Sales-process audit — eye clinic, Lima (anonymized case study).
Started from the clinic’s real sales conversations across WhatsApp, Instagram and Facebook, anonymized and extrapolated them into a 450-conversation dataset to protect patient privacy, and ran the full audit: 7-stage operations diagram, per-stage drop-off, Pareto analysis (5 causes ≈ 82% of lost sales), loss estimated in money per cause (PEN 395K–575K/yr on the modeled volume), and remediation options per cause.
- 450
- conversations in the anonymized dataset
- 7
- stages in the operations diagram
- ≈82%
- of lost sales explained by 5 causes
- 395–575K
- PEN / yr estimated loss, modeled volume
Patient privacy by design
Healthcare conversations can’t be handled like ordinary sales data. The real chats were anonymized, then extrapolated into a modeled 450-conversation dataset that keeps the sales patterns and leaves the patients out. The same approach gives any clinic a full audit without exposing a single patient.
Problem
An eye clinic in Lima knew patients were dropping off in chat, but not at which stage, why, or what it was costing.
Approach
Each conversation was classified by an LLM against an analysis rubric I designed, grounded in sales-process and data-analysis frameworks from my curated knowledge base.
Outcome
Five causes ranked by impact, each with a money estimate and remediation options across process, people, technology and commercial. The figures are modeled estimates of loss, not revenue recovered.
Architecture
- LLM classification
- rubric design
- Pareto analysis
- Python