Why human judgement still matters
Editor’s note: Justin Sutton is co-founder of Catapult Insights, with more than 20 years of experience in qualitative research and consumer behavior. His work focuses on innovation, customer decision-making and the application of behavioral science to business strategy. Find Sutton on LinkedIn.
For years, research professionals have talked about earning a seat at the table. Today, with generative AI moving quickly into research workflows, many are wondering what it will take to keep one.
That anxiety is understandable, but it can also point us toward the wrong problem. The most important question facing the insights industry is not whether AI will replace human judgment. It is whether organizations will still recognize why human judgment is needed.
Most of the conversation about AI in research has focused on efficiency, automation and speed. We hear about faster transcript review, faster summaries, faster synthesis, faster reporting and, in some cases, faster decisions. Anyone who has spent weeks buried in transcripts or trying to turn a messy body of qualitative material into something useful can understand the appeal of tools that reduce friction and make certain parts of the process easier.
But speed is not the same thing as understanding, and automation is not the same thing as judgment. In qualitative research especially, the hardest work has never been simply organizing what people said. It has been making sense of what those words mean in context, with appropriate humility about what the research can and cannot support. That work has always required restraint, interpretation and a willingness to live with ambiguity longer than most organizations naturally want to.
The real risk of AI in insights
My concern is that AI can smooth over hesitation, compress context, resolve contradiction too quickly and turn partial evidence into language that feels settled. A finding that once would have been surrounded by caveats can now be turned into a confident paragraph in seconds. A handful of moments from several interviews can become “the consumer wants” or “the key insight is” with a level of polish that obscures the thinness of the evidence underneath.
That is seductive because organizations are already hungry for clarity, direction and confidence. We are increasingly surrounded by tools that are remarkably good at producing language that feels like an answer, but qualitative research has always required us to remember that a well-written answer is not necessarily a well-supported one.
Good qualitative work is complicated because people are complicated. People describe one set of motivations while behaving according to another. Many naturally struggle to explain habits that feel automatic, emotional or socially shaped. Sometimes the most articulate people are not the most representative ones, but simply the best storytellers. Sometimes the most important meaning is not found in the cleanest quote, but in the tension between what someone says, what they avoid saying, and what their behavior suggests.
That is where craft enters the work. Qualitative research is more than the extraction of themes from language. It is the disciplined interpretation of human experience. It requires knowing when a pattern is meaningful, when it is merely interesting, when it is worth exploring further, and when it is being asked to carry more weight than it can bear. AI can assist with pieces of that process, but it cannot take responsibility for the judgment behind it. That responsibility still belongs to us.
This is why I believe the role of the research professional is at risk of losing its way. We have often described researchers as the voice of the consumer inside an organization, and that remains an important part of the job. But in an AI-accelerated world, insights leaders also need to remain the voice of methodological truth. We have to represent not only what consumers are telling us, but also what the research design allows us to responsibly claim.
Realities put researchers in a tough position. Stakeholders often come to research with more questions than any one study can answer, and that pressure has only intensified as budgets shrink, timelines compress and teams are asked to do more with less. The impulse behind those requests is usually reasonable. People are trying to make the most of the work while avoiding additional cost, reducing uncertainty and keeping decisions moving. At this crossroad, researchers must consider how to be useful in the moment while also protecting the usefulness of the research function over time.
We must prevent stretching our work so far that it makes the findings unreliable. Those are not easy conversations, especially in organizations that reward speed and decisiveness. It’s important to remember that these are acts of stewardship, not obstruction.
Always having an answer doesn’t build trust. Trust comes when stakeholders learn that the answers we do give are grounded, carefully interpreted and worth acting on.
The thing about trust is that it rarely disappears all at once. More often, it erodes gradually, as caveats get softened because a presentation needed to be cleaner, as the line between what we know and what we suspect blurs, and as recommendations become more definitive than the evidence supports. That rarely happens because researchers are careless. Instead, it happens because they’re trying to be helpful, and helpfulness without rigor eventually becomes a liability.
That dynamic isn’t new, but AI makes it easier to miss. When outputs are faster and cleaner, the distance between evidence and narrative shrinks. A polished summary can create the impression that the thinking is complete, even when the interpretation underneath still needs scrutiny. Leaders may not always be able to name the problem, but they can sense when confidence has outpaced evidence. Once that happens repeatedly, research loses influence not because it’s too slow or too cautious, but because it no longer feels dependable.
It’s one thing to know researchers should push back. It’s another to do it when your team has been cut, your budget is being questioned and everyone around you is being asked to move faster – especially in organizations that don’t fully distinguish between method, moderation, analysis and strategic interpretation, and are being sold AI as a way to get more done for less.
Under that pressure, the temptation to accommodate is powerful. Naming limitations can feel like admitting weakness rather than practicing professional discipline, and it’s tempting to become the partner who always finds a way to answer every question. But when researchers avoid difficult conversations to seem more helpful, they accelerate the very loss of credibility they’re trying to prevent. Once an organization starts seeing research as an answer factory rather than a discipline that protects decision quality, it becomes easy to commoditize – by cheaper suppliers, internal DIY tools, or AI-generated analysis that’s convincing enough to pass.
The more valuable path is harder in the short term. It means being clear about tradeoffs, protecting the integrity of core objectives, and resisting the pressure to turn every exploratory signal into strategic certainty. It also means a different kind of stakeholder management where we aren’t just telling business partners what research can’t do but instead explaining why those boundaries matter to the decision in front of them.
That distinction is important, because methodological truth should not sound like academic defensiveness and be expected to land the point. Instead, it should connect directly to business risk. If a study is being stretched beyond its design, the issue is not just purity of method; it is the possibility that the organization will make a confident decision based on evidence that cannot carry the weight being placed on it. Research leaders have to translate methodological limits into decision-making consequences, because that is where stakeholders can most clearly understand the value of restraint.
Human judgement must keep a seat at the table
The future of the profession, then, is about becoming more valuable in the places where output alone is insufficient. Research professionals will need to become stronger editors of certainty, better translators between evidence and strategy and more explicit stewards of what is known, what is inferred and what remains unresolved.
AI will continue to change research workflows, and researchers should not pretend otherwise. There will be valuable uses for it across planning, analysis, synthesis and communication. The right questions are where AI belongs, how it should be governed and what human responsibilities become more important because of it.
If insights leaders want to keep their seat at the table, we must make clear that the seat was never earned by speed or volume of output alone. It has always been earned by helping organizations make better decisions, especially when the evidence is complicated, the pressure is high and the easiest answer is not the most responsible one.
With the pace of the world accelerating around us, human judgment is as important as it’s ever been.