Rebuilding claims intake for 4.2 million members

The context

Lattice Health processes roughly nine thousand claims every weekday across three regional offices. Every one of those claims starts its life on a single internal form that has accumulated fields the way a coat accumulates patches. The form had last been redesigned in 2014. Since then each new compliance requirement, each new line of business, each new insurance product had bolted another field or another conditional on top of the last.

By the time we were brought in, the form was taking adjusters an average of fourteen minutes to complete, and roughly a quarter of all submissions were being sent back for missing or contradictory data. The cost of that rework was estimated at $2.1M annually in adjuster time alone, not counting downstream delays to members waiting on their claims to resolve.

What we inherited

Lattice had already tried two previous redesigns. Both had stalled. The first had tried to replace the form wholesale and ran into three years of compliance sign-off. The second had added “smart defaults” that adjusters universally overrode because the defaults had been tuned for the wrong population. Whatever we did, we had to survive the compliance review and earn adjuster trust within a single release cycle.

The brief I took back to leadership

After the first two weeks of research, I rewrote the brief. The original ask was “reduce intake time”. The revised brief was “reduce the number of fields an adjuster must look at, and make validation a helper instead of a gatekeeper”. Leadership signed off on the revised brief on a Friday. Engineering had a working prototype on the following Thursday.

What made the difference

Three decisions, in order of how much they moved the numbers.

First, we hid every field that could be inferred or defaulted. Not removed, compliance still needed the values to land in the database, but hidden from the adjuster unless the inference confidence dropped below 80%. That alone removed 61 fields from the visible form on a typical claim.

Second, we moved validation from blocking to advisory. The submit button stays enabled unless a field will actively fail the downstream system. Everything else is a nudge in the right rail that the adjuster can address during review. This change alone reduced the number of “save and come back” round trips from an average of 2.3 per claim to 0.4.

Third, we built a live preview of the claim as it would appear to the downstream reviewer. Adjusters stopped making formatting mistakes because they could see the formatted output update as they typed. The preview is the most used feature in the post-launch telemetry.

“

The preview pane did for our adjusters what spell-check did for writers. They stopped proofreading and started moving.

What did not work

The first prototype used a single long scrollable form with smart sections. It tested badly. Adjusters reported feeling “lost” and “unable to tell where I am”. We moved to a three-step progressive layout with a persistent progress indicator and the problem disappeared. The research clearly preferred a structured path over a flexible one, which surprised us, the second redesign had failed partly because it was too structured.

Outcomes

Twelve weeks from kickoff to full rollout. Median intake time dropped from fourteen minutes to three. First-pass rework dropped 38%. The internal NPS for the claims tool moved from −22 to −3, then to +19 over the following six months as adjusters got fully fluent.

The three-region rollout let us catch two serious issues before they reached the whole organisation. One was a date-picker bug that only manifested with the Pacific-time locale; the other was a validation rule that was rejecting claims from a rare insurance product we had not included in the test cohort. Both were caught and fixed within the 48-hour rollback window.

What I would do differently

I would have spent more of the early weeks with the compliance team directly. We got lucky, the review was approved in one pass, but it could easily have taken three months and we would have missed the release window. Next time I start with compliance on day one, not week five.