Scaled AI QA 100× to cut churn −3 % in telehealth

Overview

Found’s telehealth consult experience suffered from hidden friction — slow consults (72 h avg), inconsistent quality checks (5 %), and a 42 % churn rate.

‍
I led a mixed-methods, human-in-loop AI initiative that combined NLP, SQL audits, and qualitative synthesis to scale QA 100× in four weeks — unlocking full visibility across 5 K+ consults.
This work improved conversion +12 %, retention +22 %, and reduced churn −3 %.

‍

Business Challenge

• No visibility into consult data; Product and Clinical Ops worked in silos.
• QA covered only 5 % of cases; insights were anecdotal.
• Patients faced slow turnaround and confusion in provider messaging.

Human Problem

Patients waiting on prescriptions felt uncertainty and distrust.
Clinical teams lacked timely signals to intervene.
We needed a transparent, automated system that protected quality while accelerating care.

Approach

• Audited 5 212 consult threads using NLP, REGEX tagging, and Gemini LLM prompts.
• Built a human-in-loop AI audit pipeline — machines for scale, humans for nuance.
• Partnered with Product, Data Science, and Clinical Ops to define “Consult Confidence Score” KPIs reused org-wide.
• Cohorted users by risk and designed proactive triggers to prevent churn.

Insights

We uncovered four friction sources: Labs, Insurance, Pricing, and Tone.
Early lab requests or unclear cost explanations were the biggest churn drivers.
Fast-closing consults shared consistent behaviors — decisive users, concise provider replies, and clear upfront pricing.
AI accelerated discovery 10×, revealing high-value levers Product and Ops could act on immediately.

Insights → Action (4 Weeks)

I co-led workshops uniting Product, Data Science, Engineering, and Clinical Ops to translate findings into immediate product changes.
We prioritized four MVP fixes: delayed lab prompts, insurance clarification, provider messaging templates, and AI-based risk alerts.
All deployed within 4–8 weeks, driving measurable lift in conversion and retention.

‍

Impact

Retention +22 % | Churn −3 % | QA Coverage 5 % → 100 %Consult time dropped 72 h → 64 h, completion 53 % → 57 %.The new QA system became the foundation for org-wide AI-enabled research and operations.

Leadership & Influence

Principal-level IC (L8) — owned end-to-end research + system design, and aligned executive stakeholders across Clinical Ops, Product, and Data Science.
Unified teams around shared metrics and a common operational language for QA and consult success.
Defined the “AI-assisted QA” playbook adopted by other care domains.

Global Organizational Impact

Unified Clinical Ops and Product under one data-driven framework.
Turned a black-box process into a transparent, measurable system that scaled quality, reduced churn, and proved the business ROI of AI-enabled research.

‍