Blog · APR 5, 2025 · 3 min read
Our First Pilot: What Ten Medical Reps Taught Us in Thirty Days
Ten medical reps, thirty days, no cherry-picked users, here is everything the pilot broke that our internal testing never caught.
We ran our first FieldVoice pilot with ten medical reps across two mid-sized pharma territories in central India. Thirty days, no cherry-picked users, no hand-holding beyond a single onboarding call. We wanted the ugly data, not the flattering demo.
The first thing we learned: language switching happens mid-sentence, not mid-report. Reps would start a call summary in Hindi and drop into English for the drug name, then use a regional word for a symptom their doctor had used. Our early model treated each report as single-language. We rebuilt the pipeline to detect code-switching at the phrase level within the first two weeks.
What broke
- •Signal dropped in outdoor markets and rural clinic parking lots more often than we modeled, roughly one report in six recorded with zero connectivity
- •Reps back-dated reports out of habit, carried over from paper diaries, which meant our sync logic needed an explicit "logged for" date separate from "logged at"
- •A few reps tried to game daily call averages with very short voice notes; we added a minimum-duration flag rather than a hard block, since some visits genuinely are quick
What worked
By day ten, average logging time per call had dropped from a self-reported 45 minutes of end-of-day paperwork to under 5 minutes of voice notes recorded right after each visit. Reps stopped batching reports entirely, which is the behavior we were actually trying to produce. Batched, end-of-day reporting is where most CRM data quality problems start, because memory fades and calls blur together.
The regional sales manager overseeing the pilot told us something we now repeat in every sales conversation: he could see call notes the same evening, not three days later. That immediacy changed how he coached the team, because he could correct a messaging problem the next morning instead of the next month.
We extended the pilot by fifteen days at his request. Adoption held at nine of ten reps using the app daily without prompting, which is the number that mattered more to us than any feature list.