FNOL: Rebuilding the Front Door to Claims at Scale

When I took on First Notice of Loss, the digital filing experience was the highest-leverage and most fragile point in the entire claims journey. FNOL is the front door. It is the moment a customer reports a loss, often right after a stressful event, and it quietly sets the cost, speed, and trust of everything that follows.

Three problems sat on top of each other.

Filing was a long, multi-step flow, and customers abandoned it partway through. Only 12 percent of filings completed through the digital channel. Every customer who dropped out fell back to a call center, which costs far more to serve than a digital start.

The economics were simple and unforgiving. Serving a customer through digital filing cost about $1.30. Serving the same customer through voice cost about $2.80, more than double. So every abandoned digital filing became a contact that cost us roughly twice as much, and at 18 million customers that gap compounds fast.

The damage did not stop at cost. Incomplete or inconsistent capture at the front door pushed rework and fraud risk downstream, because the rest of a claim is only as clean as the data captured in the first few minutes. A weak front door quietly taxed everything behind it.

I owned FNOL filing end to end: strategy, roadmap, discovery, and outcomes. My job was to turn the front door from a leak into an advantage, by lifting digital completion, cutting cost to serve, and capturing cleaner data upstream so the rest of the claims lifecycle could run faster and safer.

I led a cross-functional team of 11 with no direct reporting lines, so the work moved on earned influence rather than hierarchy.

The pressure was to optimize for the obvious headline: fastest possible filing with the fewest dropouts. The easy win was to strip steps out of the flow until the number looked great. I chose not to do that, because stripping steps would have sped up filing while degrading the data captured at the front door. That would have raised fraud leakage and rework downstream and simply moved the cost out of sight.

Instead I made two deliberate calls.

First, I optimized the existing flow with evidence rather than instinct. We measured the dropout rate at every page and dug into why customers left, both quantitatively through the funnel and qualitatively through research, then fixed the points that actually lost people.

Second, I chose focus over a grand rebuild. Rather than rebuild a single universal flow covering every claim type at once, I concentrated on the highest-volume claim type first and left lower-volume types on the legacy path longer. This reached measurable value far faster, at the cost of a temporarily uneven experience across claim types, a trade I was willing to defend.

Generative AI came in selectively, not everywhere. The result was a hybrid experience: a conventional screen-by-screen flow for the straightforward steps, and an ML and GenAI-driven experience reserved for the genuinely hard problem, peril selection, where it removed real friction. AI earned its place where it solved a problem, not because it was fashionable.

Across the six-product portfolio I lead, we drove 10 percent year-over-year growth in users over three years and 35 percent year-over-year growth in adoption. Those are portfolio outcomes, so I do not claim them as FNOL alone.

What FNOL owns is sharper and fully defensible. I improved digital completion from 12 percent to 14.2 percent over three years. Because every filing kept in digital saves about $1.50 against voice, a year-one lift in digital filing saved close to $100,000, and that saving recurs and compounds year after year. Just as important, cleaner capture at the front door reduced rework and gave fraud detection and downstream automation better data to work with.

The honest summary: FNOL was a cost-side product, and I treated it like one. I won on completion and data quality first and refused the vanity speed metric, which protected the economics of the whole claims lifecycle rather than flattering a single dashboard.

Peril selection was the hardest moment in the flow, the point where customers had to classify what had happened to them, and it was where ML and generative approaches earned their place. Automating and guiding that step cut friction exactly where the conventional form failed customers most, and it captured cleaner structured data at the same time.

This selective use of AI inside FNOL was the proving ground for a larger bet. My flagship agentic AI work, a multi-agent claims system on Google's ADK, is told in full in the Tow Claims case study.