Request Intake

Every morning started with three hundred emails. An operator opened the attachment, found the addresses, weight and dates, and typed them into the system. One wrong address cost a run.

• 1,200 requests a day and 8 people on data entry alone
• Dozens of formats: emails, Excel, PDF, photos of waybills
• Typing errors only surfaced once the truck was on the road
• Half the volume lands inside a two-hour morning peak

We did not replace the accounting system. The agent sits in front of it: it reads the incoming message, extracts the fields, validates them and creates the request over the API.

• Parsing of the dispatch mailbox and attachments in any format
• Extraction of 18 request fields, each with a confidence score
• Address validation against a reference book and rates against the price list
• Unclear requests go to an operator's queue, not into the system
• An automatic reply to the sender with the request number

1. 01

   Incoming email

   The dispatch mailbox is read automatically; attachments are unpacked and recognised.

2. 02

   Field extraction

   Addresses, cargo, weight, dates and contacts are pulled into one structure.

3. 03

   Validation

   Addresses are checked against the reference book, rates against the price list, duplicates are dropped.

4. 04

   Request created

   The request goes into the accounting system over the API; edge cases go to an operator to confirm.

Next: an automatic reply with timing and price right in the email thread, before the request even enters the queue.