As I continue to digest and assess the content of last week’s $WTC Investor Day, while much of the press attention has focused on the implementation of the New Commercial Model, the session that impressed me most was the discussion on AI implementation.
I’ve decided to pull out a separate straw on this, because I think it may be of interest not only to those following $WTC, but also to the wider community interested in how firms are deriving value from AI.
$WTC’s AI philosophy
WiseTech’s AI strategy is built around narrow, specialised agentic AI, deeply embedded into CargoWise workflows, rather than broad, general-purpose models. Guardrails and human-in-the-loop verification are core design principles.
$WTC is scoping many potential AI agents. However, three are already “live” and can be accessed via the new Value Packs with 100% transaction-based pricing:
- Document ingestion (data entry automation)
- ComplianceWise – Export Controls
- Customs Classification Assistant
Each of these was described in detail at the Investor Day and provides evidence of how $WTC’s massive proprietary dataset, combined with LLMs’ ability to structure, process and reason over text-based data, can materially reduce the high labour content in existing freight forwarding, import and export processes.
Let’s look at each in turn.
1. Document ingestion
AI-native document ingestion replaces manual data entry and OCR-based bolt-on tools.
Example impact:
- Commercial invoice ingestion reduced from a ~5–6 minute manual task to near-zero
- At a global freight forwarder level, ~10 million commercial invoices per year equates to ~95 human years of data entry
Accuracy is high, driven by:
- Targeted models tuned to specific document types
- Uncertainty explicitly flagged to human operators
- Accuracy described as materially higher than OCR
From the Q&A discussion: accuracy is improving while labour is being removed, and this improvement is not being traded off against risk.
2. ComplianceWise – Export Controls
Across global jurisdictions, numerous controls restrict parties, locations and categories of goods.
AI agents assess export risk across:
- Parties (denied / sanctioned lists)
- Locations (embargoes, restricted destinations)
- Goods and end-use (dual-use items, munitions)
The AI effectively acts as a virtual compliance officer: jurisdiction-specific, always-on, and capable of scanning all available information against known restrictions and risk factors, flagging issues for human review.
Performance achieved:
- ~96% precision (low false positives)
- No missed red flags relative to expert human reviewers
3. Customs Classification Assistant
A core role of freight forwarders is ensuring goods are correctly classified. This task is exceptionally complex due to:
- The sheer number of codes
- Non-intuitive classification structures
- Jurisdictional differences
As a result, classification remains one of the most labour-intensive parts of the process.
$WTC’s AI agent is now able to complete ~90% of the classification work, leaving brokers to verify and submit (human in the loop).
Industry benchmarks:
- Human-based classification accuracy ranges from as low as ~20% to ~80%
- ~80% is typically considered a strong human benchmark
Pilot results: customers using the $WTC AI agent are reporting ~90% accuracy.
Importantly, narrow agent design and embedding regulatory data from $WTC databases (e.g. BorderWise) materially reduces the risk of AI hallucinations.
Richard White gave an illustrative example of a UPS business unit with 19 human classifiers and 1 supervisor. The AI agent is capable of performing the work of the 19 classifiers, leaving the supervisor to check and verify outputs as the only future required labour for this process.
Other AI applications
As with many enterprise software firms, $WTC is also developing AI-based customer service agents. The WiseTech agent is called ACE, supported by content from the WiseTech Academy.
Although ACE is only one month old (“a baby”, in management’s words), it is already being trained on ~20,000 historical support tickets where humans previously sent specific training materials to customers. Management estimates this could free up the equivalent of ~18 product managers currently involved in this work, allowing them to be redeployed into product development.
Finally, $WTC is also using AI internally to assist with code writing and testing.
These latter applications have quickly become ubiquitous across enterprise software, so there was little basis to assess whether $WTC is meaningfully ahead or behind peers in these areas.
Overall takeaways and management view
Richard White believes that rolling out AI agents within CargoWise and related products has the potential to reduce customer labour requirements by ~50% over two years. This is a bold claim - characteristic of RW’s visionary style - but the examples demonstrated so far lend it credibility. Time will tell.
It is difficult to overstate the complexity of freight classification and compliance. In another Investor Day session, Anthony Hardenburgh (Product Portfolio Leader, Global Trade Management) noted that across markets and products, $WTC handles ~73 million regulatory and data updates annually, all of which must be incorporated into its systems. This scale is far beyond what humans can manage alone and is precisely the type of problem suited to AI. It appears $WTC is leaning into this opportunity in earnest.
$WTC’s approach centres on carefully scoped, narrowly defined AI agents, tightly integrated into workflows. Management described a common AI agent architecture with an abstraction layer that allows A/B testing of different LLMs and rapid swapping as models improve and leap-frog each other.
Early labour reduction from AI is expected to occur primarily in lower-skill roles, many of which logistics firms have already offshored to shared-service centres. Remaining staff, employees of $WTC’s customers, will become more skilled and more valuable as supervisors of AI agents.
There was discussion around whether customers could build their own AI layers and bypass $WTC’s products. Management acknowledged this was possible and that some customers might see benefits. However, CEO Zubin emphasised that software development and workflow integration are WiseTech’s core expertise, and that embedding AI directly into CargoWise should deliver superior outcomes, because only $WTC has access to the source code and databases. That confidence seems reasonable, though real-world results will ultimately decide.
My assessment
AI adoption across large enterprise SaaS companies has become a central part of the investment thesis for my technology holdings over the past few years. The use cases showcased by $WTC are exactly the types of applications I was hoping to see, and it is encouraging that early solutions are now "live" with customers - particularly the 95% of largely small customers who have gone "live" on the new Value Packs, potentially a strong incentive for the LGFFs to follow quickly!
It will be particularly interesting to track adoption and realised benefits over the next one to two years.
This also underscores why $WTC had to move from “seat plus module” pricing to per-transaction pricing. Despite the noise surrounding the transition, there was really no alternative. Labour hours per shipment are about to be materially reduced, while shipments-per-seat should increase dramatically as these capabilities are adopted by customers. $WTC had to get ahead of that curve.
Overall, I was very impressed by the presentations, the examples, and indeed the presenters themselves. I was glad to see that $WTC are indeed doing what I expect them to be doing. It was good to learn the facts, rather than just hold the thesis!
Disc: Held
