AI in freight forwarding, autumn 2026: what already works and what a firm of 10 to 50 people can manage

For twenty years at NIKOS I saw where a forwarder's day goes. You take an order and enter it into the system. You find a truck on a freight exchange and agree on the price. You work out the rate, the loading metres and the driver's hours, and prepare a quote for the client. Then you collect the documents, send the invoice and wait to be paid.

In October 2026 I looked through job ads for freight forwarders in Latvia on cv.lv and ss.lv. Companies are hiring people for exactly the same work and paying €800 to €3,500 a month gross for it.

Over these years, software has appeared that takes on this work: AI agents. An agent reads emails and documents, writes, makes calls and works in your software the way an employee does. In the US, London and Munich, investors are putting hundreds of millions of dollars into them.

Latvia has about 750 carriers and forwarders with 10 or more people [21]. According to Eurostat, 5.9% of Latvian transport and storage firms of that size used AI in 2025. The EU average was 11.2%. In Lithuania it was 20%, four times as many as a year earlier [1]. In Germany, one in two logistics firms with 20 or more people already uses AI. In 2022 it was one in five [2].

Why logistics is first in line

The co-founder and CEO of sennder (one of Europe's largest digital freight forwarders, with about a billion euros in revenue a year) said at a public panel in Berlin: "Logistics is the best industry where you can apply AI. A lot of manual work. A lot of repetitive tasks. They're always the same" [3]. Anyone who has watched a forwarder on a Friday evening calling carriers one after another will find that hard to argue with. By TIMOCOM's count, a dispatcher exchanges messages or talks with 16 partners on average before a load goes to a carrier [4].

The second reason is pressure. Europe is short of 502,000 truck drivers, 13% of all positions. At firms with up to 50 people the gap is even wider, 15%, against 9% at large ones [5]. Arqh, a startup that optimises road freight, made the point in the same Berlin discussion: as long as you can work the way you did twenty years ago, nobody takes up the technology. Companies take it up when the pressure is on [3].

The CEO of GXO, one of the world's largest warehouse operators, put it more bluntly: "People are not going to lose their job to AI. People are going to lose their job to people who use AI" [6]. The same goes for companies.

Five areas where AI already works

1. Incoming orders

The agent reads an order from an email, a PDF or an Excel file and enters the shipment into your software: route, dates, cargo, client. If something is unclear, it asks a person.

The American startup Vooma started with exactly this, order entry for brokers and carriers, and only later moved on to quotes, finding carriers, tracking and collecting proof of delivery [7]. Since 2025 the TIMOCOM freight exchange has been able to turn the copied text of an order into a freight offer with one click [4]. In a Forto survey of European companies, automatic order entry was the second most common use of AI after chatbots [8].

A firm of 10 to 50 people can manage this, and it is usually the place to start.

2. Rates and quotes

The agent puts together a draft quote from the client's request and your price history on that route, and passes it to a person to approve.

Cargo Movers, an air freight forwarder in Frankfurt with about 85 people, put an assistant like this straight into Outlook. Of its quotes, 54% go to the client without a single edit and another 17% with one edit (according to the vendor) [9]. Nexcade in London builds agents that read the request, clarify the details, collect rates and prepare the quote. Its clients include XPO and Zencargo [10].

Drafts are within reach too. The final word on price stays with a person: it carries what you know about the client, the market and what happened last week.

3. Finding a carrier

The agent posts the load on a freight exchange, sorts the replies by your rules, answers carriers' standard questions, and writes to or calls the carriers you have a good history with.

Since May 2026 Transporeon has let carriers in Europe search for loads by typing an ordinary sentence. Since September it has also run an agent that reminds carriers and asks for missing documents on its own [11]. In September TIMOCOM showed an assistant for negotiating on the spot market. It replies to carriers and prepares options, and the dispatcher makes the decision [4]. In the US, HappyRobot and Vooma do this.

Here the agent takes only part of the work: the routine of the search. But "a small trucking company from Poland still shows up every Christmas and brings me homemade sausages," says the CEO of sennder [3]. No platform will replace a relationship like that, and the deals with your own carriers will still be yours to make.

4. Status updates to clients

The agent tracks the truck by GPS or calls the driver, and tells the client where the cargo is before the client starts calling.

In a video report on Vooma, the agent calls a driver on camera, asks when he arrived and when he finished unloading, and records the answers [12]. The driver talks to it as he would to a dispatcher. At Kuehne+Nagel, a HappyRobot voice agent checks the status of medical shipments around the clock: more than 10,000 checks, and the team handles 47% more volume (according to the vendor) [13].

Realistic if you have GPS data or the drivers' phone numbers. At sennder they said they spent years building a portal for clients, and the clients hardly used it: a client has 20 to 30 suppliers like them, and it is easier to pick up the phone [3]. So the status has to reach the client on its own, in the place where they already work: email or a messenger app.

5. Documents, invoices and payment

The agent reads CMRs, invoices, proofs of delivery and customs documents, and checks that the set of documents is complete and the figures match. It also prepares your invoice and checks carriers' invoices, then reminds the client when payment is due and matches incoming payments against the bank statement.

At Cargo Movers, checking a 500-line invoice takes minutes instead of two to three hours [9]. Wackler, a family firm from Göppingen, cut customs clearance from 39 to 14 hours a week (again according to the vendor) [14]. Kuehne+Nagel prepares customs declarations with an assistant called Claire [15].

Realistic, and this is the most mature area: reading documents works more reliably today than anything else on this list.

Who is putting money into this

The typical client of Munich-based 5U AI is a forwarder with 100 to 500 people [17]. The others work with large US brokers and global players. A 15-person firm in Riga, with emails in Latvian and Russian and local software, is of no interest to them and will not be for a long time.

As William Gibson put it, the future is already here, it's just not evenly distributed. The big players build their own. Kuehne+Nagel is bringing AI into the work of 25,000 office staff and expects at least 5% more productivity and CHF 100 to 150 million in profit a year from it by the end of 2027 [15]. At sennder, gross profit per operator grew 45% in a year, and the company credits this partly to AI [18].

What does not work yet

If someone promises you that an agent will replace a department, don't believe them.

Data is locked inside software. Arqh told the story of a client who would have had to pay thousands just to export its own data from its own system for a pilot [3].

Accuracy. "Enterprise companies don't want 90% accuracy. They want 99.99," they say at Vooma [12]. An agent that makes mistakes silently is worse than an employee. A good agent says "I'm not sure" and hands the case to a person. At Cargo Movers the assistant sends nothing by itself: a person approves every step [9].

Implementation fails. By MIT's estimate, 95% of companies have not yet had a measurable return from generative AI [19]. In supply chains, about 5% of projects reach company-wide scale [20]. The models are good enough today. What breaks is everything around them: who checks the work, who is responsible, how the agent fits into the daily routine. The same MIT data shows that projects done with an outside partner made it into everyday use twice as often as in-house builds [19].

People. The CEO of sennder says people are afraid of losing their superpower: being the one who can call ten carriers or plan the week in Excel [3]. That is a normal fear. The operator's job stays, but it changes: less "do you still have a truck?" and more talking with clients and carriers about how to work together. In Germany, 54% of logistics firms say they lack the know-how to build AI into their work [2].

The objection "we're too small for this"

The CEO of sennder raised it himself: "Only once you have a certain size does it really make sense to use technology. The moment you're small, it's not worth it" [3]. For sennder that was true. They had to build their own systems, and development like that costs millions.

Vooma explained what has changed. Every firm has its own way of working, and writing rigid software for each one used to make no economic sense. With AI agents, that equation has shifted [12]. An agent works with your email, your Excel files and your software. It does not need a new system.

So instead of asking what AI costs, count what you pay for the job it partly takes over: the salary from your job ad plus employer taxes, plus the months a new hire needs to learn the work. A small firm also has an advantage that sennder does not: one person makes the decision, and a pilot can be up and running within a few weeks. According to MIT, the best mid-sized companies get from pilot to everyday use in about 90 days [19].

Where to start if you have no programmers of your own and no investor money

1. Take the text of your latest job ad. Underline everything that is not a conversation with a client or with a carrier. Those are your candidates.
2. Pick one area where the work repeats every day and a mistake shows up at once. Usually that is incoming orders, or documents and invoices.
3. Ask your software vendor whether data can be exported from the system and entered into it from outside. Half of the timeline and the price will depend on that answer.
4. Run a one-month pilot in one area. Do the before/after measurement yourself and don't leave it to the vendor.
5. Decide in advance who will deal with the cases the agent hands back to a person. Without that, the exceptions pile up, and a month later people are saying the agent "doesn't work".

What I have already built myself

Since 2023 I have been building systems like these, first for my own companies and then for other people's. Three of them run every day:

• An arbitration court in Latvia. The system has been running since July 2026. Cases used to be handled on paper and in Word. Now a claim comes in through an online account, and every document (summons, decision, invoice) comes out of the system already filled in. By the court office's own measurement, the paperwork for one case now takes two to three hours to put together instead of several days.
• A debt collection service. Running since April 2026. It reads the client's documents, calculates the debt with late-payment interest and the court fee, writes to the debtor and prepares the claim.
• Voice call answering at my barbershops. Since October 2025 the agent has taken 1,518 calls from 952 people. It answers in three languages, checks free slots and books the client on its own.

If you would like to talk about which of this fits your firm, write or call. If now is not the time, that's fine too.

Konstantin Ahmedov · +371 29476727 · puppetmaster.agency · LinkedIn

Sources

Figures from vendor companies are given as the vendors report them; where that is the case, the text says "according to the vendor".

1. Eurostat, Use of artificial intelligence in enterprises, 2025 survey (tables isoc_eb_ai, isoc_eb_ain2, updated 15 June 2026). Firms with 10 or more people. https://ec.europa.eu/eurostat/databrowser/view/isoc_eb_ain2/default/table
2. Bitkom, survey of 502 logistics firms in Germany with 20 or more people, 10 September 2026. https://www.bitkom.org/Presse/Presseinformation/Die-Logistik-investiert-in-KI-doch-es-fehlt-Knowhow
3. Panel "Why Logistics Is the Perfect Industry for AI", Merantix AI Campus, Berlin: David Nothacker (sennder), Antonia Unger (Arqh), Tassilo Schmidt. Video, 26 February 2026. https://youtu.be/OqqOVi4p8t4
4. TIMOCOM: AI freight offer from the text of an order, 7 May 2025; negotiation assistant, IAA Transportation, September 2026. https://www.timocom.co.uk/company/newsroom/new-way-of-assigning-freight-transport-logistic-2025-783390 · https://logistik-heute.de/en/news/iaa-transportation-2026-timocom-presents-ai-negotiation-assistant-spot-market-283864.html
5. IRU, Global Driver Shortage Report 2025 (published June 2026). https://www.iru.org/news-resources/newsroom/operators-deeply-concerned-worsening-driver-shortage-new-iru-report
6. Patrick Kelleher, CEO of GXO, interview with Global Money Talk, 17 September 2026. https://youtu.be/Q0x8GFE9mDw
7. Vooma: YC W23; $16.6 million (Index Ventures, Craft Ventures), 2 December 2024. https://www.vooma.com/ · https://www.businesswire.com/news/home/20241202150186/en/
8. Forto, Logistics Trend Compass 2025. https://forto.com/wp-content/uploads/2025/05/Trend-Compass-2025-Digital.pdf
9. Cargo Movers and Nemox, transport-online.de, 28 September 2026 (figures according to the vendor). https://transport-online.de/en/news/ai-inbox-how-frankfurt-freight-forwarder-automates-offers-193553.html
10. Nexcade, $6 million seed round, 8 July 2026. https://theloadstar.com/ls_press_release/nexcade-raises-6m-to-build-ai-agents-for-freight-forwarders/
11. Transporeon (Trimble): AI load search, 11 May 2026; agent features, 23 September 2026. https://itbrief.com.au/story/transporeon-launches-ai-search-for-carriers-across-europe · https://itsupplychain.com/transporeon-unveils-agent-ready-portfolio-to-turn-manual-logistics-into-connected-autonomous-workflows/
12. "Inside a YC Startup Disrupting a $900B Industry", a video report on Vooma. https://youtu.be/zY-dqw0_6ug
13. HappyRobot: Series C, 4 August 2026; Kuehne+Nagel case study (according to the vendor). https://www.happyrobot.ai/blog/happyrobot-seriesc-fundraising-announcement · https://happyrobot.ai/customer-story/kuehne-nagel
14. Wackler Spedition & Logistik and Digicust, 10 March 2024 (according to the vendor). https://digicust.com/testimonials/wackler-logistics
15. Kuehne+Nagel: Q2 2026 analyst presentation, 23 July 2026; Annual Report 2025. https://a.storyblok.com/f/331466/x/b687283818/q2-2026-analyst-presentation.pdf · https://2025-annual-report.kuehne-nagel.com/corporate-brochure/english/interview-with-the-ceo
16. Augment: $85 million Series A, $110 million in total, 4 September 2025. https://8vc.com/resources/announcing-the-augment-8vc-partnership
17. 5U AI: $3.2 million pre-seed, 28 July 2026. https://techfundingnews.com/5u-ai-3-2m-pre-seed-freight-forwarding-ai-workers/ · https://cargoforwarder.eu/2026/07/28/5u-ai-from-football-to-agentic-ai-freight-start-up/
18. sennder, press release, 21 October 2026. https://www.sennder.com/en/press/sennder-achieves-profitability-and-hit-euro1-billion-in-annual-revenue/
19. MIT NANDA, "The GenAI Divide: State of AI in Business 2025", July 2025 (across all industries; the authors' preliminary findings).
20. GEP and UVA Darden, study of AI in supply chains, 20 May 2026. https://www.aap.com.au/aapreleases/cision20260520ae63007/
21. VID (State Revenue Service of Latvia), taxes and average number of employees by company for 2024, data.gov.lv; NACE codes 49.41, 52.21, 52.24, 52.25, 52.26, 52.29, 52.31.