Turning lab results into a conversation

Fullscript Labs shipped result delivery in 2024, but the bet behind interpretations went further: if practitioners can attach context to every biomarker — where a value sits, what it means, what to watch — the result becomes a conversation instead of a document. Two users share the artifact: practitioners need to annotate fast enough to do it for every patient, and patients need to read it without clinical training.

Every biomarker renders on a range bar — low, suboptimal, optimal, and back — in the Labs biomarker palette. The zones belonged to the clinical team; my work was making the same bar read correctly at card size, in dark mode, and for values that sit far outside any zone, where proportional rendering would break the scale.

Showing a graph from the first result implies a direction that doesn't exist, and hiding it entirely risks patients assuming something is broken. In the biomarker-trends sessions I ran, a fabricated slope on a health metric lost to an empty state: the graph appears only once a biomarker has 2 or more results. A single result gets a deliberately written blank, because an honest blank beats a misleading line.

Reassuring copy — "this looks good!" — is friendlier and tests well in isolation, but a biomarker's meaning depends on context only the practitioner has, and a cheerful label on a conditionally fine value is a claim the platform can't stand behind. The content designer's editorial rules held: no good/bad labels, no emotional framing, facts the practitioner can own. The copy reads cooler than consumer health apps, and trust here is clinical, not conversational.

The hardest feedback in the project came from our chief medical officer, who said plainly that the early interpretation content wasn't ready. We rebuilt the summaries against the editorial rules instead of softening the wording around the problem, and the bar never dropped again.

AI-generated summaries made interpretation-per-patient feasible at scale, but an unreviewed generated sentence about someone's blood work is a liability, not a feature. So in the builder I co-designed, AI drafts against the editorial rules, and the practitioner reviews, edits, and owns every interpretation before it reaches a patient. Review adds a step to their flow — the practitioner's name is on the note.

The platform has since priced that bet: plans built with its AI clinical support carry more than twice the recommendations of a practitioner-authored one, and they convert at roughly half the rate. The friction even has a name now — information overload. AI is good at generating more of something; review wasn't the tax on it.

Interpretations entered production in August 2024, AI-drafted summaries followed in September, and coverage expanded to every Quest test in October — the same month Fullscript acquired Rupa Health. Auto-sending arrived in early 2025, so a finished interpretation reached the patient without the practitioner remembering to send it.

Months after launch, a practitioner who gave Labs a glowing testimonial turned out to be managing results in a spreadsheet; she didn't know interpretations existed. Getting found is its own design problem, and launching a feature doesn't solve it.

The patient-side number still isn't mine to quote, but the clinical loop is measurable and it moved. Two years on, the platform puts the gap between a result landing and a patient holding an interpretation at about four days, down from about four weeks, and the whole diagnostic journey at less than half its old length. The lab was never the bottleneck; the wait was for a clinician to explain the result.