Case study · Boucan Factory
Boucan Factory: 100% of qualified calls analyzed by an AI agent in production.
The execution quality of every call is the product being sold. It used to be judged by ear, on about 5% of calls: we made it measurable across all of them, with an agent built on Anthropic's Claude technology, in production since September 7, 2026.
- Sector
- B2B prospecting · Meeting booking
- Scope
- 7 sales reps · 10 client accounts
- System
- Boucan platform · HubSpot
- Production go-live
- September 7, 2026
100%
of qualified calls analyzed, versus a manual sample of about 5%
+4%
booked meetings per week since deployment, over the first week of production
1,436
calls analyzed since going into production
The client
Boucan Factory runs a collective of sales reps who do phone prospecting and meeting booking for its B2B clients. The operation spans 7 active sales reps across 10 client accounts.
The problem
The execution quality of every call is the product being sold, and it was judged by ear: about 5% of calls listened to at random, four hours of supervision a week, weekly feedback at best, no shared scorecard; two listens to the same call could lead to two different conclusions. Recurring weak spots went unnoticed, getting new reps up to speed depended on the luck of the sample, and the quality delivered to end clients could not be measured.
The solution
Boucan Coach, an agent designed and deployed by Growth Wave on Anthropic's Claude technology, running on Boucan Factory's own Anthropic account.
- A scoring grid of 9 sales skills, defined with Boucan Factory's sales team
- Every call transcribed, filtered (voicemails and calls that did not connect set aside automatically), then analyzed criterion by criterion with quotes verified against the transcript
- Architecture principle: Claude judges, the system computes. No score comes out of the model; scores are computed deterministically from the judged criteria
- Quality control: a second verification pass on every analysis, calibration on 30 real calls, an evaluation set of 25 cases replayed before any change
- Delivered in the tools the team already uses: debriefs written in Boucan's tone, statistics in the Boucan platform and in HubSpot
Results
- 100% of qualified calls analyzed, versus a manual sample of about 5%
- 1,436 calls analyzed since going into production, versus a few calls a week
- A same-day debrief for every sales rep, instead of weekly feedback at best
- 4 hours a week of systematic listening refocused on coaching
- Booked meetings up 4% per week since deployment, over the first week of production, where performance used to be unmeasurable
What changed
Coaching no longer depends on how much listening time is available. Recurring weak spots are visible per sales rep and per account, and the quality delivered to end clients becomes a tracked indicator, not an impression.
Getting new sales reps up to speed no longer depends on the luck of the sample: each one gets a same-day debrief, on the same grid for everyone, with the call excerpts that justify each point. Supervision keeps its four weekly hours, now spent coaching rather than listening.
“Today I can see exactly what's working and what isn't. It saves me a lot of time on getting sales reps up to speed, especially my new ones. I have to have this tool.”
Translated from French. Original quote: « Aujourd'hui je vois exactement ce qui va et ce qui ne va pas. Ça me fait gagner beaucoup de temps sur la montée en compétences de commerciaux, en particulier pour mes nouveaux. Je suis obligée d'avoir cet outil. »
Under the hood
- Analysis and verification
- Claude Sonnet 4.6, Batch API, prompt caching
- Filtering and built-in assistant
- Claude Haiku 4.5
- Transcription
- Deepgram nova-2
- Quality control
- Second verification pass on every analysis · calibration on 30 real calls · evaluation set of 25 cases replayed before any change
- Analysis cost
- About $0.08 per call, all-in
The Anthropic API is billed directly to Boucan Factory's Anthropic account: the agent runs on the client's side.
Publication
- Published
- September 13, 2026
- Reference
- GW · CS-2026-005
- Project team
- Growth Wave · design and deployment: Raphaël Masson
- Boucan Factory: Pauline Perez
Case study published by Growth Wave OÜ, with the agreement of Boucan Factory. Production go-live on September 7, 2026; figures taken from the agent's usage log over the first week of production and validated by Boucan Factory. The increase in booked meetings is measured since deployment, with no causality claimed. Boucan Factory's end clients are not named.
Is your execution quality still judged by ear?
Tell us about your CRM, your team and where you're losing time. We reply within 24h.