How Booking Extraction Works

Airflow Core — The engine that turns unstructured emails into clean, actionable booking data.

From messy emails to clean data

Every OTA formats their booking confirmations differently. Airbnb uses one layout, Booking.com another, VRBO another still. Manually copying guest names, dates, and amounts from each email into a spreadsheet or accounting system is tedious and error-prone.

Airflow handles this in seconds.

What gets extracted

When a booking email enters the pipeline, the AI identifies and extracts:

Guest name — first and last name, used to create or match a guest contact
Check-in and check-out dates — parsed from any date format
Number of guests — adults, children where specified
Nightly rate and total amount — the gross booking value
Platform fees — Airbnb service fees, Booking.com commission, etc.
Cleaning fees — separated as their own line item
Net payout amount — what actually hits your bank account
Confirmation code — the OTA's reference number for cross-referencing
Platform source — which OTA or channel the booking came from

How the pipeline works

The extraction isn't a single step — it's a multi-stage pipeline designed for accuracy:

Email classification — the system first determines whether the email is a booking, a modification, a cancellation, or a review notification
Content parsing — Airflow extracts structured fields from the email body and subject line
Validation — extracted data is cross-checked for consistency (do the dates make sense? does the pricing add up?)
Record creation — a booking record is created in your portal, linked to the correct resource
Guest matching — if the guest has booked before, their existing profile is updated rather than duplicated

Modifications and cancellations

Airflow doesn't just handle new bookings. When you forward a modification email, it updates the existing booking record with new dates or pricing. Cancellation emails update the booking status and adjust any related invoices.

Accuracy and edge cases

Airflow's extraction engine is built on thousands of real booking email formats. For standard OTA confirmations, extraction accuracy is extremely high. For unusual formats — like a handwritten direct booking enquiry — the system does its best and flags anything it's unsure about.

If Airflow determines an email isn't actually a booking (a marketing email from an OTA, for example), it's classified as "not a booking" and doesn't consume an action credit.

One action credit per booking

Each successfully extracted booking consumes one action credit from your plan. The credit is consumed at the point of record creation, not at email receipt. This means:

Failed extractions don't consume credits
Non-booking emails don't consume credits
Duplicate emails (same booking forwarded twice) are detected and don't consume additional credits