The Translation Layer
Emily Peters, Director of Research sits down with Matt Connolly, Founder & CEO at Sønr – on why AI can answer your question but can’t tell you if you’re asking the right one.
Everyone in insurance is being told AI can do research now. And in a narrow sense, it can. Point it at a market, ask it what’s happening, and it will give you something back: quickly, confidently, comprehensively. Or at least it will look that way.
The problem is more subtle than most people realise. And it’s why, a decade into building a research practice that sits at the intersection of human expertise and emerging technology, Emily Peters is not particularly worried about being replaced by a large language model.
“AI has literally been built to please us,” she says. “It’s trying to help you. And even if it doesn’t necessarily have the right answer, it still wants to give you an answer, because that’s what it’s been built to do.”
That’s not a flaw you can patch. It’s structural. And it’s the starting point for understanding what research actually is, and why the human layer still matters.
A decade at the edge of the market
Emily Peters joined Sønr in 2017, when insurtech was still finding its shape. She has watched the market through multiple cycles since: the hype waves, the corrections, the slow integration of technology into an industry not exactly famous for moving fast. As Director of Research, she leads the team that works directly with carriers, reinsurers and brokers across more than 60 countries to help them understand what’s happening, why it matters, and what they should do about it.
Her own background is, by her admission, an unusual one for this space. Ancient history and Egyptology. The complete other end of a timeline to future technology.
It turns out to be exactly the right training. “I think it’s more about skill set than experience,” she says. “You need to have that interest and curiosity and ability to ask questions. You need to love learning. You need to be a bit of a detective. There’s not always a clear-cut answer, especially when you’re working in emerging risks and emerging trends. And you need to be comfortable challenging people. Whether they’re asking the right question. Whether they’re framing an assumption as a fact.”
That last one is where the AI comparison gets interesting.
What AI can’t do
The standard critique of AI in research is about accuracy: hallucinations, outdated training data, gaps in coverage. These are real. But Emily’s critique goes deeper.
“There’s sometimes a bit of a danger with AI of confusing breadth with depth,” she says. “You can be served up a really interesting answer. But there’s always another layer to go: that contextualising, that applying personally to a specific situation with a client.”
She gives an example. An AI could surface the best startup on paper: strong funding, the right investors, impressive technology. But the best startup on paper isn’t necessarily the best startup for you. Your business has specific integration needs, specific customer questions, specific constraints the model has no way of knowing. The gap between the answer and the right answer is where the research work actually lives.
“We’re there to translate,” she says. “It gives us the language. We’re there to turn that into actual reading.”
But the deeper limitation is the one that clients don’t always see coming. An AI will answer the question you give it, even if that question is wrong. It won’t stop you. It won’t push back. It won’t ask whether you’re actually making an assumption, or whether the real question is something else entirely.
That is what a good researcher does, before any of the actual research begins.
The report is a monologue
A report, however good, is a one-way document. It’s a monologue of findings. You can brief the work carefully upfront, but the output is static. It answers the question as it was understood at the time it was written, and it starts going out of date from the moment it’s published.
What Sønr’s research team does after the report is, in many ways, the more valuable part. Every engagement ends with a conversation: a walkthrough of findings, questions from both sides, a discussion of what it means specifically for this business, this team, this decision. Sometimes half an hour. Sometimes considerably longer.
The value isn’t just in what the researchers know about the research. It’s in what they know about the market as a whole, from having run dozens of similar projects, for comparable organisations, across overlapping themes. That accumulated context doesn’t exist anywhere in a document. It surfaces in conversation, in response to a question, in the moment someone says “actually, we’ve been thinking about this slightly differently.”
“Everyone takes us up on it,” Emily says. “I don’t know a client that doesn’t.”
It can be difficult to package this and call it a product. But that breadth of exposure and experience, brought live into the room, is what makes the difference between understanding a trend and knowing what to do with it.
The hybrid model
This is where Sønr 2.0 comes in. And the relationship between the platform and the research team is less about replacement than it is about extension.
The platform handles the always-on layer. Startup watchlists, market monitoring, pre-built intelligence workflows: the kind of continuous coverage that no research team could sustain at that cadence. It’s built on a decade of proprietary data, calibrated specifically for insurance, and it doesn’t sleep.
The research team takes it from there. “You’ve been served up this market intelligence,” Emily says. “What’s next for you and your business? What’s next specifically for you and your team, and the problems or the opportunities you’re facing?”
There’s also a more practical integration. When the research team produces a report for a client, that report can live inside the client’s Sønr 2.0 environment, and the platform continues to surface relevant updates on top of it. The research becomes a living document, not a snapshot. The monthly or quarterly check-in still happens, but when it does, the ground has already been prepared. You get straight to the conversation that matters.
“It means the research becomes this almost living, amorphous thing that changes as the market changes,” Emily says. “And we’re still there. It’s not just an automated thing.”
The platform also changes the work itself. It frees the team from the parts of research that are essentially mechanical: gathering sources, tracking activity across the market, which means more time for the parts that are genuinely difficult. More face time with clients, as Emily puts it. Which is always the part that matters most.
What’s on the desk right now
Emily’s team is seeing three big themes run consistently across the client base. They are not surprising, but the depth of how they’re playing out is.
AI, obviously. But not as an abstract concept: as an operational pressure. “There’s a very real and rational fear of how AI is going to change and redesign teams, but also specifically job roles,” Emily says. Research that was current six months ago is already out of date. The pace of change in a sector famous for moving slowly no longer feels slow.
Climate is the other constant. Physical risks, resilience, the transition to renewables, battery storage. And alongside it, a harder question about how to actually engage policyholders, particularly residential customers who haven’t experienced a loss event yet but are increasingly exposed.
Cyber runs through both. Fraud and claims are being reshaped by AI-enabled deepfakes and synthetic identity. And further out, quantum computing: quieter right now, but not gone. Q Day, when quantum capability becomes a genuine threat to current encryption, is a topic Emily’s team is watching carefully even as the immediate AI noise drowns it out.
The glorified library problem
There’s a version of market intelligence that most organisations have built, without necessarily calling it that.
Content from events. Newsletters. Industry reports. Ad hoc publications. None of it commissioned specifically for the business. None of it pertaining precisely to their needs, their problems, their opportunities. It’s generic, often recycled, sometimes second-hand. Better than nothing, but not much of a foundation for the decisions it’s meant to inform.
“You don’t know what you don’t know” is the thing clients tell Emily most often when they explain why they came to Sønr. Generic intelligence doesn’t close that gap. It just makes you feel like it might be closed.
The compound effect is significant. Businesses invest millions in solutions to problems they’ve diagnosed from intelligence that was, in important ways, incomplete. The cost of the nearly right choice, multiplied over years and major decisions, runs into the tens of millions.
Emily’s advice on building a market intelligence function comes back to a single question that most teams forget to ask: what is the end goal? What decision are you actually trying to make? Without that anchor, you can build something comprehensive, expensive and thoroughly impressive. “You’ve essentially just built a glorified library.”
The human sandwich
Emily’s framing of where humans fit in the AI-augmented research workflow is one worth holding onto.
“Humans are going upstream and downstream,” she says. “Upstream is figuring out what the right questions are: what is the ultimate decision you’re trying to come to. That’s so important, because if you are one degree off, you’re going to go off on a really big tangent that you didn’t need to go on. And then downstream, once the AI has surfaced the data and intelligence that it has found, being able to analyse that critically, take meaning from it, apply it to your specific situation, pass judgment on it.”
In the middle: the model, doing what models do, at speed and scale.
The danger she flags is one worth taking seriously. AI makes things accessible. It simplifies.
And in doing so, it creates conditions in which critical thinking, the upstream and downstream work, gets quietly eroded. “It’s never been more important,” she says. “And it’s never been more at risk.”
Sønr works with the world’s leading insurers, reinsurers, and brokers to deliver primary market intelligence, always-on platform intelligence through Sønr 2.0, and cohort-based learning through the Emerging Trends Academy. To find out more, visit sonr.global.
