Artificial intelligence is reshaping nearly every corner of the research and insights world. But for Marc Desmond and Ben Osborne, the real story isn’t about what AI can replace. It’s about what it makes possible.
From AI-moderated interviews that amplify underrepresented voices, to intelligent agents that bring customer data to life for non-research stakeholders, the opportunity is vast. We sat down with Marc and Ben to explore where the practice stands today, the dangers of overconfidence in AI outputs and why nothing beats actually talking to your customers.
Where do you draw the line between what AI can do in research, and what still requires human judgment?
Ben Osborne: I’d push back on the idea that there’s a line at all. A lot of the narrative right now is about “finding the balance,” as if there’s some perfect equilibrium between humans and AI. But really, it’s not about one replacing the other; it’s additive. AI augments human thinking; human thinking keeps AI honest. The line is always moving, and hybrid approaches will remain the richest for the foreseeable future.
Marc Desmond: Everything still requires human judgment. AI isn’t a solution you can implement and walk away from; it still requires a careful human eye and the right perspective to ensure it’s reliable and used well. That said, AI is genuinely transformative across the board. It accelerates speed to insight, it’s incredibly useful for analyzing qualitative data and we’re even seeing AI-moderated interviews proving genuinely effective.
How has AI changed the way you listen to customers?
Marc Desmond: On the front end, AI-moderated interviews are becoming more common, and audiences are increasingly comfortable engaging with them. If you have a segment that’s open to it, there’s a real opportunity there. On the back end, the bigger shift is in how you activate what you’ve heard. Before, research ended with a PowerPoint report. Now, there are new ways to bring customer insights to life: agents built from the findings, AI-generated personas and chatbots that let non-research stakeholders query the data directly. It’s a fundamentally more engaging way to make the customer’s voice accessible across an organization.
Ben Osborne: Working across multiple markets and languages already means relying on partners and moderators. You’re used to some distance in the research relationship. But AI changes what’s possible when it comes to reaching voices that would otherwise be excluded. Imagine processing dozens of transcripts from international markets simultaneously or going deeper into regional nuances without the usual barriers. For me, the most exciting thing about this wave of AI is that it expands who we’re listening to, not just how efficiently we listen.
What are the risks of over-relying on AI?
Marc Desmond: The biggest risk is that your strategy falls flat because you’ve been given a confident-sounding answer that misrepresents your customer. AI is eager to please, which means it can overlook or downplay potential red flags. If you’re compounding that with data that’s already stale or using a synthetic model that hasn’t been stress-tested against reality, you may be working with a misleading result. The worst case scenario is thinking you have a perfect solution when you might not find out it was wrong until you’ve launched your brand, campaign or product.
Ben Osborne: Absolutely. I like to think about it in terms of hindsight, insight and foresight. One of the risks is that we treat hindsight as foresight; that we mistake what was true a year ago for what’s true right now or what’s coming next. Context shifts so fast. When you’re querying AI against a data set from twelve months ago, and the world has changed dramatically since then, you have to be constantly asking: Is this still right? And honestly, there’s also a contract with respondents to uphold. When someone gives their time and opinion, they trust that it will be represented accurately. That’s a promise we can’t let AI break on our behalf.
Can AI help reach underrepresented audiences, and what are the trade-offs?
Ben Osborne: It’s a genuine Catch-22. Large data sets trained on broad populations tend to smooth out the extremes. They center on the majority and underrepresent niche groups. But as a methodology for reaching those niche groups, AI can actually be very effective. Think about audiences that are hard to find or vulnerable, that need to be anonymized or are protected by privacy regulations — people with specific medical conditions, victims of crime, highly specialized professionals and very niche demographics, including marginalized communities. AI-assisted research can model and simulate those voices in ways that weren’t possible before, creating synthetic respondents that complete surveys and take part in focus groups on this groups’ behalf.
There’s something almost ironic about it: AI can create the problem and provide the solution. So, the key is intent. If you’re just using AI to find the easy audiences faster and cheaper, you’re going to exclude the hard-to-reach ones even more than before. If you’re deliberately using it to reach audiences who would otherwise be left out entirely, that’s where it gets genuinely powerful and meaningful.
What’s one thing you wish CMOs and clients better understood about AI’s role in research right now?
Marc Desmond: That AI is a tool to make the researcher’s job easier and more efficient, not a replacement for the relationship with the customer. Talking to your customers directly is still something you need to do and do regularly. What AI gives you is a way to activate that voice more broadly: to take what your customers are saying and make it accessible, alive and engaging for everyone in your organization, not just the research team. The ability to turn a static report into something people can actually interact with and query? That’s genuinely transformative.
Ben Osborne: Don’t make the first thing you want to replace the human you’re actually trying to understand. For most companies, especially B2B clients, you’re not sitting on a massive, constantly updated stream of customer data. For those organizations, synthetic respondents and digital twins are a significant risk, not a shortcut. The best thing is still talking to your customers. And there’s something worth preserving about that — the idea that research is fundamentally a relationship, not just a data exercise.
