According to theloadstar.com, DSV’s decision to develop its own proprietary technology platform and exit CargoWise has triggered a fundamental reassessment across the freight forwarding industry—not about whether to build or buy software, but about which capabilities must remain proprietary and which can be outsourced.
From monolithic systems to modular foundations
Kristjan Lillemets, who served as chief product officer at Magaya during his interview with The Loadstar, observed that forwarders are no longer debating wholesale software replacement. Instead, they’re asking: “
‘Can I take what I consider my secret sauce and do that part myself?’… that’s the chatter we’ve been hearing recently,”
— Kristjan Lillemets, former chief product officer, Magaya.
This strategic pivot reflects a broader industry evolution toward hybrid architectures: acquiring robust, certified core systems for shipments, customs, finance, and compliance while retaining full control over custom AI-driven workflows, integrations, and customer-facing layers. As Lillemets explained, “
The foundational systems at the bottom still need to be solid. The tip of the iceberg is what we see customers build.
” That distinction underscores a hard operational reality: reliability and predictability in software partnerships have become non-negotiable—especially after WiseTech launched its CargoWise Value Packs in late 2025.
AI adoption outpacing commercial viability analysis
While generative AI demonstrations dominate headlines, Lillemets warned that the industry remains dangerously unprepared for the economic implications of large-language-model (LLM) deployment. He noted that “
The understanding of what was possible nine months ago would be severely outdated today.
” Yet most forwarders haven’t yet modeled the total cost of inference, API latency, or data residency constraints tied to frontier models like those from Amazon and Maersk, both of which announced major AI investments in 2026.
Lillemets emphasized that ROI erosion is imminent if companies treat every task as requiring state-of-the-art LLMs. “
If we all keep using the frontier models for everything, then I think that’s where the ROI question… is going to present itself in a pretty obvious and painful way soon.
” His recommendation: match model capability to task criticality—using smaller, cheaper, or locally deployed models for document parsing, exception flagging, and workflow orchestration. This approach aligns with findings from Neolink, an Australian forwarder where Director Sean Crook confirmed that traditional automation delivered greater near-term operational gains than generative AI—despite Amazon’s $1.7 billion AI infrastructure investment disclosed in Q1 2026.
Productivity gains mask a deeper human capital challenge
Early AI deployments are yielding measurable efficiency lifts: operators now process 37% more shipment records per shift, and manual data entry from PDFs and paper documents is being eliminated at scale. But Lillemets stressed that these productivity metrics conceal a structural risk. Customers report not headcount reduction—but operator upskilling: “
What they quote is not that ‘I’ve been able to reduce headcount’, but that ‘my operators now are this much more productive’.
”
This shift raises urgent questions about workforce development. With AI handling routine tasks—including document classification, tariff lookups, and exception triage—the experiential learning path for new freight forwarders is narrowing. Logistics adviser Wolfgang Lehmacher recently posed the pivotal question: “
How is the next generation supposed to learn?
” That concern, rooted in observable training gaps at firms deploying AI since December 2025, may prove more consequential than licensing models or platform architecture decisions.
Source: The Loadstar
Compiled from international media by the SCI.AI editorial team.
