Skip to content

AI data market

Logistics exception AI data market: the event feed is not the asset

By SourceX Editorial · Reviewed by Noah Loul ·

8 sections

Short answer

The logistics exception AI data market is worth backing only where records join a disruption to the recovery decision and what it cost. Tracking events alone are not scarce: in our tally of 116 event and status codes in two open logistics standards, 38 describe exceptions or responses and none records a charge, claim or credit. Rate freight claims and detention disputes first, because rules already force dated decisions there.

Value drivers

Logistics exception AI data

This page covers logistics exception AI data market: the event feed is not the asset.Illustrative

Key takeaways

  • Tracking events are standardizing: 38 of 116 codes in two open logistics standards describe an exception or a response, and none records a charge, claim, credit or liability.
  • The investable asset is the record that joins a disruption, the recovery choice someone made and what that choice cost.
  • Motor carrier loss and damage claims and ocean detention disputes come with decision clocks set by rule, which makes their outcomes gradable.
  • U.S. vessel movements are published free and aggregate freight flows are pooled, so the size of a shipment archive overstates licensable supply.
  • Ask any supplier, SourceX included, for decision-linkage and cost-linkage ratios instead of shipment counts.

What would an investor be buying in logistics exception data?

A record of what a company decided when a shipment went wrong, and what the decision cost, not the tracking feed that announced the problem. The logistics exception AI data market is only as deep as the share of records that join three layers: the event that signaled a disruption, the recovery choice someone made, and the outcome in time and money.

Agent developers want to automate exactly that middle step. An exception agent has to decide whether to rebook a container onto the next sailing, answer a damage claim, contest a detention invoice, reattempt a delivery or quarantine a load that ran warm. Each choice has alternatives, each alternative has a price, and each has a date by which it worked or failed. A model can be trained or graded on the choice only when the record keeps the alternatives in view and ties the choice to its result.

So the first diligence question is not how many shipments a company moved. It is how many of its exception records reach the third layer. This article tests that with an original count of two open event standards, a map of the U.S. rules that already force a dated decision, and a six-type scorecard naming the evidence gap that would most change each rating.

Do open tracking standards already record the disruption?

Yes, and that is a problem for anyone pricing tracking events as scarce. Open logistics standards now give the disruption itself a code, and one gives the carrier's response a code too, but neither of the two we counted has a code for what the recovery cost.

We read the code lists in two public repositories on October 27, 2026. The Digital Container Shipping Association's open-source Track and Trace reference implementation (commit af619ff, August 2023) defines 42 transport, equipment, shipment and port-call status codes. By our rule, 18 name an exception or a response: customs selection for scan or inspection and customs release, seal inspection, removal and resealing, documents rejected, held, released, voided or cancelled, and all seven port-call status codes, which cover an omitted port, a blank sailing, an ad hoc call, a vessel phasing in or out, a rotation change and a sliding call.

GS1's EPCIS 2.0 schemas and Core Business Vocabulary (commit 6f71468, September 2026) list 41 business steps and 33 dispositions. Twenty name an exception or a response: dispositions such as damaged, stolen, expired, recalled, returned, non-conformant and three kinds of mismatch, and business steps such as holding, inspecting, repairing, replacing, destroying and voiding a shipment.

Across all 116 codes, 38 describe an exception or a response and 0 refer to a charge, fee, claim, credit, refund, liability or payment. The standards also differ in what they make measurable. Every DCSA event must carry an event time and a classifier marking it planned, estimated, actual or requested, so the gap between a promised and an actual arrival can be computed for any port call. The SMDG delay reason code that would explain the gap is optional; the reference schema lets it stay empty.

The read for an investor: the event layer is standardizing, which makes it cheaper to share and harder to sell as scarce. Value sits in records that attach a recovery choice and its cost to those events. The figure shows the split, and the table below summarizes the count:

Do open tracking standards already record the disruption?
Standard and sourceCodes countedException or responseCost or liabilityWhat it makes measurable
DCSA Track and Trace reference implementation, commit af619ff (2023): transport, equipment and shipment events35110An event time and a planned, estimated or actual classifier on every event; delay reason optional
DCSA reference implementation: port-call status codes770The carrier's schedule response: omitted port, blank sailing, rotation change
GS1 EPCIS 2.0 and Core Business Vocabulary, commit 6f71468 (2026): business steps and dispositions74200The state of the goods, handling steps and sensor alarm flags on measurements
Both standards116380Where and when a disruption happened, not who paid to recover from it
Two stacked horizontal bars counting codes in open logistics standards: DCSA Track and Trace has 42 codes, 24 routine and 18 for exceptions or responses; GS1 EPCIS 2.0 has 74, 54 routine and 20 for exceptions or responses; neither has a code for cost or liability.
Codes in two open logistics standards by what they describe. Source: DCSA-TNT (2023) and GS1 EPCIS (2026) repositories, SourceX count, October 2026.

Which logistics exceptions already come with a decision clock?

Two of the clearest cases in U.S. freight are motor carrier loss and damage claims and ocean detention and demurrage billing. In both, a rule makes someone put a decision in writing by a deadline, which gives a dataset an outcome label with a date attached, the hardest part of an evaluation set to rebuild after the fact.

Under 49 CFR 370.9, a motor carrier that receives a written claim for loss, damage, injury or delay to property must pay it, decline it or make a firm compromise offer in writing within 120 days. If it cannot, it must tell the claimant in writing why, repeat that notice every 60 days while the claim stays open and keep a copy in the claim file. Every resolved claim therefore ends in one of three labeled outcomes, and every slow one leaves a dated trail explaining the delay.

For ocean freight, the Federal Maritime Commission's billing rule, most of which took effect on May 28, 2024, requires carriers and terminal operators to issue demurrage and detention invoices within 30 calendar days of when the charge was last incurred. The billed party gets at least 30 days to ask for mitigation, a refund or a waiver, and the billing party must try to resolve a timely request within 30 days. A dispute under that rule has a start, a request and a deadline built in. The section listing what an invoice must contain was still awaiting Office of Management and Budget approval when the FMC announced the rule.

Nothing comparable governs a forwarder's choice to rebook cargo after a blank sailing, a dispatcher's reroute or a warehouse's decision to accept a load that ran warm. Those choices may be the ones agent developers want most, and they are the least likely to be written down in a form anyone can grade.

A six-type scorecard for logistics exception data

Freight claims and detention disputes score best because rules date their outcomes and we found no public dataset that substitutes for them; ocean schedule disruption scores worst on scarcity because vessel movements are already public. Rate each exception type on how firmly its outcome is recorded, what public data could replace it and how many parties have a stake in the record, then name the gap that would most change the rating.

The ratings are editorial judgments built from the sources in this article as of October 27, 2026. They rate exception types, not companies, and use no market size, volume or price.

A six-type scorecard for logistics exception data
Exception type and buyer taskOutcome labelPublic substituteRights and preparation burdenEvidence gap to close
Motor carrier loss and damage claims: pay, decline or offer a compromiseStrong: written disposition within 120 days or dated status letters (49 CFR 370.9)Low: no public claim-level data foundModerate: shipper, consignee and carrier documents, photos and inspection notesShare of claim files linking the disposition to the inspection evidence and the amount paid
Ocean detention and demurrage disputes: uphold, reduce or waive a chargeStrong: invoice, request and resolution windows set by the FMC rule since 2024Low: no public dispute-level data foundHigh: carrier, terminal, forwarder and shipper each hold part of the fileShare of disputes with a recorded outcome, reason and linked container availability events
Ocean schedule disruption: rebook, reroute or re-promise after a blank sailing or omitted portMedium: planned, estimated and actual times are standard; the shipper's choice is notHigh: U.S. vessel traffic points are published free; port-call status codes are openModerate: booking and customer records sit with forwarders and shippersWhether the rebooking decision and its cost were recorded beside the schedule events
Customs holds and inspections: prepare, clear and predict holdsMedium: selection and release events are standard codes, but the codes carry no causeLow: none foundHigh: filings and broker records may carry their own restrictionsHold cause and document defect linked to release time
Last-mile failed delivery: reattempt, redirect or returnWeak: GS1's vocabulary has a returned state, but no rule requires a reason or a costLow: none foundHigh: consumer addresses, delivery notes and photos are personal dataLink from the failed attempt to the final outcome and its cost
Cold-chain temperature excursions: accept, quarantine, reject or destroyMedium: sensor reports can flag an alarm; dispositions record damaged or destroyedLow: none foundModerate: shipper, carrier, receiver and insurer recordsLink from the excursion to the acceptance decision and any claim

Does a bigger shipment archive mean more licensable supply?

No. Archive size overstates supply in logistics for three reasons: much of the event stream is public or pooled, most records belong to several companies at once, and most volume is routine handling that carries no decision.

Vessel movement is the clearest case. Through MarineCadastre.gov, the Bureau of Ocean Energy Management, NOAA and the Coast Guard Navigation Center publish broadcast vessel traffic points for U.S. coastal waters from 2009 through 2026, drawn from the Coast Guard's AIS receiver network. A supplier whose pitch rests on vessel positions or port arrival times is competing with free data.

Aggregate flow is being pooled as well. Under the Department of Transportation's Freight Logistics Optimization Works program, participants contribute supply chain data that the Bureau of Transportation Statistics anonymizes and aggregates under federal statistical confidentiality rules. Members get a market view, but the view is aggregate by design, which leaves shipment-level decisions with the companies that made them.

Shared ownership is the third filter. One exception record can involve a shipper, a consignee, a carrier, a forwarder or broker and a third-party logistics provider, and the contracts among them may limit what any single party can license. Routine volume dilutes whatever is left: in both standards we counted, most codes describe normal handling, and a large archive of routine scans adds little to a decision dataset.

What could make the logistics exception thesis wrong

Five developments could shrink the value of logistics exception records. Name them before the upside; each has a signal an investor can follow.

  • Standards reach the money. If DCSA or GS1 add codes for charges, claims or credits, the decision layer standardizes the way the event layer has, and records that join the two become easier to assemble without licensing.
  • Large carriers and shippers build their own agents. A company that holds both the events and the decisions can train in house and never license, which leaves outside supply thin however good the files are.
  • Pools move toward shipments. FLOW is aggregate by design today; a shift toward shipment-level sharing would compete directly with licensed decision records.
  • Rules narrow. The FMC's invoice-content section was still awaiting approval when the rule was announced, and a weaker rule would thin the documentation that makes detention disputes gradable.
  • Rights fragment the supply. When shipper contracts limit how a provider may use shipment records, the most decision-rich files can be the hardest to license.

Diligence questions for logistics data suppliers, SourceX included

Ask for decision and cost linkage ratios rather than shipment counts, and set the definitions yourself so every supplier answers the same question.

SourceX belongs on the same list. It turns operating records that a company chooses to license into AI datasets and keeps every stage in house, from finding and qualifying suppliers through rights checks and contracts to preparing and delivering the data and paying the supplier. Nothing moves to SourceX until the supplier has approved the scope and terms and signed. SourceX does not publish supply, pipeline or deal figures; it gives company and commercial metrics to investors directly. Its current raise, under Rule 506(c), is open only to accredited investors, and materials can be requested at sourcex.si/investors.

  • Decision linkage: what share of exception records name the recovery action taken, and the alternatives considered, in a field rather than only in email?
  • Cost linkage: what share connect the exception to a charge, a claim payment, a credit or a write-off, and from which system does that figure come?
  • Outcome clocks: for freight claims, what share carry a written disposition and its date; for detention disputes, the request date and the resolution date?
  • Plan versus actual: for schedule exceptions, what share keep the planned, estimated and actual times for the same port call?
  • Party map: which companies appear in each record, and which contracts allow the supplier to license their share of it?
  • Routine share: how much of the offered volume is normal handling with no exception at all?

Methods note: the code count and the exception scorecard

The code count read the constants of four Java enums in the DCSA-TNT repository (commit af619ff, August 4, 2023): TransportEventTypeCode, EquipmentEventTypeCode, ShipmentEventTypeCode and PortCallStatusCode. It also read the enum arrays of the bizStep and disposition schemas in GS1's EPCIS repository (commit 6f71468, September 4, 2026). DCSA's event classifiers and EPCIS error reasons were reported separately because they describe the record, not the cargo or the voyage.

A code counted as an exception or response when it names a departure from the planned flow (damage, loss, mismatch, hold, rejection, customs selection, a changed or skipped port call) or a step taken because of one (inspect, release, reseal, repair, replace, destroy, void, cancel), including the dispositions disposed and non-sellable other; the 38 codes are listed by hand in the script. A code counted as cost or liability when its name, or the description the enum gives, contains charge, fee, claim, credit, refund, compensation, liability, liable, payment, invoice or cost as a whole word. The counting script needs nothing beyond Python's standard library and goes to reviewers with this article.

The scorecard rates exception types on the criteria above, using 49 CFR 370.9, the FMC's billing rule, MarineCadastre.gov, the FLOW program's FAQ and the two repositories. The DCSA repository is a reference implementation last changed in 2023, so newer published versions of the standard may list more codes. Not measured: how many companies hold linked exception records, buyer demand by exception type, preparation cost, prices, or the share of real claims and disputes with complete files. No SourceX supply, pipeline or deal figure is used.

Frequently asked questions

What is a logistics exception record in AI training data?

It documents a shipment that departed from plan, such as a delay, damage, shortage, hold or failed delivery, together with what the company decided and how the case ended. The useful version for training or evaluation links the triggering event, the recovery choice and the outcome in time and cost, so a model can be graded on the decision rather than on the event.

Is container tracking data valuable for AI training?

Rarely on its own. Open standard codes already describe arrivals, departures, holds and omitted ports, and U.S. vessel traffic points are published free through MarineCadastre.gov. Tracking data gains value when it is joined to the decisions and costs that followed a disruption.

Why are freight claims useful as evaluation data?

Because the outcome is labeled by rule. Under 49 CFR 370.9 a motor carrier must pay, decline or make a firm compromise offer in writing within 120 days of a written claim, or explain the delay in writing every 60 days. Each claim therefore ends with a dated outcome that a model's recommendation can be compared against.

Does the FMC detention and demurrage rule create useful records?

It creates dated dispute records: invoices within 30 days of the last charge, at least 30 days for the billed party to request mitigation, a refund or a waiver, and 30 days for the billing party to try to resolve the request. Those windows make outcomes and response times measurable, although the invoice-content section was awaiting approval when the rule was announced.

Does SourceX publish logistics supply or deal figures?

No. SourceX shares company and commercial metrics with investors directly instead of publishing them, and any offering it makes is limited to verified accredited investors. Investors can request materials at sourcex.si/investors and ask how the supplier ratios in this article would be defined.

Sources

  • The DCSA Track and Trace reference implementation defines transport, equipment, shipment and port-call status codes, requires an event time and a planned, estimated, actual or requested classifier on every event, and allows the delay reason code to be empty. Source
  • GS1's EPCIS 2.0 schemas list the Core Business Vocabulary business steps and dispositions, including damaged, stolen, expired, holding, repairing and destroying, and let sensor reports flag an alarm condition. Source
  • Under 49 CFR 370.9 a motor carrier must pay, decline or make a firm compromise offer in writing within 120 days of a written claim, or give written status notices every 60 days and keep copies in the claim file. Source
  • The FMC's detention and demurrage billing rule, mostly effective May 28, 2024, sets 30-day windows for invoicing, for mitigation or refund requests and for resolving them; its invoice-content section awaited OMB approval. Source
  • MarineCadastre.gov publishes broadcast vessel traffic point data for U.S. coastal waters for 2009 through 2026, prepared by BOEM, NOAA and the Coast Guard Navigation Center. Source
  • The DOT's FLOW program aggregates and anonymizes participant data, operated by the Bureau of Transportation Statistics under CIPSEA confidentiality rules. Source

Request SourceX investor materials

SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.

See if your company qualifies