Large forwarders report big gains, wider surveys show stalled pilots, and the data points to input quality as the dividing line.

C.H. Robinson told investors on 29 July that productivity across its North American surface transportation and global forwarding businesses has improved more than 60% since the end of 2022. In the same quarter, global forwarding productivity rose more than 15% year on year, and average headcount fell 10.8%. A year earlier, Gartner had forecast that more than 40% of agentic AI projects will be cancelled by the end of 2027. Both statements describe the same industry, and the executives who read them are drawing different conclusions.

Gains reported at the top

Kuehne+Nagel is at an earlier stage. Its Q2 2026 analyst presentation projects an AI-driven productivity benefit of 5%, which the company estimates at CHF 100 million to 150 million of annualised uplift in 2027. Chief AI and Innovation Officer Alireza Nemati told analysts the company has “made our data AI-ready” and is redesigning workflows around AI before scaling adoption. CEO Stefan Paul told a media conference, as reported by AWP, that the aim is to handle higher volumes with the same headcount, and that a business unit unable to grow volumes with AI will face questions about cost.

Company executives pointed to early examples, as reported by FreightWaves. An AI-powered pricing tool in air logistics delivers quotes twice as fast as before, and AI built into the myKN platform cuts sea booking time from minutes to seconds while lowering human errors. In customs, AI-driven automation is reducing handling time per declaration.

The two forwarders are comparing a three-year result with a forward target, so their figures sit on different footings. They also differ on what productivity means for staff. Sanjeev Ghadia, CEO of Astral Aviation, told The Loadstar he believes the company “would not lose a single person” in adopting AI. Robinson’s headcount fell while volumes grew. Both companies are large enough to build and run their own systems, which few mid-sized forwarders can match.

Buyers hold back

Gartner’s analyst Anushree Verma said most current agentic projects are “driven by hype and are often misapplied”. Gartner cited escalating costs, unclear business value and inadequate risk controls, and estimated that only about 130 of the thousands of vendors claiming agentic capability offer it. The same forecast looks further out, expecting at least 15% of day-to-day work decisions to be made autonomously by agentic AI by 2028, up from none in 2024.

Shippers send a mixed signal. A BCG and Alpega survey run in January 2026 found that over 40% of shippers now expect logistics providers to offer AI-enabled services, while fewer than 10% regard AI as mandatory in their logistics partnerships. ORTEC surveyed more than 2,000 North American transportation, logistics and supply chain executives in November 2025. It found 42% still holding back on agentic AI, and 23% planning a pilot within 12 months. Among organisations yet to adopt AI or machine learning in core processes, lack of in-house expertise (23%) and unclear return on investment (21%) were the leading barriers.

The input problem

Data readiness appears in almost every account of why pilots stall. Ontegos Cloud surveyed around 800 forwarding and logistics professionals in August and September 2025. Some 48% expected AI to transform forwarding within three years and 14% were unconvinced, but only 23% said 75% or more of their company’s data was clean and reliable. Founder Oliver Gritz said successful AI projects rest on “a structured company data model and a reliable system architecture”. More than 70% of respondents spend at least a quarter of the working day on manual tasks such as chasing documents and emails.

A December 2025 survey by Parseur, a document processing vendor, found that 88% of US organisations discover errors in document-derived data at least sometimes. Respondents named logistics documents among the frequent sources.

Errors that pass unnoticed create rework later. The word “workslop” entered the vocabulary in 2025 through research from Stanford Social Media Lab and BetterUp Labs, published in Harvard Business Review. In a survey of 1,150 US desk workers, 40% said they had received AI-generated work in the past month that looked polished but lacked substance. Recipients spent an average of one hour and 56 minutes fixing each instance, and the researchers put the cost at $186 per employee per month. The sample covered general office work with no freight respondents, though anyone who has rechecked a machine-read booking confirmation line by line will recognise the pattern.

MIT’s NANDA initiative reported in its GenAI Divide study that around 95% of enterprise generative AI pilots in its dataset delivered little to no measurable impact on profit and loss. Lead author Aditya Challapally has attributed the stall to a learning gap: generic tools such as ChatGPT work well for individuals but do not adapt to enterprise workflows. The study found the strongest returns in back-office automation.

Building for the document

Founder of Deep Current, Tamim Fannoush quotes, “Most AI problems in freight forwarding begin before the AI runs. A booking, a dangerous goods declaration or a quote arrives with an error in it, and every tool downstream inherits it. Our team worked in this industry before AI arrived, and we saw how late those errors surface and how much they cost by then. Checking the document at the start means far fewer interventions later.”

Deep Current’s website lists four headline results: 70% less time spent on manual documentation, 60% fewer customer query emails handled by operations teams, 99% error-free workflows with automated validation, and 30% more shipments managed with the same headcount.

The next test is close. Kuehne+Nagel expects its AI productivity benefit to arrive in 2027, and Gartner expects the project cancellations it forecasts to land by the end of that year. Both will be judged in the same operations, using different yardsticks. Robinson reports productivity per head, Kuehne+Nagel reports francs of annualised uplift, and the workslop researchers report hours of rework.

Buyers can set those yardsticks side by side before then. They can ask what baseline each figure was measured against, how long a person spends checking the output per document, and how clean the data was before the tool touched it. Kuehne+Nagel’s own sequence gives a model to test against: data made AI-ready first, workflows redesigned second, and adoption scaled third.

For a forwarder still deciding where to begin, the evidence in this piece points to a starting place: the document that arrives first. Whatever tool comes next will work with what that document contains.

For starters, Deep Current offers forwarders an analysis of their document workflows, covering where errors enter the process and where it can run more efficiently. To arrange one, get in touch with the team at info@deepcurrent.no