AI Email Intake for Reservation Workflows

Reservation requests arrive as unstructured emails from customers and affiliates. A human operator has to read each one, interpret the trip, identify what information is missing, resolve any known context about that customer, map the request onto the reservation model, and decide what still needs follow-up.

That interpretation work happened by hand on every email. It was slow, inconsistent between operators, and easy to get subtly wrong — and the cost grew with volume.

Generation was never the hard part. A model can extract fields from text. The hard part is operating reliably on messy, domain-specific input — and that's a workflow problem, not a prompt.

I designed intake as a workflow that decides how much automation each email earns, rather than a one-shot extraction call. The model classifies intent and extracts fields, while validation and missing-field detection decide whether the request is complete enough to import. Complete requests with enough account context go straight to booking. Requests with missing information go to an operator.

Learning is downstream of that review path. If the operator fills missing fields and the booking is created, the learn agent compares the completed booking against the original draft and saves verified addresses and preferences back to account memory. That memory is what lets future requests skip review when the same gaps can be filled safely.

An inbound email is classified for intent, then fields are extracted and checked against what the account already knows. The draft is validated against the reservation schema, and missing fields are surfaced rather than guessed. Complete drafts for accounts with enough verified context pass the auto-import gate and become bookings. Incomplete or low-confidence requests go to operator review and only trigger learning after the operator fills the gaps and the booking is imported.