From a three-day queue to three minutes

Harborlight Health had a queue problem, not a model problem. Requests came in by email, by form, and by phone note, and a small team read every one in the order it arrived. On a busy week, the first response took three days. The urgent cases waited behind the routine ones, because nobody had time to sort them first.

The gap

The work wasn’t the responding — the team was good at that. The work was the reading and sorting that had to happen before anyone could respond. That was the gap, and it was invisible on every org chart.

The bridge

We built a triage layer that sits in front of the queue. It reads each incoming request, drafts a first response, and routes it to the right team with a suggested priority. A person reviews every draft before it goes out — the model never has the last word.

“The triage layer doesn’t replace the team. It hands them a sorted desk every morning.”

• Unstructured intake — email, forms, phone notes — becomes one normalized queue.
• Each item arrives pre-summarized, pre-categorized, and pre-prioritized.
• The urgent surfaces to the top automatically, instead of waiting its turn.

The result

Time-to-first-response dropped from three days to about three minutes for the common cases — a 94% cut — and the team stopped starting each day underwater. The headline number is nice. The quieter win is that the people doing the work got their attention back.