Conversational Design Machine Learning Automated Handling

AI-Assisted Self-Diagnosis

An end-to-end web-based diagnostic application embedded in BestBuy.com, letting customers self-assess hardware and software issues against 25 years of Geek Squad troubleshooting.

• 112% of customers completed the full flow in production — high engagement on a site built for buying, not troubleshooting
• 280% of testers — all screened for a real device issue in the past six months — said they’d use it
• 3Now the enterprise triage standard at BestBuy.com — the live site runs on the framework and flows established here, redesigned by Best Buy since

Current live version (same framework, but redesigned)

Created Business Value by Reframing a Call-Deflection Project

• 1Showed customers Best Buy repairs computers — most didn’t know (new demand)
• 2Gave an early estimate — so they’d know if it’s worth fixing (fewer dead ends)
• 3Sold the next step — repair or replace, trade-in, new device (more per sale)
• 4Sped up calls — passed what the customer typed to the agent (lower cost)

Three Facets of AI, Doing Three Different Jobs

This wasn’t one “AI feature.” It ran on three distinct pieces, each doing a specific job.

Conversational DesignCarried answers forward — naming the device, using transitions between steps — so a guided flow felt like a conversation, even with pre-set answers

Machine LearningMatched symptoms to likely issues, trained on 25 years of Geek Squad calls

Automated HandlingRouted each customer to a fix, a chat agent, or an appointment

Pre-set answers weren’t only a constraint. Selecting is often faster than typing, and faster than putting a problem into words — a good interface shortens the conversation on purpose.

This was the services team’s first step into conversational AI. Running the triage tool live taught us about intent and natural language — and fed the conversational agent the team built next.

Looking Back

Much of what made this work wasn’t in the original scope. The standalone experience, the non-logged-in path, the adaptive credential model, the conversational-design groundwork — each one I had to identify, make the case for, and drive through, often by building alliances with other teams to get it done.

The organization was optimizing for the simplest possible version; I was building for the version that actually served users and captured the business value the simple version left behind.