Source: The Future of Insurance Podcast S8E25 with Chris Tunnicliffe – Transcript
TLDR:
- The real blocker to AI in insurance is people and fear, not the technology, and the savviest carriers guarantee jobs for two to three years while framing AI as speed and capacity.
- “Digital lipstick,” meaning AI layered on top of legacy systems, creates the illusion of progress; the value comes from going below the surface and placing humans in the loop deliberately.
- Vibe coding could finally bring real tooling to small specialty and E&S books that never justified a big system.
Chris Tunnicliffe has spent decades on both sides of the insurance table, running IT for carriers and building products for the market. In this closing episode of season eight, he and host Bryan worked through where AI is genuinely changing carrier operations and where it is mostly noise. His most useful ideas are reframes that cut against the common script.
The data and experience problems are old, the tools are new
The hunger for better data is not new. What changed is that AI can now attack it differently. Tunnicliffe describes carriers that modeled claims data over eight or nine years, built a taxonomy, and can now tell an insurer which three or four indicators predict a complex claim so it routes to the right place faster. Others use AI to leave legacy data where it sits and pull it together on demand, turning analysis that took weeks into something that takes minutes. On experience, he draws a persona line: with a minor motor claim he just wants to submit photos and be served, but when his 84-year-old father has an accident, the man wants to hear compassion in a human voice.
Human in the loop, placed on purpose
Tunnicliffe’s sharpest operational point is about where the human sits. Do not park the reviewer at the very end as a rubber stamp. Instead, compare outputs: if the human reaches decision A and the model reaches decision B, route it to a person. If both agree, ask why you would add a human at all. He is clear that fully autonomous claims are not here yet, so keep sanity checks running until the data earns more autonomy. He also pushes back on the perfectionism that stalls projects. Imperfect data already exists, because Janice and Jim enter messy data after a sleepless night with a crying baby. He recalls a peer running a loss ratio around 120 to 130 who refused a predictive model because it “might be wrong,” while his current approach was clearly losing money for years. Anchoring in the devil you know is still a choice with a cost.
The people play, and tooling for the small books
His boldest move is reframing AI away from cost-cutting. Some leaders squeeze every department to force AI in. The savvy ones instead tell staff their jobs are secured for two to three years, position AI as faster speed to market and easier data processing, and free up time for people to be innovative, what he calls time around the water cooler. That cultural space, not just offloading drudgery, is what keeps a company transforming. He also sees vibe coding, meaning describing a system in plain language and having AI build a working version, as a way for tiny specialty or E&S teams to create their own apps and workflows without a large tech team, finally breaking the scale barrier that kept good tools away from niche lines.
The Bottom Line
AI is not a silver bullet, it is a shovel, so the value comes from the work you put in and the culture you set. Stop layering digital lipstick on legacy systems, place humans in the loop where outputs disagree rather than at the end, and tell your people the truth about job security so fear does not quietly stall the whole effort.