On 30 June 2026 we ran a webinar, AI in Insurance Will Your Claims Operation Break, Stall, or Scale as Volume & Complexity Rise?.
We received many more questions than we could answer live, so our expert speakers have addressed them here. For more information please contact us or request a demo.
Q: How do you build a business case when the savings are spread across leakage, LAE, and cycle time, not one line item?
A: Pick the one problem you’re really solving – customer experience, efficiency, or leakage – and make that your headline number, but keep an eye on the others too. Push too hard on one lever, like headcount, and another, like indemnity accuracy, can quietly get worse. A 1% gain in indemnity accuracy usually moves the P&L more than an equivalent cut in LAE.
Q: Given NAIC requirements for human oversight, what needs to be in our audit trail?
A: In the US, the NAIC’s position is that human oversight has to remain part of the decision, and that’s showing up in state and federal frameworks now. In practice: you need to show not just what was decided, but how, and by whom. That means a full reasoning trail for each decision, override logging, and clear escalation rules.
Q: How do you decide where AI decides, where it just recommends, and where a person has to stay in the loop?
A: It comes down to claim type and risk. High-volume, standard claims can run fully automated once you trust the outcome. Anything higher-value or more complex still goes to an adjuster, whatever the AI’s confidence level. Human review has to be a genuine check, augmented by AI, not just a rubber stamp. It is important to keep monitoring for drift as trust builds.
Q: What kind of results are carriers seeing as they move up the maturity curve?
A: Straight-through processing up to 95% depending on claim type, 67% less handler time on coverage decisions, and one Sprout.ai customer saw a 19-point jump in Net Promotor Score after AI came into their claims process.
Q: Is Automator (workflow automation) really where most carriers are stuck?
A: Yes, that or Experimenter, piloting AI in some form. Getting all the way to Architect takes well-defined strategy and strong underlying operations, not just ambition. That’s why most carriers are currently clustered around automation and AI pilots. Those who have progressed beyond that have not waited for perfect conditions, they made a start and built on it over time.
Q: What does it feel like once you hit Stage 4?
A: When AI is augmenting claim decisions, cycle time drops, adjuster time per claim falls, and ROI shows up within the budget cycle. The part people miss is that adjusters are actually using the recommendations and seeing value in them, and the system keeps improving with every claim processed.
Q: Do we have to move up the curve one stage at a time?
A: No. A greenfield operation functioning manually can jump straight to AI-augmented decisioning. What matters most is knowing your primary goal – cost, accuracy, or customer service – since cycle time and satisfaction are usually the easiest wins to prove early.
Q: What brings adjusters on side?
A: Start with the parts of their day that cause most pain, and automate those first. And put a few adjusters who are genuinely curious about the tech onto the project team. They end up doing more to bridge the gap than any training session will.
Q: How do you decide where to start? What’s the best first use?
A: FNOL coverage checking is a strong first move. It’s early in the process, it is consequential, and the data you need is well understood. According to research conducted by Sprout.ai in January, half of complex claims take over a week to confirm coverage, and 15% take three months or more (get The State of Policy Coverage Checking report here). That delay is what slows everything downstream.
Q: Is adoption more about adjusters trusting the AI’s reasoning?
A: Trust in AI significantly increases adoption. If adjusters see AI as something that helps them do the job better, rather than another box to tick or a system to comply with, you move up the curve faster. If they don’t, no amount of extra AI fixes that.
Q: What’s the single biggest thing leadership can do to support this kind of change?
A: Embrace the change themselves, not just endorse it from the sidelines. Paint a picture of where you’re headed, not just the burning platform you’re escaping. And track how adjusters are feeling about it alongside your cycle-time numbers. Both matter.
Q: Before starting, what are the warning signs that a pilot may stall once it hits production volume?
A: Check these before you launch: does the data pipeline work at real volume, not just test volume; can your system take AI recommendations back in; were adjusters involved in designing the pilot; is there an override and audit framework in place. Miss any of these and production rollout is at risk.
Q: What kills AI pilots, and how do you get ahead of it?
A: Sponsor turnover and mismatched budget cycles are quiet killers, along with success metrics nobody agreed on upfront. One we see a lot is that an innovation team runs the pilot, but they’re not the ones who’ll own it in production. It is important to get operations, tech, innovation, and the business in the room together from day one.
Q: What’s the most common mistake teams make moving from pilot to production?
A: Treating it as a technology problem when it’s usually an operational one: messy data, decision logic that’s never been written down, limited integration, adjusters who weren’t brought along early. The good news is those are all within your control and fixable, and the gap is smaller than it feels when you’re stuck inside a stalled pilot.
Q: How long does it take to go from a working pilot to production?
A: There’s no single industry number for this, but Sprout.ai customers typically move in weeks to a few months, not years, as long as the operational blockers get fixed upfront.
Q: Our CTO wants to build claims AI in-house. What should we weigh up first?
A: You need real insurance expertise – policy wording, exclusions, claim-specific reasoning – not just general ML skill. AI talent is brutal to hire and keep right now; the wage premium over non-AI roles has passed 56% and doubled in a year. Also consider that a partner who works across many customers’ claims gets a flywheel effect that an in-house build won’t.
Q: If there’s one thing we could do today to progress our AI project, what would it be?
A: Ian Thompson: Start now, and treat all of it as a learning exercise. Don’t wait for perfect conditions, or whatever’s coming next.
A: Roi Amir: Pick one concrete decision for AI to support, get honest about where you are today and where you want to be (check the eBook). Then build it for production from the start, not just as a proof of concept.