Customs brokering and the skills gap: AI as the answer, with one big caveat
Half of Australian licensed brokers are 50 or older. Trade volumes are forecast to rise 50% by 2032. AI won't replace the broker — but it can multiply their capacity.
Australia's customs brokers are ageing out faster than new ones are coming through. AI will not replace the licensed broker, but one broker can handle several times the volume with AI doing the first pass on extraction, classification, and rules-checking. The real risk is not the technology. It is complacency in the review step. AI buys the industry roughly a decade of breathing room on the headcount problem. It does not fix the pipeline.
The Australian customs broker profession is in a slow-moving crisis. Nearly half of licensed brokers are 50 or older. Fewer than 11% are under 40. The 20–29 and 30–39 cohorts are down more than 12% year-on-year. Border Force forecasts a 50% rise in trade volumes by 2032.
A piece in the DCN recently made the case that AI is not the answer to this — that it can only assist, not replace, because statutory responsibility under the Customs Act 1901 sits with the licensed broker.
We agree with the legal point. We disagree with the conclusion.
The framing problem
"Will AI replace the broker?" is the wrong question. It produces a binary answer to a workforce problem that is fundamentally about capacity.
The real question: can AI close the gap between declining headcount and rising volume? On that, our view is yes — and the maths is not close.
Consider what a customs broker actually does on a typical entry. Extract data from commercial invoices, packing lists, and bills of lading. Classify goods to HS codes. Calculate duties and taxes. Check for permits, anti-dumping, country-of-origin rules. File the entry. Manage the back-and-forth on missing or ambiguous information.
The first four of those — extraction, classification, calculation, rules-checking — are where models are already strong. Not perfect. Strong enough that a broker reviewing AI-prepared entries can process 3–5× the volume they could from a blank screen, with statutory sign-off intact.
What "augmentation" actually means in this context
The DCN piece is right that AI assists rather than replaces. But the word assists understates the change.
If one licensed broker can sign off on the work of an AI-prepared queue that previously needed three or four people, the workforce maths changes. You don't need to recruit your way out of a 50% volume increase. You need to recruit your way out of maybe a 15% volume increase, with the rest absorbed by throughput per broker.
That's the answer to the shortage. Not headcount substitution. Capacity multiplication on the people you already have.
Where the real risk sits
There's a legitimate concern in the DCN piece that's worth sitting with: automation complacency. Brokers who stop scrutinising AI-generated classifications, then sign off, then own the consequences when something is wrong.
This is a real risk and we've seen versions of it in other regulated industries. The mitigations are operational, not technical:
- The interface has to make the AI's confidence and reasoning visible, not hidden.
- The review step has to feel like review, not rubber-stamping.
- Edge cases — unusual HS codes, anti-dumping flags, novel products — need to route to the broker with elevated friction, not slipped into the same queue as routine entries.
- Audit trails need to be traceable through the AI step, not around it.
Get those wrong and you have a compliance bomb. Get them right and you have a broker whose statutory liability is better protected than it was on a manual workflow, because every decision is now logged.
The succession problem AI doesn't fix
One thing the DCN piece is clearly right about: AI does nothing for the long-term capability pipeline. The judgement that experienced brokers bring — on novel classifications, dispute handling, regulator relationships — is built over years of cases. If younger people aren't entering the profession to accumulate those years, no model will substitute.
Our honest view: AI buys the industry maybe a decade of capacity relief. What it doesn't buy is the next generation of senior brokers. That requires the boring, expensive work of cadetships, university pathways, and licensing reform. AI is the bridge, not the destination.
What we'd tell a brokerage today
Three things.
Start with the highest-volume, lowest-variance entries — repeat customers, standard goods, well-understood lanes. That's where AI extraction and pre-classification pays first.
Don't build. The off-the-shelf tooling for customs document parsing has caught up enough that custom builds rarely earn their cost for a single brokerage. The integration into your CMS and AOS — that's where the work is.
And keep your senior brokers in the loop on the design. The mitigations above only work if they're built by people who know what a real classification dispute looks like.
