the executive can't see those differences. And if research leaders spent years presenting their teams as study providers, they built the frame the comparison is made in.

That's why positioning matters as much as capability. Research leaders don't only manage people and deliver studies. They shape what the company believes research is for and how it judges research's contribution. If leadership defines research by its outputs, it's easy to compare researchers with systems that make similar outputs cheaper. A readout is an event, and knowledge is a state. If leadership defines research as responsible for evidence quality, uncertainty, measurement, and high-stakes decisions, the comparison gets harder to make. That is the case for running research as an intelligence function.

Positioning can't be branding, though. Calling a team strategic doesn't make it strategic, calling researchers trusted advisors doesn't give them method expertise, and claiming ownership of evidence quality doesn't mean the team can evaluate all the evidence the company uses. Positioning has to rest on skills the team has and responsibilities it carries, or it's a more sophisticated way of defending the current org chart.

So leaders have two jobs at once. They have to improve what their teams can do and change how the company understands and values it. Neither works without the other.

Democratizing research well can make the research team look unnecessary

One common answer to all of this is that research teams should lead the democratization of research instead of resisting it. I largely agree. Researchers should set up the tools, training, standards, workflows, and reviews that let other people run research well.

Not every product decision needs a study, and not every usability test needs a specialist watching it. Making researchers the gatekeepers of every customer conversation would be an excellent way to make the whole company resent the research team.

But democratization has its own paradox. Imagine a research leader who builds an excellent self-serve program. PMs run basic discovery interviews, designers test early concepts, and teams have approved recruiting, tested templates, training, and a shared place for findings. Researchers consult when asked and lead the complex studies.

It works. Research happens more often, teams learn faster, and the backlog shrinks. Then an executive asks why the research team still needs the same headcount when so much research now happens without it.

A research function can increase its impact while reducing the company's sense that it needs researchers. The work that made its contribution visible is now happening elsewhere, and if leadership can't explain what the team still does, its own success weakens it.

That's why guardrails alone aren't the answer. Guardrails matter, but they can be automated, turned into a commodity, or seen as bureaucracy. A team that answers by requiring approval for every study becomes an obstacle to the democratization it built. A team that tries to keep all execution for itself to protect headcount is defending territory, not improving decisions.

The more durable approach separates access to research from accountability for the evidence behind important decisions. A designer exploring reactions to an early concept should move fast. A PM doing early discovery should have good tools and guidance. A company making a major pricing decision, running a high-stakes experiment, studying an unfamiliar population, or drawing causal conclusions from observational data should have the right methodological expertise involved.

The line isn't between researchers and everyone else. It's between levels of methodological complexity, uncertainty, and decision risk, the same reasoning behind minimum viable rigor: the threshold below which evidence can't support the decision it informs. Researchers should have a defined role in designing the system that draws that line, and in writing the standard for what counts as evidence.

Research leaders have to design the function around the decisions it serves

If we built a research organization from scratch today, would we build it the way most teams are built now? Would we hire the same mix of people, favor the same methods, write the same career ladders, and keep the same relationships with Product, Design, Data Science, and Engineering? Would we still organize around requests for studies, or around the decisions the business has to make?

I suspect many research leaders would answer differently if they were starting from zero, and that's where the profession should look much harder.

We've spent years telling individual researchers to be more strategic, learn quantitative methods, build technical skills, adopt AI tools, and grow their influence. Much less attention has gone to whether research leaders are building teams that can use those skills. I made a version of this argument in UXR Team, Not UXR People.

A researcher can build deep method expertise and still work for leadership that doesn't understand it, value it, or know how to use it. They can learn advanced quantitative methods while their manager judges them on stakeholder satisfaction and study counts. They can be ready for strategic questions while the company keeps bringing research in after the decision is made. Individual growth matters, but the design of the organization decides whether that growth turns into opportunity and influence.

A research leader's job should start with what the business needs to know and then deliberately assemble the skills to answer it. For some companies that means deep experimentation, measurement, and behavioral analysis. For others it means strong qualitative, ethnographic, longitudinal, or participatory work. Most mature product companies need a mix, plus good partnerships with neighboring teams. Those choices should follow the problems the company faces, not the backgrounds of whoever was hired first.

That doesn't mean research has to own every method. Data science, marketing science, and analytics may already have the expertise. The question is how the company makes sure the right expertise is involved in the decisions that need it, without duplication or turf fights.

It also means accepting something uncomfortable. Some research teams are overstaffed for what they're responsible for. Some roles no longer justify their old scope, and some work can be done more efficiently by AI or by other teams. Keeping every existing responsibility can't be the goal. The goal is a research function whose contribution to decision quality justifies what it costs, even if that's a different team, with different skills and a different structure. The profession should be willing to have that conversation.

The research function should own the quality of evidence and let execution spread

What would that function look like? The model I find most promising spreads execution across the company and makes method accountability explicit.

Research leadership sets the standards and infrastructure that let other teams run appropriate research on their own: recruiting, data governance, study design, documentation, interpretation, and responsible use of AI tools. The system scales with the risk and complexity of the decision. Low-risk exploratory work is mostly self-serve, more complex projects get method consultation, and high-stakes questions get specialists and closer scrutiny.

It can't be a checklist of approved methods. A simple-looking study has large consequences if the company treats its findings as settled, and a sophisticated method is overkill for a low-risk decision. The right level of oversight depends on what the research is expected to support, how uncertain it is, and what happens if the answer is wrong.

The research function also needs ways to check the quality of evidence produced across the company. Can a finding be traced to its data? Do teams know the limits of their samples? Are claims sized to the strength of their evidence? Are conflicting findings investigated or ignored? Are AI-generated analyses checked when the decision warrants it? This is what I built the Frame for: an organization's actively maintained model of its users, where every claim carries its evidence, its scope, and how sure anyone can be.

That responsibility doesn't have to sit only with a standalone research team. It can be shared with data science, research operations, privacy, legal, or anyone else relevant. What matters is that someone is explicitly accountable.

And researchers have to keep building expertise. A team can't be the company's methodological backbone by managing templates, holding office hours, and running a repository. Those support the work, and they don't replace research skill. The function has to keep the ability to run hard studies, challenge weak conclusions, evaluate unfamiliar methods, and take on problems other teams can't handle responsibly alone. Otherwise we've swapped research practitioners for research administrators, and that's not a stronger position.

Counting studies stops meaning anything once studies are cheap

If AI keeps lowering the cost of producing research, the usual productivity measures tell you less and less. Counting studies, interviews, reports, and presentations says very little about whether the company is making better decisions. A team that runs twenty studies can contribute far more than a team that runs two hundred.

One study can stop a major investment in the wrong direction. Another can show the company has misunderstood a key customer segment. A well-designed experiment can overturn an assumption that several rounds of qualitative feedback seemed to support.

Those contributions won't always turn into revenue figures, and we should be skeptical of simplistic attempts to put a dollar value on every insight. But research leaders still have to show how expertise changes what the company does. Did the work expose an important assumption? Did it change a product decision? Did it find an overlooked population? Did it resolve uncertainty that mattered? Did it stop a misleading conclusion from gaining credibility?

When research is spread across the company, leaders should also measure the systems researchers build for others. Are teams using appropriate methods? Are studies interpreted more carefully? Do teams recognize when they need a specialist? Is existing evidence reused instead of collected again? Are important decisions better informed? These are harder questions than how many interviews someone ran last quarter, and they're much closer to why a company should invest in research at all.

Even these measures won't settle the economics. A company can agree its research team is valuable and still decide a smaller team, a more distributed model, or a different mix of specialists is enough for less money. The right answer isn't always a bigger or more independent research organization. How many researchers a company needs isn't a moral question. It's a question of skills, responsibilities, costs, risks, and alternatives, and research leaders should be ready to make that case without assuming research belongs to people with research titles.

Being right about how complex research is does not keep a research team funded

Whenever technology makes a profession's work more accessible, practitioners point out that the job involves much more than the tasks being automated. They're often right. Being right about the complexity of your work doesn't guarantee your team survives.

The economic question is whether the company sees that complexity, believes it matters, and will pay for the expertise to handle it. AI is changing all three. It makes research activities more accessible, lets vendors sell research to much larger audiences, gives companies new ways to spread the work, and makes familiar research outputs easier to produce without making them any more reliable.

It also opens up opportunities. Researchers can take on questions that used to be too expensive, spend less time on repetitive work, and put more into methods, combining evidence, and hard problems. Companies can involve customers in product decisions more often, without every bit of learning waiting on a researcher's calendar.

None of that requires professional research to disappear, and none of it guarantees research stays relevant. Which brings me back to leadership.

If research teams grew around the skills of their first hires, many will need to take a hard look at what they can and can't do. If leaders spent years defining research by the studies their teams produce, they'll need to change how the company understands their contribution. If they want research to own the standards for evidence, they'll need the expertise to back that up. And if they believe researchers should stay indispensable simply because they're researchers, their companies may disagree.

I don't think the answer is defending the profession's borders harder. It's being much more deliberate about what research functions are for, how they're built, and what makes them worth keeping. That takes method depth, a range of skills, leaders who understand research beyond their own experience, and an organization that separates producing research from the reliability of what it produces. It also means admitting when AI or another team can do certain work just as well, or better.

The future of research probably isn't fewer people doing research. It's the opposite: research-like work becoming a normal part of almost every product role. That still doesn't answer what happens to professional researchers. Their future depends partly on their skills, partly on the technology, and mostly on decisions about how organizations are designed, who owns what, and who is accountable.

Research leaders aren't just managing a change in technology. They're managing a change in the economic case for their own teams. So here is the question I'd like more of them to ask: if we were building our research organization today, knowing what AI can do and what the business needs, would we build the team we have now?

If the answer is no, the problem isn't only AI. The economics changed, and the organization didn't.

🎯 If this post made you look at your org chart a little differently, subscribe. The next one will do it again, and it lands in your inbox, no algorithm required.

📖 If you’re the research leader who has to answer that last question, start with my book: AI-Powered UX Research, the operating manual for running research at the speed your team needs. It’s cheaper than a reorg, and much faster.