Associate Product Manager · Carfax · Apr 2026 – present

I led the 0→1 design of Carfax Canada’s first agentic AI product: a vehicle-buying companion that helps used-car sales managers decide what inventory to buy and when to reprice aging units. I owned the PRD, prompt design, data schema requirements, and phased roadmap, then partnered with S&P Global Mobility's AI team to build it, unblocking issues and translating between technical and business stakeholders across a 20-person cross-functional effort.

Problem

A used-car sales manager is a capital allocator. Every week they decide which vehicles to buy, sourcing from auctions, trade-ins, and private listings, and those decisions are made largely on intuition and past sales history, supported by Carfax reports that surface accidents, salvage and stolen titles, and open recalls.

That process leaks money in two places.

The decision loop is slower than the market. A manager identifies a promising unit, confirms with their GSM, and works out logistics. By the time the answer comes back, the vehicle is frequently gone.

Aging inventory quietly drains margin. Every day a vehicle sits on the lot accrues floorplan interest and softens demand for that unit. The longer it sits, the worse both get.

Both are the same underlying failure: high-stakes inventory decisions made on instinct, in a loop too slow for the market they operate in.

Discovery

I ran three parallel tracks:

• Working sessions with the Carfax sales team, who hear dealer frustrations continuously and had pattern knowledge no single interview would surface
• User interviews with dealers, to understand the buying workflow and where it actually breaks
• Competitive analysis of the AI tools already sitting in dealer workflows

The finding that changed the build: dealers were already using AI to support buying decisions, in the form of general-purpose assistants like ChatGPT and Gemini. That reframed the question. It was never "will dealers adopt AI for this." It was "why would a dealer use ours instead of the one already open in another tab."

The answer defined the product. A general model reasons well but knows nothing about the specific VIN in front of the manager. Carfax has aggregated data across the vehicle lifecycle and flagship products like the Vehicle History Report. Our advantage was never going to be reasoning quality. It was grounding.

Decision

We positioned away from the chatbot category entirely. Not a conversational assistant that happens to know about cars, but a vehicle-buying companion: it reviews a dealer's current listed inventory and sales history, recommends which vehicles to buy, flags units that have been sitting too long, and suggests pricing action. Every recommendation is grounded in proprietary Carfax data rather than general web knowledge.

Three trade-offs made that possible.