AI takes shape in freight forwarding
Sonia Sali
AI is fast becoming part of freight forwarding, but its real impact is still taking shape. As companies explore AI across documentation, operations and decision-making, connected systems and clean data remain key. In an interview with CargoTalk Middle East, Toby Edwards, Director, Trade Tech, highlights the opportunities and challenges shaping AI adoption in freight forwarding.
AI adoption in freight forwarding remains at an early stage, with companies still experimenting with the technology and seeing limited results at scale. Toby Edwards points to research from The Loadstar and Raft, which found that only 22.2% of organisations have deployed AI at scale. He also cites MIT’s (Massachusetts Institute of Technology) Project NANDA study, which found that 95% of enterprise generative AI pilots delivered no measurable return. “The announcements are running well ahead of the outcomes,” Edwards says.
According to Edwards, document processing is where much of the early progress with AI is taking place. However, this depends heavily on the quality of the underlying data. “It only works where the data underneath is already clean and connected, which comes from automation,” he says. Without this foundation, AI can simply move messy data around faster. “The state of the data matters more than the area.”
AI adoption
Most freight forwarding companies are currently adopting bolt-on AI tools that sit on top of their existing systems. These include document readers, chat assistants and exception alerts. However, fragmented data remains a challenge. Edwards says forwarders seeing real value are those that have first automated repeatable work on a connected platform and are then adding AI on top.
Several business pressures are also driving logistics companies to explore AI. These include margin pressure, increasingly complex compliance requirements across Customs regimes, staff turnover and growing customer demands for visibility. Boards are also asking companies what they are doing about AI. Edwards says this is a major driver, but adds that wanting to be seen as doing something with AI is not the same as knowing what problem needs to be solved.
Data Foundation
For freight forwarders, Edwards believes most of the real gains come from automation rather than AI itself. Converting rates into quotes, quotes into bookings, generating Customs filings, triggering milestone notifications, allocating costs and producing invoices are all rules-based activities that automation can handle when data is connected.
AI can then work as a layer on top of this foundation, helping with more complex tasks such as reading bills of lading and invoices or identifying exceptions across large volumes of shipments. Accounting is one area where this distinction is particularly important. Shipment-level costs, revenues and reconciliations remain difficult for many forwarding businesses because of disconnected systems. “AI cannot fix it. Connecting the data and automating the flow can,” Edwards says.
The same principle applies to decision-making for complex shipments involving multiple routes, carriers and regulations. Edwards notes that many decisions are not judgement calls but repeatable steps that follow rules. Automation can handle these when the data is connected, removing much of the workload.
AI can then help by flagging conflicting requirements, suggesting routing alternatives and warning about likely exceptions. However, Edwards stresses that the quality of these recommendations depends on the data being used. Many forwarders operate systems that were never designed to work together across different languages and regulatory environments. As a result, applying AI to fragmented information can produce inconsistent or incorrect recommendations.
“A single shipment record shared across sales, operations, compliance, and finance is the prerequisite,” Edwards says.
Cybersecurity is another area companies need to consider as they implement AI. Edwards identifies three key risks. First, every AI tool added to a business creates another connection that needs to be secured, increasing the surface area for something to go wrong. Second, data may leave the business when shipment information is sent to a third-party AI provider for processing. In such cases, the security of that provider becomes part of the company’s own security. Third, AI itself can be manipulated. Document processing tools, for example, can be fooled by forged or altered paperwork if the appropriate checks are not in place.
Edwards comments that these risks are not a reason to avoid AI, but a reason to approach it with the same discipline as any other critical system.
Companies also need to be careful about how they use generative AI tools with sensitive supply chain information. Edwards recommends treating any AI tool as a third party handling company data and asking where the data goes, how long it is retained, whether it is used to train the model and who can access it.
Public AI tools should not be used for information containing customer names, rates, commercial terms or shipment details unless an enterprise agreement clearly excludes the data from training. Where possible, Edwards recommends using AI within the operational platform so that the data does not leave the business.
Policies, employee training and controls that prevent sensitive information from being pasted into public AI tools are also important. “Assume anything pasted into a public tool has left the business for good,” Edwards says.
Comments