← All case studies
Logistics / freightWorkflow automation

Turning 400 daily emails into dispatch records at Northline

Booking requests arrived as free text from 90 customers in no consistent format. Now they arrive as structured jobs, checked by one coordinator.

92%
Bookings auto-processed
11 min
Saved per booking
Volume handled, same team
Problem

Every customer had their own idea of what a booking looks like

Northline handled around 400 booking emails a day. Some were spreadsheets, some were three lines of text, some were photographs of a printed manifest. Four coordinators retyped them into the transport management system, and the errors that slipped through surfaced as failed collections.

Approach

Extraction with an explicit uncertainty threshold

An extraction pipeline reads each email and attachment, pulls out collection and delivery addresses, dates, weights and reference numbers, and validates them against the customer record and the address database. Above the confidence threshold the job is created automatically. Below it, the email lands in a review queue with the uncertain fields highlighted and the source text beside them.

Build

The review queue was the real product

We spent as long on the review interface as on the model. Coordinators correct a field in one keystroke, and those corrections feed the customer-specific rules, so a format that needed review in week one is usually automatic by week four.

Result

Coordinators moved from typing to exceptions

92% of bookings now process without human touch. The team that was retyping is now handling exceptions and customer calls, and Northline has taken on roughly four times the booking volume without adding headcount.

Before / after
Manual entry per booking
12 min1 min
Failed collections
2.1%0.4%
Coordinators on entry
41
Same-day confirmation
61%98%

We stopped hiring for data entry and started hiring for customer service. That was the whole point.

NFOperations Director, Northline Freight