Everyone wants to believe AI can just figure out messy data on its own. AJ McGowan, VP of Research and Development at Reynolds and Reynolds, says that belief is exactly what stops most AI projects from ever reaching 100%. He joined Auto Collabs to break down Ray, Reynolds' proprietary AI agent, and three ideas worth stealing before your own dealership bets big on AI.
1. Stop Treating "Messy Data" as an AI Problem to Solve Later
The myth vs. the reality:
The Common Assumption | What McGowan Actually Sees |
|---|---|
AI can just "figure out" messy data | Fuzzy matches never fully resolve, they just get less wrong |
More AI flexibility = less engineering work | Engineering discipline still determines whether results can be trusted |
A 90% accurate answer is close enough | 90% is often the hardest wall to clear, not the easiest |
If I'm going to have an AI in my dealership and treat it like an employee, I want all of my data to interoperate the same way I want all of my software systems to interoperate if I was going to teach a human employee.
Consider this today
Before evaluating any AI tool, audit whether your CRM, DMS, service, sales, and inventory data can actually talk to each other. That's the real starting line, not the AI itself.
🎙️ Listen to today's Auto Collabs episode for the complete discussion.
2. A 90% Answer Isn't Progress. It's a Trap.
Here's the counterintuitive part: giving AI more freedom to guess at relationships between systems doesn't get you closer to a reliable answer, it gets you stuck at "pretty close" forever.
We're not going to get the best result if we give it the hardest possible problem to solve. Let's not tie its arm behind its back from the working gate and hope it gets it right every time, since it's fundamentally non-deterministic anyway.
An AI agent guessing at connections every time will always be a little bit wrong. An agent working across data that's already properly connected doesn't have to guess at all.
Consider this today
If a tool's pitch is "don't worry about your data, we'll figure it out," treat that as a yellow flag, not a feature.
3. What Ray Can Do Now vs. What's Coming Next
✅ Today: Answers cross-departmental questions on demand, inventory counts, meeting prep, surfacing wins and opportunities across the whole Reynolds ecosystem
🧪 In testing: asked who to "yell at" before a meeting, Ray refused and gave management advice instead, a response the team is actively tuning out
🔜 12 months out: moving from answering questions to actually taking action inside dealership workflows
Consider this today
Start thinking now about which repetitive manager tasks you'd hand to an AI agent first, once it's ready to act, not just answer. That list will matter sooner than it feels like it should.
The dealerships winning with AI in a year won't have the fanciest chatbot. They'll have data that actually deserved to be trusted in the first place.
Thanks to AJ McGowan and the team at Reynolds and Reynolds for the conversation. The gap between a 90% answer and a real one isn't a technology problem, it's a decision about how seriously a store takes the data underneath everything AI is about to be asked to do.
