We automated inbox triage, shipment matching, action identification, and data extraction so operators could review what requires attention instead of repeatedly reading, interpreting, and cross-referencing every email.

Impact

64% less operator time spent processing operational email

Measured operating results.

MetricBeforeAfterChangeOperator time spent on email triage3.9 hrs/day1.4 hrs/day-64%Average handling time per actionable email3.6 min1.1 min-69%Emails requiring full manual review100%27%-73 ptsAverage daily inbox backlog146 emails31 emails-79%

Operating scale

~18,000 active shipments per month, with approximately 1,100 operational emails arriving across shared and individual inboxes each business day.

Context

Freight operators were effectively using email as an unofficial workflow system. Every arrival notice, booking confirmation, schedule change, invoice, customs request, POD, detention warning, and pickup request had to be opened, understood, matched to a shipment, and mentally translated into the next action.

The problem was not simply the minutes spent typing. Operators were continuously switching context: email, TMS, shipment, attachment, customer history, back to email. Even emails requiring no action consumed attention because someone still had to determine that they required no action.

The operating inbox contained materially different freight events, from critical bills of lading and customs requests to schedule changes, invoices, PODs, detention notices, and routine warehouse communications.

What was broken

• Operators manually opened and interpreted almost every operational email before deciding whether it mattered.
• Identifying the corresponding shipment required searching across booking numbers, container numbers, B/Ls, customer references, vessels, and other identifiers.
• The same inbox mixed urgent exceptions with routine documentation, invoices, confirmations, and informational messages.
• Operators had to remember what each message implied operationally: update ETA, arrange pickup, submit documents, process payment, notify the customer, or simply archive it.
• Interruptions destroyed workflow continuity. An operator working one shipment could be pulled into several unrelated shipments within minutes.
• Important emails were not necessarily the newest emails, so chronological inbox order was a poor proxy for operational priority.

What Twenty built

• Automatic classification. Every incoming message was classified into freight-specific categories such as arrival notice, booking confirmation, schedule change, invoice, customs documentation, detention, POD, pickup request, or exception.
• Shipment resolution. Twenty extracted references from the email and attachments and matched them against bookings, shipments, containers, bills of lading, customers, and counterparties.
• Action extraction. Instead of merely summarizing the email, the system identified the operational implication, such as update shipment, arrange pickup, submit documentation, process payment, or notify stakeholders. Different freight email classes imply very different operator actions.
• Priority and exception routing. Routine, high-confidence messages were separated from ambiguous, urgent, or potentially high-impact events.
• Review-ready changes. Extracted data was presented alongside the existing shipment record and proposed changes, allowing operators to approve, reject, edit, or hold rather than re-key information.
• Auditability. The original email remained linked to the proposed and committed change, preserving the source behind every AI-assisted action.

Operational result

The biggest improvement was not simply faster email processing. It was lower cognitive load.

Operators no longer began every email with four questions: What is this? Which shipment is this? What changed? What do I need to do? Twenty answered those questions before the operator opened the work item.

The inbox shifted from being a stream of unstructured messages into a prioritized operational review queue. Operators could focus their attention on exceptions, ambiguous matches, customer communication, and decisions requiring judgment rather than continuously reconstructing context.

The result: fewer inbox interruptions, less searching across systems, faster action on important events, and substantially more operator capacity without increasing headcount.