AI for Claims Handling in Freight Forwarding

The email lands on a Friday afternoon. A warehouse manager at your customer has photographed a pallet leaning badly, shrink wrap torn, cartons crushed at the base. The reference in the subject line belongs to their buyer, not to you. The receipt was signed earlier in the week, and under the signature the driver wrote a few words nobody has read since.

You did not carry that pallet. You sold a door to door service, then bought the ocean leg from a carrier and the haulage from a subcontractor. Your customer holds your trading conditions. You hold terms you did not draft. The claim against you and the claim you bring are different claims, with different limits, evidence and deadlines, and the gap between them is money you keep or lose.

Very little of the work in between is judgement. It is identification, retrieval, arithmetic and diarising.

Matching the complaint to a job file nobody quoted

The sender quotes their own reference, or a container number, or nothing beyond the shipment from Shenzhen last month. An agent reads the mailbox, recognises that a message describes loss or damage rather than asking for a status, and resolves whatever identifier it finds against your records: equipment numbers, house and master bill numbers, air waybill numbers, or consignee and delivery date. It opens a claim against the matched file and extracts the delivery date, the parties and the complaint. Where it cannot match with confidence it offers candidates rather than guessing.

When the sender is not your customer

Often the person complaining is not the person who pays you. A consignee writes in directly, or an overseas agent forwards a note from a receiver. The claim still has to come from whoever carried title and risk under your conditions, so an agent should flag when the complainant is not the contracting party.

Protective notice goes out before anyone knows what happened

Every regime gives a short window to put the carrier on notice, and the shortest are measured in days from delivery or from handover itself. Air, road, rail and sea all differ, concealed damage is treated separately again, and your trading conditions give your customer a window that lines up with none of them.

Liability can be plain, the evidence complete, the carrier willing, and the recovery still gone because the file spent a fortnight being investigated rather than protected. An agent calculates both windows from the delivery date as the record opens, taking the governing regime from the transport document rather than from habit, and drafts the holding notice straight away. It commits you to nothing and preserves everything.

A bundle built from a warehouse phone and a photocopied packing list

The file a carrier will accept is a specific set of documents, and in forwarding they sit across the booking system, the origin office, the warehouse and somebody's handset. An agent assembles it: the bill of lading or air waybill with the terms on its reverse, the delivery receipt and any annotation, the packing list, the commercial invoice supporting value, stuffing and arrival photographs, logger output where the traffic carries one, and the correspondence from both ends of the chain. It indexes that against the carrier's requirements and names what is missing.

Files stall when a logger is reused before anyone downloads it, or the warehouse clears photographs off a shared phone, so the ask goes out on day one.

Reading the driver's biro, then showing the image beside it

The annotation on the receipt usually decides the argument. A clean signature makes your position against the carrier much harder. An exception naming cartons, counts or damage type is close to decisive. A scrawl about the load looking rough is worth whatever its reader wants. An agent transcribes the handwriting from the scan, records its confidence, classifies the note as specific or generic, and puts the transcription next to the original image. It does not interpret, and it makes reading the original cheap, because a confident misreading of a driver's biro is the most expensive small error here. An image of a crushed carton is an index to evidence, never evidence itself.

Damage found after the receipt was signed clean

Concealed damage is the hard case. The receipt was signed without exception, the pallets went into racking, and the damage surfaced when a carton was opened days later. Your customer is often genuinely right and your evidential position is still weak. An agent treats this as its own category: confirm what the receipt said, establish when the damage was first recorded, pull the seal records and logger output, and apply the rule the governing regime sets for damage not apparent at delivery, usually a different window from visible damage. Then state plainly what you can prove.

The limit is calculated on weight, not on what the goods cost

Customers assume a claim is worth the value of the goods. Carrier liability rarely works that way. The conventions cap recovery by reference to the weight of what was lost or damaged, or to a unit of the consignment, in an accounting unit that has to be converted. A light, valuable consignment can be capped far below invoice value. A heavy, cheap one may be capped above it.

An agent builds the working: weights affected against weights shipped, value substantiated from the commercial invoice, freight and duty separated, salvage where goods remain saleable, downstream loss identified as such, then the cap shown beside the claimed amount. Tedious arithmetic, which is why it happens only on the large files.

The mode changed, and so did the limit

One movement can cross several regimes. Air, road and ocean legs are governed differently, and a multimodal bill of lading may apply one set of terms throughout or whichever regime governs the leg where the loss occurred. An agent should name the regime it applied and cite the document that put it there, so a reviewer can argue with the reasoning, not only the number.

You are defending one claim and bringing another

The claim against you runs on your trading conditions, your caps and your time bars. The claim you bring runs on the convention and the carrier's terms. The two can give different answers about the same pallet. An agent should hold both positions in one record: what you owe, and what you can recover and from whom. It should surface early the files where you are liable for more than you will ever recover, because that conversation belongs before anyone writes a number down.

Cargo insurance and carrier liability are not the same product

The customer declined cargo insurance, assumed your liability covered the goods, and discovers at claim stage that carrier liability is capped by weight and excludes a long list of causes. An agent checks at intake whether the shipment carried a policy placed through you, and whether cover was offered and declined. Where cover exists, the claim goes to the insurer with the bundle already built, and the carrier recovery runs alongside. Where it was declined, the correspondence showing the offer belongs in the file immediately.

General average, and a demand from an average adjuster

On ocean cargo an unrelated casualty produces a demand against your customer's goods. General average is declared, security is required before release, and salvage may follow. None of it is damage to the consignment, but it lands on the same desk with its own deadlines. An agent should identify every shipment of yours on the affected vessel and notify those customers and their insurers. Uninsured cargo means a customer posting a bond to get their own goods back, which nobody wants to discuss late.

When the complaint is about a late sailing and nothing is broken

Much of what the claims inbox receives is not cargo loss. The box missed its sailing, the airfreight went on a later flight, the delivery arrived after a retail window closed. The tone is identical and the substance is not: no physical loss, usually no carrier liability for delay, and an exposure that is contractual rather than conventional. An agent should split these at intake and route them into service failure handling with the booking history, the cut off evidence and the rollover notice. Mixing them is how real cargo claims queue behind an argument about a missed vessel while a notice window quietly expires.

What the carrier's claims desk does with your recovery file

Once you have settled with your customer you are the claimant, dealing with a department whose job is to apply the cap, question the evidence and wait. An agent prepares the submission in the carrier's format with proof of payment to your customer, the notice served, the custody timeline and the limit calculation. Then the unglamorous part: chasing silence, tracking the limitation period that ends the right to sue, and escalating as it approaches so pursuing or writing off is a decision rather than a drift. Where your insurer has paid and taken the claim over by subrogation, it builds the same pack for them.

Where AI fails here

Admitting or denying liability cannot be delegated. Anything leaving in your name binds you, and a sentence of sympathy to an angry customer reads later as an admission. The agent drafts, a named person releases, and the system enforces that rather than the policy.

Settling is judgement. Paying a claim you do not owe to keep an account, or paying above what you can recover because the relationship is worth more than the file, is a commercial decision an agent has no basis to make. It should present the evidence, the quantum, the cap and the recovery prospects, then record which decision was taken, so a commercial settlement is never logged as admitted liability and quoted back at you next quarter.

Pursuing a carrier is human too. It depends on how much capacity you buy from them and whether a thin recovery is worth the relationship.

Then the quiet failure: a summary that misread the driver's note, or applied the wrong regime because the bill of lading terms were on a reverse nobody scanned. It will read beautifully and be wrong. Keep the source beside every extraction and make checking cheaper than trusting.

Where to start

Start with intake and the notice diary. Both are high volume, both are easy to check against what your team did, and both fail safely. A wrongly matched claim record costs a minute. A notice served after the window costs the whole recovery.

Run it in parallel on real files long enough to meet the awkward ones. Leave limit calculations, liability positions and anything customer facing until file assembly, the biggest time cost, has earned trust.

If you want a view on which parts of your claims desk an agent should own and which must stay with a named human, describe the process to Adopt AI Partners at hello@adoptaipartners.com.