Freight has always run on paperwork. Commercial invoices, packing lists, bills of lading, customs entries. A growing share of it is now read, sorted and checked by software before a person looks at it.
For a business shipping across borders, that raises a practical question about the global logistics provider it hires: which parts of the work can AI tools handle well, and which parts still depend on people and on the shipper’s own data? The answer changes from shipment to shipment, but a few patterns hold fairly steady.
Key points
- AI tools in freight are often used for reading documents, estimating arrival times and flagging exceptions.
- Customs classification still depends on clear, accurate product descriptions from the shipper.
- Appointing an agent doesn’t automatically move responsibility for customs declarations away from the business.
- The useful questions for a provider are about how its tools reach their outputs and who reviews them.
Where AI Shows Up in Day-to-Day Freight Work
Most of the visible change is in document handling. Some document-processing systems can extract shipper names, weights, item descriptions and values from scanned invoices and packing lists, then check them against a booking. That used to be manual data entry, the kind of repetitive task where small errors creep in.
Tracking is the other big area. Some visibility platforms use carrier updates, port information or historical transit times to estimate arrival dates, then send an alert when a shipment drifts off schedule. For a finance or operations team, an early warning about a delay is often more useful than a perfect forecast.
Then there’s quoting. In some systems, collecting rates from several carriers, comparing routes and building a quote can be partly automated, which mostly saves time on routine lanes.
Adoption is uneven, though. A January 2026 survey by BCG and Alpega, covering more than 180 logistics providers and shippers, found that only about one in ten logistics providers had embedded AI into core operations at scale. BCG’s published findings list documentation handling, predictive arrival times and exception management among the use cases in play. And these tools work from the data they’re given, which can be messy.
Product Data Is Still the Weak Link
Customs is where messy data costs the most. Goods crossing borders are classified using the Harmonized System, which the World Customs Organization describes as a product classification built on six-digit codes and used by more than 200 countries and economies. According to the WCO, over 98% of merchandise in international trade is classified under it. Many countries then add digits of their own for tariff and statistical purposes.
Some AI classification tools suggest a code based on product descriptions and other product data. A contributed article in WCO News on AI-assisted customs classification describes these outputs as recommendations that a classification expert still validates or corrects. In other words, a suggested code isn’t a customs ruling.
It also puts a lot of weight on how the item is described. “Plastic parts” tells a model, and a customs officer, almost nothing. A description that covers material, function and use gives the suggestion something to work with.
This is where the master data conversation turns into a logistics conversation. If product records inside a business are incomplete or inconsistent, the same gaps travel straight into shipping documents, no matter how capable the provider’s software is.
What to Ask a Provider About Its Digital Tools
Plenty of forwarders now mention AI in their sales material, and the label alone doesn’t say much. Before choosing a global logistics provider, businesses should ask how its digital tools support shipment visibility, documentation and exception handling, and which decisions receive human review.
A few questions tend to get useful answers:
- Which tasks are automated, and which are still reviewed by a licensed broker or trade specialist?
- How are suggested tariff classifications checked, and can you see the reasoning or rulings behind them?
- What shipment data will you be able to see yourself, and how often does it update?
- What happens when the system flags an exception at 2 a.m. on a Friday?
- Where is shipment and customer data stored, and who can access it?
How a provider answers says a lot. One that can explain the limits of its tools plainly tends to be easier to work with than one suggesting software has taken the risk out of shipping.
Responsibility Doesn’t Move to the Software
Here’s the part that sometimes gets lost. Hiring an agent, or relying on an agent’s AI tools, doesn’t automatically hand over responsibility for what’s declared at the border.
UK guidance is a clear example. HMRC’s page on getting someone to deal with customs for you says a business will still be responsible for due diligence on its customs declarations after appointing someone. Rules differ between countries, and liability can depend on how the agent is appointed, so it’s worth checking the arrangement in each market you ship to.
In practice, that means keeping your own records, reviewing classifications for new products, and not treating a fast automated entry as a checked one.
Where People Still Make the Calls
Some freight doesn’t fit a template. Oversized project cargo, shipments split across several transport modes, goods that need special permits, or a container held for inspection all involve judgment that’s hard to automate.
Negotiation is another one. When capacity tightens or a carrier rolls a booking to a later sailing, someone has to pick up the phone and find an alternative. Software can show the options, but closing the deal usually still takes a person.
And claims, disputes and audits still tend to come down to people reading documents carefully and explaining decisions. If an automated suggestion was wrong, someone has to own the fix.
A Sensible Way to Start
For a business reviewing its logistics setup, a small test usually says more than a sales demo. Pick one regular lane, compare the automated outputs against what your team would have filed, and track how often corrections are needed.
It also helps to tidy product data before switching providers or tools. Clear descriptions, materials, country of origin and any existing classifications make a system more accurate, whether a person or a model is reading them.
AI is making freight paperwork faster and shipment status easier to see. Decisions about what’s being shipped, how it’s described and who answers for it still sit with the business and the people it works with.
