Comparison
Do you need AI shipment tracking if carriers do it?
Usually not. Carriers already publish tracking for free. What is worth automating is reconciling feeds, handling deviations and the paperwork.
By the Foxnut team · Updated
Do you need AI shipment tracking if the carrier already tracks it?
You usually do not need AI shipment tracking just to see where a consignment is, because the visibility most buyers are shopping for already exists, is produced by somebody else, and costs nothing. Every carrier, forwarder and parcel operator runs a tracking portal, so automatic package tracking is already there for the asking. The container lines have gone further and published a common Track and Trace API through their own association, so the same events can be read the same way from any of them.
We are Foxnut AI, a senior-only practice in Bengaluru and Paris that builds AI systems and hands them over, so we earn when you decide to build. There’s a link to talk to us at the end of this article. Let’s first work out what is left to automate once the carrier has done its part.
In Europe the obligation runs deeper than customer service. Under the Union Customs Code, goods brought into the customs territory must be covered by an entry summary declaration lodged before they arrive, and the party who lodges it is the carrier. The company holding the movement data holds it because the law requires it to. That is why no downstream system will ever know more about a consignment than the carrier already does.
So this page is narrower than the category it sits in, and we would rather say that plainly. If the question is where a consignment is, a party under a legal duty to answer it answers for free, and no automation project should be sold against it.
What remains is three smaller questions: how to reconcile several feeds that disagree into one record a business can act on, what happens when a promised date moves, and how to produce the paperwork that has to travel with the goods. Those are worth ruling on, and the ruling is what the studio’s position on systems you have to keep running exists to make legible before anyone commits to a build.
What are your options for tracking shipments?
You have four arrangements to choose from. The choice is not between tracking and not tracking. It is about where the reconciliation happens and who is left holding it.
| Axis | The carrier’s own portal and alerts | A multi-carrier tracking aggregator | A visibility module in the TMS or ERP | A built integration against the carriers’ APIs |
|---|---|---|---|---|
| Where the data comes from | One carrier at a time, in that carrier’s own vocabulary | Many carriers, normalised by the aggregator into its own vocabulary | Whatever the vendor has connected, inside the system that holds the order | The carriers directly, plus the standard where a carrier implements it |
| What it can answer | Where this one consignment is, authoritatively | Where all consignments are, in one list | Where this order is, joined to the order that caused it | Whatever the business can define, joined to whatever it already holds |
| What it cannot answer | Anything about the other carriers, or about the order behind the consignment | Why a date moved, or what the business should now do about it | Anything about a carrier the vendor has not connected | Nothing structural, which is the trap: the scope has no natural edge |
| What breaks it | Nothing, and that is the point | A carrier changing its feed, absorbed silently as stale or missing events | The connector roadmap, which belongs to the vendor and not to the business | Every carrier-side change, forever, and each one arrives without notice |
| What it costs to keep running | Somebody’s browser tabs, and the time to open them | A subscription, and trust in a normalisation nobody outside can inspect | A licence line, and an escalation path through the vendor | An owned integration, its exception rules, and a person who reads them |
The last row is the one that gets skipped. An integration built against several carriers is not a project that finishes. It is a maintained connection to systems belonging to other companies, and it costs a permanent, small, unglamorous obligation to keep reading and repairing feeds.
That obligation is what the buyer is actually taking on. It is why the fourth arrangement is right less often than it is chosen.
What decides whether shipment tracking is worth automating?
Six structural dimensions decide whether a process is worth automating. Shipment tracking reads unusually on three of them: its input is produced by a third party and not observed, its exceptions are built into the data format and not discovered, and its regulatory exposure sits entirely on the documents and not on the tracking.
| Dimension | What it reads on shipment tracking | Source |
|---|---|---|
| Exception variance | The exception is not a share of consignments; it is a property of the data format. In the container industry's own standard every transport and equipment event carries a classifier saying whether it is an actual, a planned or an estimated event, and only document-status events are required to be actual. A carrier may also publish an event that retracts an event it sent earlier, and the standard gives that retraction its own field. Alongside these sit a three-character delay reason code, whose valid values are deferred to a list published by a separate industry body, and a free-text change remark whose documented example is 'Bad weather'. Read together: a large part of what arrives is provisional by design, revision is expected rather than exceptional, and the reason a date moved arrives as a code plus prose rather than as a fact a system can act on. | DCSA EVENT_DOMAIN 3.1.0; DCSA_DOMAIN 3.0.0 |
| Volume | Not consignments. The carrier already tracks every one of them at no marginal cost to the shipper, so counting shipments measures the carrier's work and not the buyer's. What scales is the number of moments a person has to decide something: an estimate that moved past a threshold somebody cares about, a document that entered a state needing action, a promise that has to be remade to a customer. The standard's own vocabulary shows the shape - two transport event types against seventeen equipment event types and seventeen document-status codes, so most of the traffic in a tracking feed is paperwork changing state rather than goods changing place. The payback condition is therefore a condition rather than a figure: the automation pays when the deviations needing a human decision outrun the hours available to notice them, and that count, not the shipment count, is the number to measure first. | DCSA EVENT_DOMAIN 3.1.0 |
| Cost of an error | Split, and the two halves differ by orders of magnitude. A wrong arrival estimate costs a wasted dock slot, an unnecessary expedite or an apology, all absorbed inside the week. A wrong declaration is a legal act: lodging one renders the person concerned responsible for the accuracy and completeness of the information and for the authenticity, accuracy and validity of every supporting document, and a representative lodging on someone else's behalf is bound by the same obligation. The exposure also moves without warning. Goods valued at 800 dollars or less arriving in the United States other than by post could previously be released against a manifest; from 24 June 2026 they must go through formal or informal entry, a change applied to a flow the same rule records at 1.36 billion consignments in Fiscal Year 2024 against 139 million in 2015. | Regulation (EU) 952/2013, Article 15(2); CBP final rule, 91 FR 37789 |
| Reversibility | The tracking record is freely revisable and the customs record is not, which is the reverse of what buyers expect from a system that feels like paperwork. Retraction is a first-class feature of the tracking standard. A declaration is the opposite: amendment is permitted on application, but not once the authority has said it intends to examine the goods, not once it has established the particulars are wrong, and not once the goods have been released or presented. Two of those three cut-offs are acts of the authority rather than of the declarant, so the window closes on somebody else's decision and without notice. An unlodged entry summary declaration is invalidated automatically 200 days after lodging where the goods never arrive. The connection between the two records is the sharp part: the tracking feed itself carries a Customs Selected for Inspection event, so the same stream a business is automating is the stream that announces its ability to correct the filing has just ended. | Regulation (EU) 952/2013, Articles 129, 173 and 174; DCSA EVENT_DOMAIN 3.1.0 |
| Regulatory exposure | Nil on the tracking and heavy on the documents, with nothing in between. No rule requires a business to know where its own goods are, and nobody audits an arrival estimate. Everything riding with the consignment is regulated: the entry summary declaration lodged before arrival, the customs declaration, and in Europe an electronic-freight regime that obliges authorities to accept regulatory information presented electronically, from a certified platform, in machine-readable form over an authenticated connection, with a human-readable view produced on the operator's own screen when an officer asks. The acceptance obligation begins 30 months after the first implementing or delegated act takes effect, and the implementing act was adopted on 5 July 2024 and published on 20 December 2024. The design consequence is that the tracking half can be built casually and the document half cannot. | Regulation (EU) 2020/1056, Articles 4 and 5; Regulation (EU) 952/2013, Article 127; Implementing Regulation (EU) 2024/1942 |
| Vendor market maturity | Standardising, from a low base, and the evidence is the industry's own conduct rather than a market share. No verified public measure of concentration in shipment visibility software was retrieved, so none is quoted here. What is documented is that the container lines built a common Track and Trace API through their association, and that the association's own statement of the problem is that current methods for generating and communicating operational data are antiquated, manual, unaligned and unpredictable, with data often exchanged inconsistently, with delays or not at all. The standard also records its own limit: the poll model is mandatory to implement, the push subscription model is optional, and a compliant implementation may answer a subscription request with 501 Not Implemented. On the document side the European regulation legislates a certified-platform market into existence rather than describing one that already worked. Both facts say the same thing, which is that the format is arriving before the market is settled. | DCSA Track & Trace standard and API v3.0.0-Beta-1; Regulation (EU) 2020/1056 |
Two of those rows carry the argument. The volume row says the thing being counted is usually wrong. A business that describes its scale in consignments per month is describing the carrier’s workload, and the number that decides the purchase is how many times a week somebody has to stop and make a decision about one.
The exception row says the input is not a stream of facts. A tracking feed is a mixture of estimates, plans, confirmed events and withdrawals of earlier events. Any system reading it has to hold that distinction explicitly, or it will present a forecast to a customer as a promise.
The reversibility row is the quiet one, and it sets the whole design. The tracking record can be corrected freely and the customs record cannot, so the two halves need opposite treatment inside the same project.
On the tracking side, an automated action should generally wait for a confirmed actual event and be cheap to undo. On the document side, the automation should stop short of the filing and put the last step in front of a person, because the window in which a mistake can be fixed is closed by an authority acting on its own timetable.
What is a delivery exception in shipment tracking?
A delivery exception is something you derive yourself, because the published standard has no exception event. Its vocabulary has two transport event types, arrived and departed, and seventeen equipment event types covering loading, discharge, gate movements, stuffing, stripping, pick-up, drop-off, inspection, resealing, removal and the three customs states. None of them means “this went wrong”.
An exception is therefore always worked out and never received. It is an expected event that has not arrived, an estimate that has moved beyond a tolerance somebody defined, or an event that was published and then retracted.
This matters to anyone scoping the work. A carrier cannot supply the exception logic and a standard cannot hand it over, because it is a set of business decisions about which deviations matter, to whom, and within what time. What counts as late for a promotional launch is not what counts as late for a replenishment order, and neither is a property of the shipment.
The reason vocabulary makes the same point from the other side. The standard carries a short delay reason code whose values are maintained by a separate industry body, plus a free-text remark. A system can route on the code, but the sentence explaining what actually happened arrives as prose written by whoever keyed it.
Nobody searches for this. The searches under delivery exceptions returned nothing at all in every question shape we tested, which usually means the work is real but has no name buyers use for it. It shows up instead as a person opening portals each morning and writing emails, and that kind of work is easy to automate and easy to over-buy.
What is customs documentation, and can software file it?
Customs documentation is the paperwork that travels with the goods, and it is the half of this process where automation has real standing and also has to stop before the end. The tracking standard’s own list of document types is a useful map: booking, shipping instruction, transport document, delivery order, container release order, arrival notice, verified gross mass, customs clearance, dangerous goods declaration, and the certificates of origin, health, phytosanitary, veterinary, fumigation, analysis and inspection. Every one of them is a structured artefact assembled from data the business already holds, and software is good at assembling them.
What the software should not do is press send. Under the Union Customs Code the person lodging a declaration is responsible for the accuracy and completeness of what it says and for the validity of the documents behind it, and a representative lodging on their behalf carries the same obligation.
Amendment is available on application, and stops at three defined events. Correcting after release is possible only within three years, and only to bring the declarant into compliance with the obligations of the procedure the goods went under. So the cost of a wrong filing is not the effort of redoing it. The ability to redo it can end at a moment chosen by somebody else.
Two developments make this a live design question. In the United States, low-value consignments that used to clear against a manifest now require formal or informal entry, applied to a flow the rule itself records as having grown roughly tenfold in under a decade.
In Europe, electronic presentation is becoming the expected form and not a convenience. Authorities will be obliged to accept regulatory information from a certified platform in machine-readable format, and to be shown a human-readable view on the operator’s own device on request. Both push in the same direction: more data, assembled earlier, in a stricter form. Neither changes who is liable for it.
Important note: this page will not tell anyone how goods should be classified, what duty is owed, or which procedure a consignment belongs under. Those are the questions most searches in this space are asking. They are regulated advice, and a software studio has no standing to answer them.
What do buyers get wrong about AI freight tracking?
The first error is buying visibility that already exists. The pitch for this category is a map with dots on it, and the map is nearly always available for free from the party that produced the dots. A business that cannot see its shipments usually has an access problem or an attention problem and not a software problem, and both are cheaper to fix than an integration.
The second error is treating an estimate as an event. A feed that mixes planned, estimated and actual events into one timeline is easy to render as a tidy status and hard to render honestly. The temptation is to show the latest value with no indication of what kind of value it is, and that is how a forecast reaches a customer as a commitment.
The standard makes the distinction machine-readable so it can be preserved. A system that flattens it has thrown away the one piece of information that decides whether an action is safe.
The third error is scoping the integration by carrier count and not by change rate. Connecting to five carriers is a bounded piece of work. Staying connected to five carriers is not, because each of them will change something without asking, and the failure arrives quietly as events that stop appearing and not as an error somebody sees.
The published standard helps where a carrier implements it, and its own text says implementations may differ. The polling model is mandatory but the push model is optional, so a system designed around subscriptions has to fall back to polling for anyone who declined to build it.
The fourth error is automating across the customs boundary because the two halves look like one workflow. They are one workflow to the operator and two different risk regimes to the business: one record is designed to be corrected and the other is designed to be closed.
There is also a prior question this page does not answer, which is whether the purchase orders, expected dates and consignment references are complete and current enough to reconcile the tracking data against. That is a readiness question with its own answer. A tracking system built over an order record nobody trusts will produce confident, well-presented, wrong arrival dates.
Should you automate shipment tracking?
Automate the reconciliation and the paperwork, and not the visibility. Visibility is already supplied by the party legally obliged to supply it, and rebuilding it is the most commonly bought and least useful part of this category.
The work that pays is joining several carrier feeds to the orders that caused them, so the business has one record and not five portals. It is also defining which deviations deserve a human’s attention, and assembling the documents that travel with the goods, up to but not including the moment they are filed.
The reason is structural and is no comment on how good the available products are. The volume row says the number that decides the purchase is deviations needing a decision, not consignments, and almost nobody has measured it. The exception row says the value of any system here is the discipline with which it keeps estimates, plans, confirmed facts and retractions apart. The reversibility row says the two halves need opposite designs, because a tracking record can be corrected at leisure while a filing stops being correctable at a moment an authority chooses.
The useful first move is a count and not a demonstration. Over four weeks, record every time somebody had to open a carrier portal, chase a status, re-promise a date to a customer, or key the same consignment detail into a second document.
That produces three numbers nobody currently holds: how many deviations a week actually need a decision, how many carriers and portals the answer is spread across, and how much of the paperwork is retyping data the business already stores. Those numbers decide whether anything here is worth building, and in a good number of businesses the honest answer will be better use of the portal that is already there.
We build AI systems and hand them over, and that is why we are direct about this. The carrier’s own tracking is not a competitor to be worked around. It is the source, and a proposal that treats it as a gap to be filled has misread the problem.
At Foxnut AI the people who sign the proposal do the work, and there is one tier. We designed, manufactured and launched a physical product of our own, so a manufacturer going dark and stock running short are things we have paid for and not read about. We would rather tell you what your count shows than sell you the rebuild.
The first step is a free expert audit: a senior practitioner reviews your setup, with no RFP and no pitch deck. You can read about how Foxnut Consulting takes on this work first.
If your four-week count turns up real deviation volume worth building for, tell us what you counted. If it turns up mostly a portal nobody opened, that is a cheaper and better answer, and we will say so.
The part most pages leave out
When not to choose Foxnut AI
Situations where another option is the better call, and where we say so in the first conversation rather than the fourth.
- The only question anyone in the business actually asks is where a consignment is, and the carrier's portal, the forwarder's portal or a free multi-carrier lookup already answers it in a few seconds. That portal, and we say so. The party holding the data is under a legal obligation to produce it, publishes it at no marginal cost, and knows things no downstream system can reconstruct. Paying to rebuild an answer that already exists is the wrong purchase.
- You have few enough lanes, carriers and consignments that one person holds the whole picture in a browser and a spreadsheet, and deviations arrive at a rate that person can absorb inside a normal working day. The person already doing it, unchanged. What decides payback here is not shipment count. It is the number of deviations somebody has to notice, chase and record. Where that number fits in a morning, an integration adds a system to maintain and removes nothing.
- On this lane the estimate moves so often that nearly every consignment ends up an exception, so an exception queue would contain most of the traffic and not just the tail of it. A redesigned process, not an automated one. A different carrier, a different incoterm, a different port pair or an honest buffer changes the arrival distribution. Routing the same unreliability faster produces a queue nobody can clear and a team that stops reading it.
- The payoff you want is an automatic downstream action fired from a tracking event, such as releasing a replacement consignment, crediting a customer or closing a purchase order the moment a status changes. A human control point, or a simpler deterministic rule that acts only on a confirmed actual event and never on an estimate. The published standard lets a carrier withdraw an event it has already sent, so an action taken on a signal that is later retracted has spent real money on something that turned out not to have happened.
- The output being automated is a filing to an authority and not a note to a colleague: an entry summary declaration, a customs declaration, or regulatory information handed to an inspecting officer at the roadside. The incumbent certified vendor, whose attestation is the product. The person who lodges the filing carries the liability for its accuracy and completeness, the representative who lodges on their behalf is bound by the same obligation, and the European electronic-freight regime requires a certified platform and not merely a working one. Certification is the thing being bought, and it cannot be built.
Sources
- Regulation (EU) No 952/2013 laying down the Union Customs Code, Official Journal 10 October 2013 - Article 15(2) on responsibility for the accuracy and completeness of a lodged declaration and the binding of the customs representative, Article 127(3) and (4) on lodging an entry summary declaration before the goods are brought into the customs territory and on the carrier being the party who lodges it, Article 129 on the three events that end amendment of an entry summary declaration and on invalidation 200 days after lodging, Article 173 on the three events that end amendment of a customs declaration and the three-year post-release window, and Article 174 on the two grounds for invalidation. Read in the Official Journal full text on EUR-LexRetrieved
- Regulation (EU) 2020/1056 on electronic freight transport information, Official Journal 31 July 2020 - Article 4 requiring regulatory information to be made available from a certified eFTI platform in machine-readable format over an authenticated connection with a unique electronic identifying link, Article 5(1) obliging competent authorities to accept it as from 30 months after the entry into force of the first of the acts adopted under Articles 7 and 8, and Article 18 on application. Read in the Official Journal full text on EUR-LexRetrieved
- Commission Implementing Regulation (EU) 2024/1942 of 5 July 2024 laying down common procedures and detailed rules for accessing and processing electronic freight transport information by competent authorities, Official Journal 20 December 2024, adopted under Article 8 of Regulation (EU) 2020/1056 - read for its adoption and publication dates and its entry-into-force clause. Read in the Official Journal full text on EUR-LexRetrieved
- Digital Container Shipping Association, 'DCSA OpenAPI specification for Track & Trace', version 3.0.0-Beta-1, Apache 2.0 - the poll and push models, the statement that the push subscription model is optional to implement and may answer 501 Not Implemented, and the worked response examples including a transport event that retracts a previously published eventRetrieved
- Digital Container Shipping Association, EVENT_DOMAIN schema version 3.1.0 - the eventClassifierCode enumerations (ACT, PLN, EST for transport and equipment events; ACT only for shipment events), retractedEventID, the two transport event type codes, the seventeen equipment event type codes including Customs Selected for Scan, Customs Selected for Inspection and Customs Released, the seventeen shipment event type codes, the changeRemark and reason free-text fields, and the documentTypeCode listRetrieved
- Digital Container Shipping Association, DCSA_DOMAIN schema version 3.0.0 - delayReasonCode, a three-character field whose valid values are deferred to a code list published separately by SMDGRetrieved
- Digital Container Shipping Association, 'Track & Trace standard' - the association's own statement of the problem the standard exists to fix, that current methods for generating and communicating operational data are antiquated, manual, unaligned and unpredictable, and that data is often exchanged inconsistently, with delays or not at allRetrieved
- US Customs and Border Protection, 'Indefinite Suspension of the De Minimis Exemption for Merchandise Arriving Through All Modes Other Than the International Postal Network', final rule, 91 FR 37789, published and effective 24 June 2026 - the requirement that all entries valued at 800 dollars or less arriving other than by post use formal or informal entry procedures, the description of the superseded release-from-manifest process, and the recorded volumes of 1.36 billion de minimis shipments in Fiscal Year 2024 against 139 million in 2015. Read in the Federal Register full textRetrieved
Foxnut AI takes on work like this from Bengaluru and Paris. If you want to see how we do it before you talk to anyone, here is what an AI engagement covers and what you keep.