Agentic ≠ chatbot: a working definition
Last week three different vendors at NADA pitched me their "agentic AI for dealerships." Two of them were chatbots with a thesaurus.
The word is getting muddied fast. Here's how we draw the line internally:
A chatbot answers questions. An agent takes actions.
A chatbot needs you to ask clearly. An agent figures out what you want from context.
A chatbot lives in a chat box. An agent runs on whatever surface the user is already on — phone, SMS, email, in-app.
A chatbot is constrained to producing text. An agent has tools — real ones that mutate state in your real systems.
A chatbot is one-shot. An agent maintains state across turns and across surfaces.
Why does this distinction matter? Because the value calculation is different.
A chatbot deflects support tickets — saves you ~$3 per ticket times some volume. The math works out to maybe $50k/year for a mid-size dealership. Real money, not transformational.
An agent books appointments your team would have missed entirely. A 24/7 receptionist that handles every after-hours call adds up to an extra 8-15 sales per month for a typical store. At a $3,000 PVR, that's $300-540k/year.
Same letters in the marketing copy. Two-orders-of-magnitude different in dealer P&L impact.
When you're evaluating any AI vendor, ask exactly two questions:
1. "What real action does this take in my real systems, with what authority?"
2. "If your AI is wrong, what happens to my data, my customer, my OEM relationship?"
Vendors who can't answer #1 specifically are selling chatbots. Vendors who can't answer #2 calmly haven't thought hard enough about the deployment.