What changes logistics
Sixty-five points, sorted by two questions: when does it bite, and how hard? Measured against a mid-sized forwarder — not against the industry.
When it bites, how hard it hits
Nine fields of time and impact. The number says how many points sit in each; the link leads to them.
What applies when
Nine dates, in order. Three already apply, six are coming — and build the connection only for the deadline and you will not have it on the day.
-
Emissions to one standard
Anyone stating transport emissions computes them to EN ISO 14083 — measured data before estimates. Forty-eight months to comply.
Open the CO₂e calculator → -
Customs and tachograph
Every consignment from a third country is dutiable. Light commercial vehicles above 2.5 tonnes need the second-generation tachograph in cross-border work.
Short term → -
AI Act, Article 50
Anyone talking to a system has to be told. The transparency duty applies whatever risk class the model sits in.
Liability → -
NISG 2026
A duty to report incidents; transport is named explicitly in the annex. Nobody writes that documentation retrospectively.
Attack surface → -
Electronic freight information
Authorities accept it — but only what arrives through a certified platform. A connection does not appear in the June before.
Regulation → -
Carbon border adjustment
First surrender of certificates for the 2026 import year, with the supplier’s verified emissions data.
Short term → -
AI Act, Annex III
The obligations for high-risk applications. Postponed is not withdrawn — and the documentation is written during operation, not after it.
Liability → -
Emissions trading for transport
A year later than planned; national schemes run on until then. The surcharge arrives in the buying price, not at your own pump.
Capacity → -
EU Customs Data Hub
Capture customs data once, use it many times. The date hangs on the ongoing negotiations and is the least certain in this row.
Long term →
What is changing
Sixty-five points, each in one sentence — ordered by when they bite and how hard. Nine are my own position, the rest come from published sources.
Short term
Bites now to 2028. Wait here and someone else decides for you.
-
High impactShort term
No hit rate, no autonomy
The jump from “the model suggests” to “the model acts” rarely fails on the model. It fails because nobody knows the hit rate.
my position
read on
The open point: A hit rate can neither be estimated nor taken over from a vendor. You measure it in shadow operation: the procedure runs alongside, decides nothing, and every recommendation is held against what was actually done. That costs weeks in which it returns nothing — and it is the only route at the end of which someone can take responsibility for the release. What you measure with has recently become a legal question too: the Data Act (EU) 2023/2854 has applied since 12 September 2025 and expressly covers connected products, vehicle telematics included. Whoever uses the data from their own devices has a claim to it rather than a request. How wide the gap runs shows in two figures from the same week: Gartner expects the agent market to jump from under 2 to 53 billion dollars by 2030 — and MIT’s NANDA project measures roughly 95 per cent of AI pilots with no measurable return, RAND over 80 per cent failure. Both are true at once, and the difference is the question of who built the process around the model.
From practice: The company’s own trailers deliver data continuously — GPS, cameras in the loading space, condition readings — and that is the basis against which a recommendation can be checked at all. Measured and calculated stay separate fields: an estimated position never belongs in the same field as a measured one, even where the display shows both alike.
What would move it: When the first hit rate comes out of shadow operation and somebody signs it off.
In the industry radar: tracked for the first time as a trend of its own, high impact, under five years — explainability as a precondition, not an add-on.
-
High impactShort term
The seventy percent nobody needs to look at
The consignment note arrives as an image; a human reads and releases it. Money in both directions hangs on that release.
my position
read on
The open point: The hurdle is not the model but handwriting and sloppily completed consignment notes. Per hundred incoming documents: around twenty are handwritten or carry a handwritten note, around seven are captured so badly that nothing reliable can be read from them, two to three carry damage — rounded experience from live operation, not an analysis with a sample and a cut-off date. Whoever promises “the document is read automatically” promises the hundred and delivers the seventy.
From practice: Then the seventy it is, and explicitly so: not cracking the hard thirty — handwriting cannot be read with the certainty a release needs when money hangs on it in both directions — but recognising the clean ones reliably and letting them run through. Everything else goes visibly to a person instead of silently passing. As of today a person checks after the upload; the model that reads the document itself is being worked on. What stands here is the goal, not the state.
What would move it: When the clean documents run through for a week without rework — measured on the inbox, not on a demonstration.
In the industry radar: covered under image processing and generative AI in the inner ring, both aimed at reading — that recognition fails on handwriting and not on the model is not said there.
What has to be on a CMR – open the checker →Where documents become the bottleneck →
-
High impactShort term
Workflows decide for themselves
Assistants make preliminary decisions where a person checks today — starting with recurring cases.
all three sources
read on
What it means here: The step is small and the consequence large: whoever hands over a pre-decision must afterwards be able to say why it came out that way. For a forwarder that does not mean “the agent buys”, but: the assistant prepares, the dispatcher releases — and both are logged.
Source: all three sources
-
High impactShort term
Explainability before autonomy
Traceability becomes a precondition, not an add-on — listed as a trend in its own right for the first time.
industry radar
read on
What it means here: Traceability is not an ethics question here but an operating condition: whoever quotes a customer a price has to be able to explain it — even when a model proposed it. Without that line every automation stays stuck in suggestion mode.
Source: industry radar
-
High impactShort term
From pilot to regular operation
95 percent of the companies surveyed run AI initiatives, about 20 percent take one past the pilot — and only 44 percent measure what it returns at all.
executive survey
read on
What it means here: The gap between ninety-five and twenty per cent is not a technical question. It appears where nobody defined beforehand what success is and who signs it off — and where the pilot ran on an extract instead of on the daily business.
Source: executive survey
-
High impactShort term
Why pilots fail
Not on the model. More than half name data quality and access, 39 percent silos and unclear ownership.
executive survey
read on
What it means here: Data quality, access, ownership: three questions that must be answered before the first model. In a forwarder they read: who owns the master record — and which figure counts when two systems deliver different ones?
Source: executive survey
-
High impactShort term
Regulation delivers the data everything else waits for
eCMR, e-invoicing, digital freight contracts: what arrives as an obligation produces machine-readable records of the whole case as a side effect.
my position
read on
The open point: The deadline is read more sharply than it is written: from 9 July 2027 authorities in the EU must accept freight information provided electronically — Regulation (EU) 2020/1056. That is an obligation on the authority, not on the company. Anyone who nevertheless makes it their own cut-off date is right: only what comes via a certified platform is accepted, and a connection does not appear in June 2027.
From practice: The CMR already arrives through an upload portal today and is therefore on hand before a deadline demands it. Money hangs on its release in both directions: the credit note to the sub-carrier and payment by the customer. Whoever gets this document structured early saves not paper but a recognition step.
What would move it: When a customer demands the certified platform — that will be long before 9 July 2027.
In the industry radar: not tracked as a driver of its own — the obligation appears as a framework condition, not as the thing that supplies the data.
What has to be on a CMR – open the checker →Where the obligation becomes the lever →
-
High impactShort term
End to end digital, no paper
Documents, identities and payments without a break in the medium.
all three sources
read on
What it means here: The break is rarely in your own house; it sits at the ramp and at the subcontractor. A document that arrives as an image is digital and still not usable — end-to-end means structured, not paperless.
Source: all three sources
-
High impactShort term
Access to your own data
Data structures that allow fast access become business-critical — the department has to reach its own numbers.
key trends
read on
What it means here: The bottleneck is rarely the database but the question of who may build an analysis. Where every figure goes through IT, spreadsheets grow in the dark — and in the end they are the truth nobody can check.
Source: key trends
-
High impactShort term
One reliable source instead of many truths
Carrier, port, warehouse and haulier report the same event differently. Only a reconciled record can carry a decision.
megatrend study
read on
What it means here: For a forwarder this is not a data project but a claims question: when customer, carrier and your own system carry three different arrival times, it is not the correct one that decides but the one shown first.
Source: megatrend study
-
High impactShort term
From tracking to prediction
Tracking says where something is. The next stage says when it arrives and which route holds instead.
megatrend study
read on
What it means here: The difference is the phone call: “the shipment is stuck” is information, “the shipment will be two hours late and the consignee has been told” is a service. The second needs an estimate somebody stands behind.
Source: megatrend study
-
High impactShort term
When the forgery itself comes out of a model
The same tools that read a consignment note also write one. A single false order moves an entire load.
my position
read on
The open point: The familiar tells fall away: broken language, skewed logos, an imprint that does not add up. What remains is checking against data rather than against impressions — and it has to sit before the award. After that the goods are gone and the paper trail looks clean. On top of that comes a deadline closer than most here: from 1 October 2026 Austria’s NISG 2026 applies, implementing Directive (EU) 2022/2555 (NIS2). The transport sector is in Annex 1; whoever it catches, it catches with incident reporting duties — that is, with exactly the documentation you cannot create retrospectively.
From practice: Vetting of new carriers runs against hard attributes: trade licence, proof of insurance, and binding via a definitive identifier rather than via a search term two companies can share. For AI the same applies here as everywhere on this site: checking and presenting yes, releasing is done by a person.
What would move it: When a forgery gets through that hard criteria do not catch.
In the industry radar: tracked as a trend of its own, high impact, under five years — logistics networks expressly named as a preferred target and AI-driven attacks as a class of their own.
Whether a document is unchanged – open the checker →Where the release stays with a person →
-
High impactShort term
Attacks come out of models
Fake hauliers, phishing, entire cargo thefts — harder to spot thanks to AI. Logistics networks count as a preferred target.
industry radar · freight exchange
read on
What it means here: The old tells are gone: broken language, crooked logos, an imprint that does not add up. What remains is checking against hard data — and it has to happen BEFORE the award, because afterwards the goods are gone and the paper trail looks clean.
Source: industry radar · freight exchange
-
High impactShort term
When the subcontractor lacks drivers, the buying price rises
For a forwarder the driver shortage is not a staffing problem but a price: scarce capacity shifts bargaining power to the haulier.
my position
read on
The open point: The age structure is the most convenient explanation and the least complete. Long-haul demands weeks away from home and is hard to reconcile with a family, waiting time at the ramp is rarely paid, and the licence costs money before it earns any. On top of that comes the war: drivers from Ukraine, Belarus and Russia are largely unavailable for European long-haul. None of these causes reverses because the economy picks up; whoever names only demography underestimates how long this lasts.
From practice: No calculator helps against the shortage — against paying blind it does. The should-cost calculator says before the negotiation what a trip may cost; the empty-running rate shows where capacity travels unused that nobody has to procure first. That is the honest lever: not more trucks but better utilisation of those that run. The answer that lowers demand is automated long-haul driving — and no forwarder builds that. As of today: in the ATLAS-L4 research project a driverless truck runs on the motorway under real conditions, and a multi-year trial is running in Spain. Series production towards the end of the decade — the moment this moves the price is further out than the headlines suggest.
What would move it: When the buying price rises without diesel, tolls or demand explaining it.
In the industry radar: tracked as a demographic trend, with automation as the answer — that a shortage at the subcontractor arrives as a buying price at the principal is not said there.
What a trip may cost – open the calculator →What travels empty – open the calculator →
-
High impactShort term
Drivers are running short
Around 420,000 truck drivers missing across Europe. Demographics is only one cause; waiting time at the ramp is rarely paid.
key trends · freight exchange
read on
What it means here: At a forwarder the shortage does not arrive as a staffing problem but as a price: the subcontractor finds nobody, withdraws from the lane, and the buying price rises — on a trip whose selling price was agreed long ago.
Source: key trends · freight exchange
-
High impactShort term
The seller's market takes hold
Scarce capacity keeps prices up, bargaining power shifts to the haulier, the price gap closes.
freight exchange · market report
read on
What it means here: The price gap closes from the wrong side: buying rises, selling hangs on annual contracts. Negotiate in this market without a target price and you notice the loss only in the monthly report.
Source: freight exchange · market report
-
High impactShort term
Freight awarded on platforms
High impact, under five years. That awarding happens is undisputed — whose price applies is not stated there.
industry radar
read on
What it means here: The platform is convenient and it shifts something: award there and you adopt its pricing logic and its view of the market. The question is not whether to take part but with which target price of your own you go in.
Source: industry radar
-
High impactShort term
Tariffs move flows, not only prices
US tariffs on European vehicles could cut exports by up to seven per cent. That hits whole lanes and the feeder runs to the North Sea ports, not single shipments.
market report
read on
What it means here: For transport capacity it is not the tariff rate that counts but the lane: if an export falls away the truck runs back empty — and the empty-run rate bites faster than any tariff.
Source: market report
-
High impactShort term
Proof instead of intent
Large customers demand verifiable records; without them you are not even considered in their tenders.
all three sources · customer magazine
read on
What it means here: The sentence is two years old and has turned: proof used to be an extra in the tender, today it is the ticket in. Deliver it per shipment and you get asked; deliver it as an annual report and you no longer do.
Source: all three sources · customer magazine
-
High impactShort term
Disruption is the normal case
In early 2026 storms, snow and ice halted road traffic across half of Europe and North America at once. Weather is now a fixed part of daily planning.
industry radar · key trends · market report
read on
What it means here: The exception has become a planning variable. For dispatch that does not mean “more buffer” but: the second route must be computed in advance, otherwise it gets improvised during the disruption — and improvised routes are the expensive ones.
Source: industry radar · key trends · market report
-
Moderate impactShort term
Analytics instead of reporting
From report to decision support. The bottleneck is not the model but data quality and access.
industry radar
read on
What it means here: The difference is the question in front. A report answers “how was it”, a decision aid “what do I do now” — and the second kind needs a figure somebody can act on, not one that is merely correct.
Source: industry radar
-
Moderate impactShort term
Machines reading images
Depth perception, 3D reconstruction, making sense of blurred originals.
industry radar
read on
What it means here: The technology is there, the documents are not: handwriting, crooked photos, stamps across the line. About twenty in a hundred carry a handwritten note — that is where any recognition ends that was not meant as a suggestion.
Source: industry radar
-
Moderate impactShort term
Code comes out of the model
Three stages worth telling apart: the prototype in hours, reviewed agent-built code, and assisted programming. Only the latter two hold up in production.
customer magazine
read on
What it means here: The prototype in hours impresses and carries nothing. What carries in production is the second case: reviewed code somebody owns. For a company without its own development that mostly means knowing the difference before a vendor blurs it.
Source: customer magazine
-
Moderate impactShort term
The obligation brings a new data source
Since 1 July 2026 light commercial vehicles above 2.5 tonnes in cross-border work also need the second-generation tachograph: border crossings via Galileo, loading and unloading, remote interrogation.
EU Mobility Package I
read on
What it means here: The tachograph is usually read as an obligation. It is also a source: border crossings, loading and unloading, all time-stamped and readable remotely. Anyone wanting to evidence waiting times has them — at the subcontractor, though, not in their own house.
Source: EU Mobility Package I
-
Moderate impactShort term
Email as the open flank
Anyone running processes on email is especially exposed: attachment and link are the usual way in.
freight exchange
read on
What it means here: In a forwarder almost everything runs on email: order, notification, invoice, claim. That is exactly why a single altered attachment is enough. The countermeasure is unspectacular — never take bank details from an email, always from the master record.
Source: freight exchange
-
Moderate impactShort term
Reporting duty, not goodwill
Austria’s NISG 2026 brings a duty to report incidents from 1 October 2026; transport is named explicitly in the annex. Nobody writes that documentation retrospectively.
NISG 2026 · Directive (EU) 2022/2555
read on
What it means here: A duty to report means: in an incident what counts is what was written down before. Start documenting then and you document a reconstruction. The effort is not in reporting but in keeping the records beforehand.
Source: NISG 2026 · Directive (EU) 2022/2555
-
Moderate impactShort term
Smaller hauliers give up
Staff, energy and running costs push thinly capitalised operators out. The result: less capacity, more concentration.
freight exchange
read on
What it means here: Every failure costs twice: the capacity and the knowledge of the lane. The successor drives it for the first time, and that shows in the first price. Buy broadly and you pay a little more all the time — and much less when one drops out.
Source: freight exchange
-
Moderate impactShort term
Rigid structures cost capacity
Fixed slots, inflexible ramps, no part loads. Efficiency comes from better organisation, not more trucks.
freight exchange
read on
What it means here: The missing capacity is often already there, running empty. Fixed time slots, inflexible ramps and separated orders prevent consolidation — that is the lever you have without a single extra truck.
Source: freight exchange
-
Moderate impactShort term
Volatile energy prices
Carbon charges and swinging power and gas prices call for variable contracts.
freight exchange
read on
What it means here: Diesel is one half, contracts are the other: sell without an escalation clause and you carry the swing yourself. For costing that means knowing the fuel share per lane instead of carrying it as a lump sum.
Source: freight exchange
-
Moderate impactShort term
The true price is more than the freight
Add carbon charges, insurance, the risk of an interruption and the revenue lost when goods arrive late.
megatrend study
read on
What it means here: The freight rate is the visible part. Add levies, insurance, the risk of a disruption and the revenue lost when goods arrive late. Compare only the freight and you award to whoever hides the other items best.
Source: megatrend study
-
Moderate impactShort term
Small consignments are no longer duty free
Since 1 July 2026 every consignment from a third country is dutiable; below 150 euros in value a flat three euros applies. The trigger was 4.6 billion such consignments in 2024.
EU customs reform
read on
What it means here: The amount is small, the volume is not: 4.6 billion such consignments in 2024. What changes is not the duty but the number of declarations — and that hits everyone doing first and last mile for e-commerce.
Source: EU customs reform
-
Moderate impactShort term
Importing now requires authorisation
The pricing phase of the carbon border adjustment has run since 1 January 2026: covered goods may only be imported by an authorised declarant — with a threshold of 50 tonnes a year for cement, iron and steel, aluminium and fertiliser.
Regulation (EU) 2023/956
read on
What it means here: For the forwarder this is not a climate question but a lead time: without an authorised declarant the goods stop. If you import for customers, know whether they hold the authorisation — otherwise the question comes at the terminal.
Source: Regulation (EU) 2023/956
-
Moderate impactShort term
Whoever states emissions follows one standard
Since 1 June 2026 only the whether is voluntary: anyone naming transport emissions — in a report, a contract, an advertisement — computes them to EN ISO 14083, measured data before estimates. Forty-eight months to comply.
Regulation (EU) 2026/1030
read on
What it means here: Only the whether is voluntary now. State a figure and it must be computed to the standard — measured data before estimates. That applies to the tender, the quote and the invoice alike.
Source: Regulation (EU) 2026/1030
-
Moderate impactShort term
Those who can prove it gain share
Especially where it comes without notable extra cost.
key trends
read on
What it means here: The advantage lies less in the value than in availability: deliver the figure per shipment within hours and you are on the short list — supply it a quarter later and you are long out. The effort is in the data chain, not the formula.
Source: key trends
-
Moderate impactShort term
Several routes instead of the cheapest
Broaden the supplier base and the routes, even where normal operation gets more expensive.
industry radar · key trends
read on
What it means here: Redundancy costs in normal operation and pays in a disruption — and because normal operation is more frequent, it is optimised away first. The honest comparison sets the running extra cost against the expected damage, not against zero.
Source: industry radar · key trends
-
Moderate impactShort term
Rail does not show up
Terminal bottlenecks, construction and the withdrawal of the EU directive slow combined transport; road stays dominant.
freight exchange
read on
What it means here: For the CO₂e balance rail would be the biggest lever, and it is the hardest to pull: terminal bottlenecks and construction turn a transit-time promise into a risk. Offer it anyway and you should have costed the fallback to road first.
Source: freight exchange
-
Low impactShort term
The market grows double-digit
Machine vision 19.6 percent a year, generative AI 47.5 percent to 2030 — figures from an industry radar, not from my own shop.
industry radar
read on
What it means here: A market figure says nothing about your own benefit — it says something about the vendors’ pace. For a mid-sized company double-digit growth mostly means: what is expensive and exotic today is in the standard package in two years.
Source: industry radar
Medium term
Bites 2028 to 2031. Prepare now, do not build now.
-
High impactMedium term
Interoperability instead of point-to-point
Every new partnership starts with an interface someone builds and maintains. The next step is not another one, but a shared language.
my position
read on
The open point: An open door is quickly built and hard to close again. Whoever opens one decides beforehand what is visible behind it and what is explicitly out of scope — binding awards, for example. Whoever waits inherits someone else’s boundary. And a protocol is not yet a shared language: MCP governs how an assistant calls a tool — not what a consignment note is. That second layer is being built elsewhere: the Open Logistics Foundation on open-source components including a European eCMR standard, the eFTI Regulation on certified platforms with a common data model, and the DCSA on container standards at sea. Whoever has only the protocol can call out and will not be understood.
From practice: This is already built. In the operation I write from, public calculators hang off an MCP server as tools and can be called by an outside assistant — loading metres, CO₂e, demurrage, cabotage. Queries yes, binding awards no; that boundary sits in the code. The calculators on this site come without that connection: they compute for readers, not for machines.
What would move it: When a customer makes the connection to their assistant a condition instead of a trial.
In the industry radar: not tracked — agent-to-agent does not appear there as a trend of its own.
What an agent may do and what it may not →Where the interface becomes the bottleneck →
-
High impactMedium term
Agents directing other agents
A layer grows above the single agent: a conductor that builds an entire fallback scenario when something breaks.
megatrend study
read on
What it means here: The conductor is the point where automation turns from busywork into decision — and where the question of release has to be asked a second time. A contingency nobody signed off is a disruption with extra cost.
Source: megatrend study
-
High impactMedium term
Let it negotiate — but on whose price?
That a model can bid is no longer the question. Which price applies is decided not by the model but by whoever sets the parameters.
my position
read on
The open point: An agent in the same house negotiates two things: what a carrier is paid and what a customer is charged. Between them lies the margin a forwarder lives on — and a competition-law boundary: two algorithms settling on a price level can amount to a concerted practice within the meaning of Art. 101 TFEU, even where nobody agreed anything.
From practice: The parameters are set in three places: the platform’s target price, your own calculation, a market value from someone else’s data. The should-cost calculator on this site shows openly how such a target price comes about; an agent would be given it, not left to interpret it.
What would move it: When a vendor ships an agent that closes bindingly and the competition-law question is settled.
In the industry radar: tracked as digital marketplaces, high impact, under five years — whose price applies is not said there.
What a trip may cost – open the calculator →Where a model decides and where it does not →
-
High impactMedium term
The carbon price on diesel arrives later
Emissions trading for transport and buildings starts in 2028 instead of 2027 — finally adopted in March 2026. National schemes run on until then; postponed is not cancelled.
EU climate package
read on
What it means here: Postponed is not cancelled, and the postponement is the actual news: it gives two years to compute the cost effect per lane instead of suffering it. Until then national schemes run on — so the surcharge is not zero.
Source: EU climate package
-
High impactMedium term
Sourcing is spread wider
South-East Asia, India, nearshoring: diversification is no longer an intention but under way — and it creates new pre- and on-carriage at other ports.
market report
read on
What it means here: Diversification sounds like sourcing and is a transport question: other origins mean other ports, other pre-carriage, other transit times. Fail to move the lanes with it and you sell a transit time the new route does not have.
Source: market report
-
High impactMedium term
The provider as architect of the chain
First rebuild the status quo, then compute scenarios where cost and lead time count together. It turns out some redundancy is cheaper than removing it.
customer magazine
read on
What it means here: This is the role a forwarder can hold when volume falls: the advantage is not the vehicles but the view of both sides — shipper and carrier. The precondition is being able to compute the current state before offering scenarios.
Source: customer magazine
-
High impactMedium term
Net zero becomes an obligation
A voluntary disclosure turns into a cost item that has to be evidenced per shipment.
megatrend study · customer magazine
read on
What it means here: A voluntary disclosure turns into a cost item — and it has to be evidenced per shipment, not per year. For costing that means: the proof belongs in the price, not in the sustainability report.
Source: megatrend study · customer magazine
-
Moderate impactMedium term
Who covers the shortfall when the agent books?
While an agent only suggests, the question is academic. The moment it books bindingly, it is an amount.
my position
read on
The open point: Liability stays with the operator; a model error is no excuse towards the customer. The AI Act (EU) 2024/1689 turns that into processes with documentation duties, and it has recently had two different clocks: the transparency obligations under Article 50 have applied since 2 August 2026 — anyone talking to a system must be told. The obligations for high-risk applications under Annex III, by contrast, have moved to 2 December 2027, and for Annex I products to 2 August 2028; they were postponed by the Digital Omnibus Regulation (EU) 2026/1744, in force since 27 July 2026. Anyone who reads calm into that postponement is confusing the two clocks.
From practice: Whoever retrofits the documentation is retrofitting for transactions that ran long ago — and that is exactly where what is needed is missing. Hence the reverse route: every figure that goes out carries its method and data source, a model value is marked as a model value, and anyone talking to an assistant learns so in its first sentence. The refusals belong to that: no agent awards bindingly — not because it could not technically, but because that responsibility cannot be delegated.
What would move it: When the first case is argued in which an agent placed a binding award.
In the industry radar: carried along under ethics, not as a question of its own — who is liable when an agent books bindingly is not taken apart there.
Bounded autonomy and its limits →What an agent may do and what it may not →
-
Moderate impactMedium term
Agents against paperwork
Documentation duties are not shrinking. The hope rests on agents that pull data from several sources and file it.
freight exchange
read on
What it means here: The hope is justified and the order is backwards: an agent that collects data from five sources first needs the five sources. Where they are missing it automates the retyping — and that costs more than the retyping.
Source: freight exchange
-
Moderate impactMedium term
Clerical work shifts
Where cases look alike, the machine takes over — flatter organisations, fewer intermediate levels.
key trends
read on
What it means here: What disappears is not the job but the uniform part: order entry, status queries, invoice checking. What remains is the exception — and it takes more experience than the routine, not less.
Source: key trends
-
Moderate impactMedium term
Connected assets and devices
Assets report their own condition instead of being polled.
industry radar
read on
What it means here: For a forwarder the interesting devices are not the plant but the trailers: temperature, door opening, standing time. The benefit only appears when those signals are attached to the order — otherwise it is a second dashboard.
Source: industry radar
-
Moderate impactMedium term
Digital twins
Full integration is years away — in handling it has arrived: overhead scanners identify, locate and measure every package without a hand touching it.
industry radar · customer magazine
read on
What it means here: The full twin is years away, the small version is useful today: your own network rebuilt once so scenarios can be computed instead of debated. The effort is in the rebuild, not in the computing.
Source: industry radar · customer magazine
-
Moderate impactMedium term
Not just defend — restart
If the system stops, the goods stop. The question shifts from “how do we keep it out” to “how fast do we come back up”.
megatrend study
read on
What it means here: The question shifts from “how do we keep it out” to “how long are we down”. For a forwarder that is a question about the paper trail: what still goes out without systems, and how long can operations run on shouted instructions?
Source: megatrend study
-
Moderate impactMedium term
The subcontractor cannot afford the electric truck
The bottleneck is not the vehicle but the connection: transformers with long lead times and some 800 grid operators in one country alone.
customer magazine
read on
What it means here: The bottleneck is not the vehicle but the grid connection: transformers with long lead times, many operators, uncertain dates. Demand electric runs from subcontractors and you demand an investment that only pays for them with a multi-year commitment.
Source: customer magazine
-
Moderate impactMedium term
Growth sits in the east
Poland, Bulgaria, the Baltics and the Balkans are forecast above two percent for 2026/27, the large western markets between zero and one.
market report
read on
What it means here: Where the market grows, so does the number of trips subject to rules: cabotage, posting, return obligations. A lane’s price advantage disappears quickly when an inspection finds one trip too many.
Source: market report
-
Moderate impactMedium term
The network is computed, not grown
A large groupage network has its own model determine the number, location and size of its terminals — the rebuilt status quo is the yardstick.
network report
read on
What it means here: The sentence is about large networks and the idea holds for small ones: number, location and size of hubs are a calculation, not a habit. The benchmark is the rebuilt current state — otherwise you compare a model with a feeling.
Source: network report
-
Moderate impactMedium term
Better planning creates fewer trips
Over thirty percent of food produced is never eaten. Better forecasts cut that loss — and the transports moving it.
research
read on
What it means here: Fewer trips sounds like less business and is both: the shipper saves, the carrier loses volume. Offer the saving yourself and you sell the planning with it — wait and you only get the volume cut.
Source: research
-
Moderate impactMedium term
Manufacturers sell direct
The path through retail gets shorter. That pushes stock forward, creates smaller shipments and turns returns into a process of their own.
megatrend study
read on
What it means here: The shorter route to the customer is a longer one for logistics: more small consignments, more addresses, more returns. That is not lost volume but a different shipment profile — and per-shipment costing decides whether it carries.
Source: megatrend study
-
Moderate impactMedium term
From cost item to the customer's selling point
No longer “can you deliver?” but “can you deliver reliably, transparently and sustainably?” — the manufacturer differentiates through its supply chain.
customer magazine · research
read on
What it means here: The customer sells his supply chain along with the product — and needs figures from his forwarder for that. Deliver them and you become part of the sales argument; do not, and you become the line item that gets tendered.
Source: customer magazine · research
-
Moderate impactMedium term
Networks are built for restart
The yardstick shifts from the cheapest chain to the one that survives a disruption.
megatrend study
read on
What it means here: The benchmark shifts from the cheapest chain to the one that survives a disruption. For buying that means not keeping the second supplier in reserve but running a share with them — a reserve that never drives is no reserve in an incident.
Source: megatrend study
-
Low impactMedium term
Speech processing
Spoken instructions instead of a keyboard — rated low impact by the source itself.
industry radar
read on
What it means here: Rated low-impact by the source itself, and that fits: in dispatch typing is rarely the bottleneck. It gets interesting at the ramp and in the cab, where hands are busy — and there it fails on noise and dialect.
Source: industry radar
-
Low impactMedium term
Own energy, different fuels
On-site generation, sustainable fuels, circular packaging — the highest impact in the radar's technology section.
industry radar · key trends
read on
What it means here: For a company without its own fleet this is mostly a procurement question: the subcontractor’s fuel sits in their costing, not in ours. What remains is your own site — and its share of the balance is small.
Source: industry radar · key trends
Long term
After 2031. Keep it in view, keep the effort low.
-
High impactLong term
Autonomous driving
Held back in Europe by approval law, long since running on private sites. It is the only answer to the driver shortage that lowers demand.
industry radar · freight exchange
read on
What it means here: It is the only answer to the driver shortage that lowers demand — and it arrives later than the headlines suggest. For buying that means: until the end of the decade the price is a matter of negotiation, not technology.
Source: industry radar · freight exchange
-
High impactLong term
When less is transported, organising is the business
A peer-reviewed paper predicts AI weakens rather than strengthens classic logistics firms — not because they fail to use it, but because their customers do.
my position
read on
The open point: Two routes lead there, and both are walkable. One is volume: production closer to the market, spare parts from a 3D printer, smaller stocks, less spoilage — for spare-parts production roughly twelve per cent less material input and around nine per cent less CO₂ are projected, and each of those movements produces fewer trips. The other is the role: whoever masters demand planning, stock and procurement with AI themselves needs their service provider for less. The authors’ counter-strategy is the role of the organiser: take-back, disassembly, sorting and return transport are multi-stage and coordination-heavy, and in long chains with many small participants nobody has taken that on so far. Source: Kokkinou and Mitas, Cleaner Logistics and Supply Chain 18 (2026), openly accessible.
From practice: Not the business model — the preconditions. What a forwarder can set against that is not the vehicles but the information flows: it sits between shipper and carrier and knows both sides. What is missing is the first case where a take-back chain pays for itself. That is more honest than an announcement — and the reason this topic sits in Watch and not in Extend.
What would move it: When a shipper tenders a take-back chain instead of announcing one.
In the industry radar: the circular economy is tracked, but as a sustainability topic. That it could at the same time be the answer to shrinking transport volume is not said there.
What a shipment costs in CO₂e – open the calculator →Where the figures for it come from →
-
Moderate impactLong term
Connectivity everywhere
Low-orbit satellites close the dead spots. Five to ten years out.
industry radar
read on
What it means here: Dead spots are the quiet reason for half data sets: a signal created in a tunnel arrives late or not at all. Anyone computing arrival times from telematics should know how often the chain has gaps before promising them to a customer.
Source: industry radar
-
Moderate impactLong term
One filing instead of many
The planned EU Customs Data Hub is meant to capture customs data once and reuse it. Start in 2028 at the earliest, subject to the ongoing negotiations.
EU customs reform
read on
What it means here: The benefit is far off and the preparation is not: file structured data today instead of sending documents and you have what the hub will expect. The date is the least certain in this row.
Source: EU customs reform
-
Moderate impactLong term
The digital product passport
To take a product apart and sort it, its composition must be known along the way. It fails because nobody who gains nothing wants to supply the data.
research
read on
What it means here: So far it fails on how the effort is spread: the data has to come from whoever gains nothing from it. For logistics it is still interesting because it makes returns plannable — without knowing what is inside there is no sorting.
Source: research
-
Low impactLong term
Blockchain
Moved distinctly outward: low impact, late adoption.
industry radar
read on
What it means here: The placing is clear and it holds for this business: a chain pays where several parties must trust the same record without trusting each other. In a forwarder’s day there is one such place — the consignment note — and there regulation solves it anyway.
Source: industry radar
What I take from the industry radar — and where it stays silent
The points above carry their origin, because otherwise they would be claims. That leaves the question of what I take from it, where I disagree, and what it does not mention at all. The last of those is the actual reason this page exists.
What I take from it
Four lines hold here too. The bottleneck is not the model but data quality and access. Explainability comes before autonomy. Attacks will come out of models themselves. And automation is to be built as relief, not as replacement — which is exactly the decision to settle for the seventy clean documents rather than the hard thirty.
Where I disagree
Blockchains have moved outward there: low impact, late adoption. I have said so for a while, only more precisely — a chain earns its cost only where several parties have to trust the same record without trusting each other. And from “generative AI, high impact, three to five years” no instruction follows: how a company gets there is not said.
Where it stays silent
Agent-to-agent, regulatory obligation as a driver, and liability on binding awards do not appear there as trends of their own. That is no reproach: a radar published every two years cannot carry a development that is twelve months old. It is the reason this page exists.
What I think the next step is
All nine topics converge on the same bottleneck: the models can do enough, what is missing is the process around them — and that has to be built for every case. Twenty of those are feasible, two hundred are not. So I do not see the next bigger model as the next step, but the AI that writes the workflow itself instead of merely executing it.
The pilot and failure-rate figures come from MIT’s NANDA project and from RAND Corporation, the market estimate from Gartner. The AI Act deadlines are in Regulation (EU) 2026/1744 of 27 July 2026, those for the electronic consignment note in the eFTI Regulation (EU) 2020/1056. The autonomous driving statements refer to the ATLAS-L4 research project and to manufacturer information on ongoing trials. I deliberately do not name large logistics providers: that companies of that size run agents in production carries the point — which ones does not.
Where the best solution sits today
What is named is the pattern, not the vendor — a product name ages, an idea does not.
Bounded autonomy
The pattern that wins in real deployments — clearly defined authority, a clean escalation path. It comes from the studies, not from a vendor brochure.
Structured beats recognised
The eCMR makes document recognition unnecessary in many places. The best solution here is not a better model but the disappearance of the problem.
Organisational, not technical
The failure causes named are without exception questions of ownership, definition and measurement. No tool changes that.