What we learned about full resolution, guardrails, and why that last 10% is harder than it looks.
Most AI tools in insurance stop at triage. They categorize a ticket, maybe route it. SageSure's AI reads an email, updates the policy, and replies to the customer — no human touches it. That's not a roadmap item. It's live today, across 850,000+ policyholders in catastrophe-exposed residential property.
We sat down with Pete Rizzo, AVP of Service and New Business Onboarding Optimization at SageSure, and Alex Taylor, Global Head of Emerging Technology at QBE Ventures, to hear how they got there. Pete runs AI across one of the largest MGUs in the US. Alex evaluates and invests in AI from inside one of the world's largest insurers. Isharna Walsh, Lorikeet's Head of Product and Design, moderated.
If you're running CX at an insurer, MGA, or carrier and evaluating where AI fits — here's what two people further down that road wish they'd known.
From categorization to full resolution — in four phases
Pete described a journey that started small and expanded deliberately:
Phase 1: Categorize. AI reads incoming emails and identifies what the request is about.
Phase 2: Route. The right email gets to the right person, with priority flagging.
Phase 3: Reply. Standard FAQ-type responses — "you get an email for this, I tell you here's what you reply with."
Phase 4: Full resolution. End-to-end — no human in the loop.
"Not only can Lorikeet receive an email on my behalf, they could tell me what the email is, they could make updates to the policy, and they could reply to the customer without me ever needing it to cross the desk of one of my employees." — Pete Rizzo, SageSure
This is where most AI implementations stall. Getting from triage to actual resolution — where the AI takes action in your policy system — is the gap between a helpful tool and operational leverage.
Email was the unlock
Pete made a point that surprised the audience: the channel that mattered most wasn't voice or chat. It was email.
"So many companies that I was exploring were limited to either phone or phone and chat. For me, I really wanted to lean in on the email work because we're still receiving a high volume of emails from our producers around the country." — Pete Rizzo
Insurance still runs on email. Independent agents across the US send status requests, document uploads, and policy change requests via email every day. If your AI tool doesn't handle email well, it's missing the highest-volume channel — and the one where your producers already work.
Alex reinforced this from the channel consistency angle. It doesn't matter how good your AI is on one channel if the experience fragments across others:
"There's nothing worse than coming in through one channel only to then look at that same thing… and it's just not there." — Alex Taylor, QBE Ventures
The takeaway: evaluate AI tools on how they handle your actual channel mix — not just the channel that demos best.
You don't need full automation to get massive value
Alex brought up claim history summarization as an example of AI creating real value without replacing anyone:
"Being able to compress that 10 to 15 minutes down into literally 30 seconds and under from what we've seen so far is a massive opportunity." — Alex Taylor
Before AI, an agent picking up a complex claim had to spend 10–15 minutes reading through case file notes before they could even start the conversation. That's dead time for the agent and the customer on hold. Compressing that to 30 seconds means the same team handles more cases, faster, with better context.
This is an important reframe for anyone who thinks AI adoption is all-or-nothing. You don't need to hand the entire conversation over to a machine. Even something as straightforward as claim summarization can change the economics of how your team operates.
Guardrails and auditability earn trust
Both speakers kept returning to the same theme: control.
"Every workflow that I've gone live with, I'm fully in control that I have access to a UI experience where I can audit that work." — Pete Rizzo
"Guardrails is the thing for me. It's the same as it is with people, with AI systems." — Alex Taylor
Pete described his approach as "sharpening the pencil" — auditing each workflow and pushing accuracy north of 95% before moving to the next one. He doesn't scale until he's earned the right to. Alex compared it to the training programs companies already run for human agents: we already hold people to a compliance bar, and AI should meet the same standard.
This matters for anyone building the internal case for AI. The auditability story is what gets compliance and risk teams comfortable. The old model — sampling a fraction of a percent of calls and hoping for the best — is replaced by continuous, 100% coverage. That's not a concession to regulators. It's an upgrade.
Build vs. buy: that last 10% is harder than it looks
Alex was direct about the temptation to build internally:
"That last 10% is a lot harder than it looks. Making sure that you can do it efficiently and effectively and continuously in your organization — it's probably not something that you want to bite off." — Alex Taylor
Pete flipped the frame: even if your long-term goal is to build, the work you do with a partner now isn't wasted.
"All of the pre-work that you're doing now to set your AI partner up for success is only pre-work for when you may make that decision to actually bring it in-house." — Pete Rizzo
This was the most practical advice from the session. Partnering now builds the operational muscle — documented processes, tested workflows, accuracy benchmarks — regardless of where the technology lives later. And the economics of recruiting and retaining the engineering talent to build it yourself are often worse than they look on a spreadsheet.
Where do you start?
Pete's four-phase journey is a useful framework for anyone early in this process:
Categorize — let AI tell you what's in your inbox
Route — get the right request to the right team, fast
Reply — handle the FAQ-level work automatically
Resolve — end-to-end, no human in the loop
Most teams can start with phase 1 in weeks. The question isn't whether you'll get there — it's how quickly you start building the muscle.
The full session is available on demand:
If you're exploring AI for insurance CX — whether you're starting with categorization or ready for full resolution — book a conversation with our team.
Book a call
See what Lorikeet is capable of
Share this article
