Leads that route themselves, and flag when they're unsure

The situation

Event inquiries came in through the website and landed in a shared inbox without a reliable owner. Some sat for a day. Some got worked twice by two different reps. Nobody could say with confidence how long a lead waited before someone picked it up.

What we built

• A signed webhook endpoint that takes inquiries straight from the web form, with duplicate detection on the way in
• A routing rules engine keyed on location and event type, so the right rep gets the lead without anyone triaging it
• Immediate dispatch to the assigned rep by SMS and email, and an automatic push of the lead into that rep's CRM record
• A review queue for the sales director, and response-time reporting that didn't exist before

The details that mattered

• Every lead is scored by a language model on a 0 to 100 scale. Anything below the confidence threshold routes to the director for a human ruling rather than going out automatically.
• Each ruling is written to a training log with the model version attached, so the routing gets measurably better and you can see exactly when and why it changed.
• SMS opt-in is captured and stored at the point of submission, and STOP requests are honoured automatically.

What it proves

The question every owner asks about AI is what happens when it's wrong. This is the answer: a person sees the uncertain cases, their correction is recorded, and the system's mistakes are on the record instead of buried.