Skip to content
Three cropped photographs on a dark brown ground: a person’s eyes seen through a glass prism, blurred pink and green petals, and white snowdrops

AI can, and should, scale human insight

At Made by Many, we’ve always argued for the importance of human research. For most of our history, that meant in-person, qualitative in-depth interviews, and it took weeks. We recruited and screened participants, ran the sessions and transcribed them. Then we copied hundreds of quotes onto post-it notes, grouped them into themes and wrote insights. It was slow and sometimes arduous, but it was worth it, because the time we spent with people shaped everything we made afterwards.

Over the last year, that process has changed more than it did in the previous ten. AI-moderated interviews, synthetic users and research agents are here, and some argue research can be replaced by AI altogether. We don’t agree. But we do think teams that ignore these tools are missing a real opportunity to get breadth and depth at the same time, and will struggle to keep up with the pace product teams now work at.

Here’s how our approach has changed across preparing, running and making sense of research, and where we think human judgement still needs to be applied.

AI allows us to begin projects from a richer starting point

We do, and have always, resourced small, dedicated teams on projects over time, it ties into our forward deployed ways of working that we’ve written about in proximity does the work a brief can’t and going all-in on the forward deployed designer. By being deeply embedded in an organisation, we get close enough to its challenges and stakeholders to spot the opportunities which have the best chance of succeeding. We also work across a vast number of industries, and have done over the years, allowing us to cross-pollinate learnings from seemingly disparate spaces, from EV charging to electronic music, and come up with more creative solutions.

Yet no matter how many processes you put in place, ensuring that rich tapestry of context is continually spread across the business is hard. AI helps to make that dream a reality, stitching together every project brief, research plan and discussion guide we’ve ever written into a collective company ‘brain’. This means when we plan a new project, we can see what methods we used for similar problems and audiences. We start from what we’ve already learned, helping to sharpen our research practice over time.

Research starts with what you’re trying to achieve

But while AI provides a richer starting point, each project still needs its own research strategy. Creating a new proposition and improving an existing product are entirely different problems. With an existing product, there’s usually plenty of data to draw on, the problems are better understood and the questions can be quite specific. When you’re creating something new or finding new opportunities for value creation for organisations, nobody knows the right questions yet, and you need to follow tangents to find them.

That difference shapes every planning decision. It covers what we want to find out, what kind of research to run, who we need to speak to, how many people and in what mix, and what stimulus we need to make. It also decides how much of the work agents can take on — for a new proposition, human insight and imagination is essential.

That’s the reason why a brief which could be broadly understood as “future-proof our legacy business by developing an entirely new offering”, required a scope of imagination that would have been naturally narrowed by AI. AI can be incredibly helpful looking into the past, but speculating into the future is nuanced work, requiring a deep understanding of human nature and emergent social signals.

On the other hand, a more recent project asked us to validate a selection of potential propositions. The task was more clearly defined which meant we could use AI as an effective sparring partner to refine the research questions and recruitment strategy, and to quickly make and iterate working versions of the product directions we were testing. Unlike the first example, you don’t risk narrowing your own imagination, or that of your users.

In summary, the planning stage must remain deeply human. If you don’t spend the effort upfront shaping the right approach for the specific problem in front of you, everything that follows is at risk.

Empathy still comes from speaking to people

In-depth interviews are still where our understanding starts. Designers, strategists and technologists talk to people about their lives, their work and the things we’re making. We think running sessions ourselves is valuable for three reasons:

  1. It builds empathy. The team hears people describe a problem in their own words, and can hear the frustration or excitement in their voice. Great products are built on an understanding of people’s emotions and motivations, not statements decoupled from the sentiment behind them.
  2. A person can dig deeper. When someone says something surprising, you can go off-script and follow it. Some of the most useful moments in our research have come from doing that.
  3. You observe everything, not just what AI decided was important. Hesitation, or someone not quite convinced by their own answer, doesn’t survive into an automated summary and can be entirely misleading further down the line.
Hesitation, or someone not quite convinced by their own answer, doesn’t survive into an automated summary and can be entirely misleading.

The “yeah, maybe I could see myself using it” problem

AI can read responses very literally, building on the final point, someone who says “yeah, maybe I could see myself using it” to an interviewer is usually being polite, and has no intention of using your product. Their tone, their hesitation and their choice of words make it a ‘maybe’ at best. An AI summary will often record it as a yes, which looks great in a client summary, but it isn’t true. This problem, for now, can only be solved through human intuition and the deeply embedded ability to discern body language.

AI-moderated research gives you more data, but it can’t yet pick up nuance: the language that gives away uncertainty, or the sense that someone isn’t really sold on an idea. And it can’t give a team the shared understanding that comes from speaking to people together and helps to align on the direction you’re taking the product in.

Crunching data isn’t the same as writing actionable insights

While AI is good at compiling data and spotting patterns, it struggles with meaningful synthesis. Agents rank points by how often they come up, not by how much they matter to people. In one project, AI-generated insights kept surfacing onboarding, which made it look like one of the most important areas to focus on. But, in reality, every participant had been through the onboarding flow, so naturally everyone mentioned it. In short, it came up consistently without being important.

While AI can quickly summarise the ‘what’, it rarely defines the ‘so what’. We’ve been thinking a lot about that skill, one we deem invaluable and uniquely human, but that can be quite intangible and hard to define. The exercise of thinking more deeply and making connections is crucial, and it avoids mistaking the fluency LLMs are known for, for real substance.

Superficial summaries might work for those external to the project, to get a gist of what was concluded, but as someone in the project, you have to be just that, in it. Research has shown that we can’t comprehend what’s merely been summarised for us, that work to build conclusions, is where the deep level of comprehension comes from. And that understanding is the point of departure from which you define the next steps and allows you to have conviction in the direction you’ve chosen to take as a team.

A final note is to say being in the weeds matters not just for the project you’re on, but for everything that comes after. Whilst discernment might be an intrinsically human capability, formed by decades of experience being in and around the world, it can become blunt and off the mark if you stop participating in that world. In other words, the nuances and social signals you absorb over time is how that discernment was built in the first place, continuing to be in environments where you continue to exercise those skills is how you keep it sharp.

We can now validate at a scale and pace we couldn’t reach before

While we still run in-depth interviews ourselves, AI has completely transformed what happens next.

Using AI-augmented platforms, we can take what we heard in a handful of interviews and verify it against a much wider data set. AI-enabled social listening tools allow us to check if a signal holds across thousands of threads, while AI-moderated interviews help us test ideas with many more people, across markets and time zones. The AI asks the questions, follows up on answers and runs sessions around the clock and in other languages. Often they run over the weekend and the results are waiting for us on Monday.

Most importantly, we use this to validate what we’ve learnt, not to generate new insight. Depth interviews give us the insight, and AI-moderated research shows how far it holds for a larger group. Clients have often been nervous that we only ran interviews with eight to ten people. While we still believe smaller groups give better insight — surfacing the nuances which confirm a conviction or disprove an earlier assumption — we are excited about the ability to validate that thinking at scale and give our clients more confidence about what to build and why.

What this means for clients

This is all to say that agents have changed the mechanics of research. Planning starts from everything we’ve done before, recruitment takes days, interviews run overnight and in other languages, and a first read of the data takes minutes. They haven’t changed where the value comes from, which is the time a team spends with real people and the judgement it takes to decide what those people meant.

Clients understandably ask why research takes a week if AI can run interviews overnight. Some of that challenge is fair — faster recruitment, transcription, AI moderation and AI analysis have cut our research timelines dramatically on recent projects. But speaking to people and working out what their conversations mean hasn’t shrunk in the same way, and we don’t think it should. That’s where the time agents give back should go.