Opportunity landscape

Where AI can help.

Logistics runs on plans that start to go wrong as soon as they are made. Carrier delays, port congestion, weather and demand swings change what is possible. AI helps where teams must predict, compare options and reconcile documents faster than they can by hand.

Priority applicationWhat it improves
Route and capacity optimisationHow loads, vehicles and capacity are assigned, and how plans adjust as conditions change.
Shipment exception managementWhich shipments are at risk, and who acts, before a commitment is missed.
Document and freight reconciliationWhether invoices, customs entries and delivery documents match what was agreed and shipped.

These are examples of work that can be scoped. The right starting point depends on your data, systems and priorities.

Priority applications

What we can build, in detail.

01 / 03

Route and capacity optimisation

Business impact

Improves how loads, vehicles and capacity are assigned, reducing cost and distance while meeting service commitments.

Suitable situations

Fleets or networks with many stops, time windows or lanes, where plans are built by hand or by rules that ignore cost trade-offs.

Full application detail
What we build
An optimisation service that produces routes or load plans from orders and constraints, a comparison of options, and re-planning when conditions change.
Required inputs
Orders and stops with locations and time windows, vehicle and driver constraints, travel times, costs and service rules, plus historical routes.
Technical approach
Constrained optimisation with heuristics for scale, travel-time and service-time estimation from history, and evaluation against current plans on past days.
Workflow integration
Plans flow to dispatch and driver applications. Planners review and adjust, and changes are logged to improve the rules.
Human responsibility
Dispatchers own the final plan and exceptions. Operations leaders own service and cost trade-offs.
Measurement
Distance and cost per stop, on-time performance, vehicle utilisation and planner time, against a baseline period.
Limitations
Plans are only as good as the location, time and constraint data. Driver acceptance and regulatory limits shape what can change.
Adjacent opportunities
Shipment exception management, which uses the same status data.

Next action

Take the assessment for a first view of where this fits, or contact us to discuss a specific case.

02 / 03

Shipment exception management

Business impact

Identifies shipments likely to miss a commitment early enough to act, and directs the right person to the right action.

Suitable situations

Networks that depend on several carriers or modes, where status arrives in different formats and delays are found by customers first.

Full application detail
What we build
Status ingestion from carriers and systems, risk and arrival-time prediction, a prioritised exception queue, and drafted customer and carrier messages for approval.
Required inputs
Carrier and tracking events, schedules, historical transit times by lane and carrier, orders and commitments, and contact rules.
Technical approach
Harmonising status data, transit-time prediction compared with published schedules, risk scoring, and tracking of prediction accuracy by lane.
Workflow integration
Exceptions open as tasks for the responsible team. Messages are drafted and sent after approval, and outcomes are recorded.
Human responsibility
Customer and operations teams own the response. Approvals apply to credits, rerouting and customer commitments.
Measurement
Accuracy of arrival predictions against actuals by lane, share of exceptions caught before the customer, time to respond and missed commitments.
Limitations
Prediction depends on timely carrier data, and disruptions can break historical patterns, so estimates need buffers and frequent refresh.
Adjacent opportunities
Route and capacity optimisation. Document reconciliation.

Next action

Take the assessment for a first view of where this fits, or contact us to discuss a specific case.

03 / 03

Document and freight reconciliation

Business impact

Checks invoices, customs entries and delivery documents against contracts and shipment records, catching errors before payment or filing.

Suitable situations

Businesses handling high volumes of freight invoices, bills of lading, customs declarations or proof-of-delivery documents.

Full application detail
What we build
Document extraction, matching to rates, contracts and shipment records, an exceptions queue with the evidence for each mismatch, and audits of past records.
Required inputs
Invoices, rate agreements, bills of lading, customs entries and delivery documents, with shipment and order records to match against.
Technical approach
Document parsing and OCR, entity resolution across documents, rule and model checks against contracts and tariffs, and sampling to measure accuracy.
Workflow integration
Matched items proceed. Mismatches go to a reviewer with the documents side by side, and accepted corrections update the record.
Human responsibility
Finance and compliance own approvals and filings. Reviewers resolve exceptions. Customs responsibilities stay with the licensed party.
Measurement
Match rate, extraction accuracy, error rate on filings or payments, dispute outcomes and review time.
Limitations
Poor scans and inconsistent formats lower extraction accuracy. Rules differ by country and contract and must be maintained.
Adjacent opportunities
Exception management, using document status as a signal.

Next action

Take the assessment for a first view of where this fits, or contact us to discuss a specific case.

Published deployments

What others have reported.

These are results published by other organisations, shown as their sources report them, with what each figure measures. They are not TJC Labs results and are not a forecast of what a project will deliver. A prospect’s own business case is estimated from its own data.

Freight forwarding, global

Flexport

Company-reported
Problem
Carrier schedule reliability in 2024 ranged from about 50 to 55%, making plans built on announced transit times unreliable.
What was implemented
Machine learning models that predict ocean transit times, reviewed by experts, with added buffers and more frequent updates during the Red Sea disruption.
Technical approach
Models are trained on historical transit times and carrier-advertised transit times, with fixed adjustments for inland origin and destination.
Reported result
As of August 2024, Flexport’s predictions on the Transpacific Eastbound lane achieved 75% on-time performance, against 53% market schedule reliability for the same period (Sea Intelligence).
Scope, baseline and period
One lane only. The comparison is Flexport’s own data against the market figure, and the methodology and sample size are not given.

Reported by: Company-reported. Source: Flexport (opens in a new tab), 24 October 2024.

Customs brokerage, United States

Flexport

Company-reported
Problem
Mistakes and compliance errors in customs entries.
What was implemented
An AI auditor, including an agent that audits past customs entries for mistakes and compliance errors, built on a data layer that exposes container shipping network data to its agents through MCP.
Technical approach
The source gives no further technical detail.
Reported result
A 0.2% error rate on US customs filings after piloting the agent on Flexport’s own entries. The source gives no prior baseline.
Scope, baseline and period
Pilot on Flexport’s own entries. A separate claim of a tenfold improvement over other brokers is an estimate by the speaker and is not used here.

Reported by: Company-reported (founder and CEO, quoted). Source: Supply & Demand Chain Executive (opens in a new tab), 27 February 2026.

Parcel delivery, United States

UPS

Customer-reported
Problem
Giving each driver an optimised sequence for the day’s pickups and deliveries. Early route optimisation methods worked in the lab but were hard to put into practice.
What was implemented
ORION, which generates an optimised route each day for each of 55,000 US drivers, built on UPS’s Package Flow Technology. It went through about three years of field testing, and UPS changed its management practices so that users and executives would accept it.
Technical approach
Route optimisation algorithms from UPS’s operations research group using a meta-heuristic, keeping routes consistent from day to day.
Reported result
More than $320 million saved as of December 2015, and savings of $300 to $400 million a year expected at full deployment. About 100,000 metric tons of CO2 reduced a year.
Scope, baseline and period
More than 35,000 of 55,000 US drivers were using ORION as of December 2015. Savings are estimates and are in addition to those attributed to earlier systems.

Reported by: Customer-reported; peer-reviewed in Interfaces; INFORMS summary. Source: INFORMS (opens in a new tab) · Interfaces abstract (opens in a new tab), Interfaces 47(1), 2017; INFORMS page undated.

More published examples

Freight brokerage, United States

Uber Freight

A new carrier-side pricing algorithm replaced a Markov decision process with truncated backward induction and a tree-based estimate of continuation cost. A combined A/B and switchback test showed a statistically significant reduction in the total cost to cover loads. The post gives no figure.

Reported by: Company-reported. Source: Uber Freight (opens in a new tab), 8 December 2023.

E-commerce logistics, India

Flipkart

Flipkart’s supply chain work includes capacity planning and real-time capacity orchestration. The authors report that capacity planning cycle time fell by 90%, and that two-day deliveries for high-priority orders rose by 20%. See the Consumer & Retail page for the full case.

Reported by: Customer-reported; peer-reviewed. Source: INFORMS Journal on Applied Analytics (opens in a new tab), 2026.

Technical implementation

What shapes the build.

  • Status data from many sources. Carrier, port and system events arrive in different formats and need harmonising before prediction.
  • Prediction and uncertainty. Arrival estimates carry ranges, are checked against actuals by lane and refresh as events arrive.
  • Optimisation at scale. Large routing and allocation problems use decomposition and heuristics, validated against current plans.
  • Document extraction. Scan quality and format variety are measured, and low-confidence items go to a reviewer.
  • Event-driven integration. Webhooks, queues and idempotent writes keep tasks and records consistent when systems are slow or unavailable.

Readiness and data requirements

What helps you start.

  • Shipment, order and status data from your TMS, WMS or carriers, with identifiers that join across them.
  • Historical routes, transit times and delivery outcomes, ideally by lane and carrier.
  • Rate agreements, contracts and tariffs in a form that can be checked against invoices.
  • A clear owner for exceptions and the rules for who may approve changes.
  • Access to carrier APIs or data feeds, or an agreed way to receive status.

Common questions

Before you ask.

Do we need carrier APIs?

Status can come from carrier APIs or data feeds, or from an agreed way of receiving it. Event-driven integration keeps tasks and records consistent when systems are slow or unavailable.

How accurate are arrival predictions?

We measure predictions against actual arrivals by lane and track them over time. Disruptions can break historical patterns, so estimates need buffers and frequent refresh. We do not promise a figure in advance.

Who stays responsible for customs filings?

Customs responsibilities stay with the licensed party. The system checks entries and routes mismatches to a reviewer.

Next step

What could AI mean for your business?

Take our free Enterprise AI Adoption Assessment to explore where AI could make a difference. Get two scores, a recommended next move and a full report by email in about three and a half minutes.