Why we built Herald — and why it isn't a chatbot

Most "AI for dealerships" you see today is a chat widget on a website. A box you type into. Maybe it answers an FAQ.

That's not what dealers need.

Dealers need their phone line to actually be answered. They need test drives booked outside business hours. They need follow-ups on stale leads that don't feel like spam. They need their service department to stop losing same-day appointments because the receptionist couldn't get a tech on the line in time.

That's not a chatbot problem. That's an agent problem.

We built Herald around a single observation: in any service business, the highest-leverage place to put an AI agent is the inbound phone call. Not the website chat. Not the marketing email. The call.

Why? Because every call to a dealership is a customer who has done their research, decided to engage, and is willing to spend 90 seconds talking to a human. If you don't pick up, they call the next dealer on their list. If you pick up but can't book the test drive — they call the next dealer.

A chatbot in this context is worse than nothing. The customer wanted to talk; instead they got a form.

So Herald is built differently. It picks up the actual phone (Twilio + Vapi, sub-700ms latency). It speaks naturally (Claude Sonnet 4.6 with a voice warm-up loop). And it has tools — eight of them, all real — that let it look up your inventory, check sales availability, hold a slot in the calendar, send the confirmation SMS, and write the call summary back to the CRM.

When a customer hangs up, three things have happened: they got their answer, your team has a fresh lead in the CRM with full context, and the customer has a calendar invite and a directions link in their texts. Total elapsed: usually under three minutes.

We don't call this a chatbot because it isn't one. It's an agentic receptionist. The distinction matters.