Alpin Motel - An AI Email Assistant for the Front Desk

Client Overview

Alpin Motel & Conference Centre is a 40-unit motel in Whakarewarewa, Rotorua, with two conference rooms that host meetings, sports teams, family events and weddings. Its front desk inbox handles everything from direct booking enquiries and group requests to booking channel notifications, supplier invoices and council paperwork.

The Challenge

One shared mailbox was doing several jobs at once. Guest questions about dates and rates sat alongside reservation notices from Booking.com, Expedia, Agoda and Airbnb, cancellations, invoices and a steady stream of spam, including scam "reservation enquiries" designed to look like real guests.

Answering a booking enquiry properly meant checking live availability and rates, and replying in the motel's own friendly tone. Sorting the rest meant knowing which of the property's folders each email belonged in. The motel wanted help with both, without handing control of guest communication to software.

The Solution

Automate AI built and runs an AI email assistant that works inside the motel's existing mailbox. It reads new mail, works out what each email is, and either drafts a reply for staff or files the email where the team already looks for it.

The most important design decision was what it cannot do. The assistant has no way to send email at all, and it never creates, changes or cancels a booking. Every reply is saved as a draft with a review banner and an internal note explaining what it found, so a staff member reads it, edits it if needed, and presses send.

Replies are written in a tone profile based on the motel's own past emails, and the drafting rules forbid quoting a price or availability that was not read live from the booking engine, or sharing payment details.

Under the Hood

• Claude AI models: a fast model classifies each email and extracts dates, guest numbers and booking references; a stronger model writes the drafts.
• Preno: the assistant reads the motel's public booking engine for live availability and rates, and looks up existing bookings in Preno.
• The existing mailbox: it works over standard IMAP with the motel's current email host, so staff keep the email client they already use.
• AWS serverless: scheduled and queue-driven AWS Lambda functions, with DynamoDB for state and an audit trail, and alarms that alert us if anything fails.

Tuned to How the Motel Works

Before launch, the classifier was tested against a labelled set of example emails written to mirror the real inbox, and it cleared the 90% accuracy target. Since going live, the motel's team has fed back on individual emails, and each one has become a specific rule: direct website bookings go to the Website folder, booking channel one-time codes stay in the inbox, the property's own scanner is never treated as spam, and invoices from accounting platforms, booking channels and the local council land in Invoices.

Because the motel empties its Spam folder, filing an email there is the one decision the assistant makes that could lose something. So a second, independent AI check now has to agree before anything is spam-filed, and genuine guests who mention payments are kept well clear of it.

The Outcome

The front desk now starts with booking replies already drafted, priced from the booking engine where it has the dates, and written in the motel's voice. Admin and channel mail is filed into the right folders without anyone dragging it there, and scam enquiries are moved out of the way.

Staff stay in charge of every word a guest receives. The same setup also supports small, personal outreach, such as drafting individual thank-you and review requests to past direct guests for the team to check and send.