There is an obvious attraction in removing routine work from trade finance. Few people would argue that an experienced practitioner should continue spending time extracting dates or searching manually through documents for information that technology can identify almost instantly. If AI and other forms of automation can reduce that effort while improving consistency, experienced people should have more time available for the questions that actually require their expertise.
There is, however, a consequence that deserves rather more attention than it has received so far because the routine work we are understandably keen to automate has traditionally played another role. It was part of the process through which people learnt the business.
For generations of trade finance practitioners, expertise developed gradually. Someone starting in documentary credits did not arrive on their first morning capable of interpreting UCP or determining whether an unusual document satisfied international standard banking practice. They learnt by examining presentations and discussing difficult cases with more experienced colleagues, while repeated exposure to similar issues in slightly different circumstances gradually changed the way they approached the documents in front of them.
Much of that work was routine, but the routine provided the environment in which judgement developed.
This becomes important because knowing the rules has never been quite the same as knowing how to apply them. Someone can read UCP 600 article 14 and understand its provisions without necessarily being able to examine a complicated presentation and immediately recognise where the real issue lies. ISBP provides much greater detail about aspects of international standard banking practice, but applying that practice still requires an understanding developed through exposure to actual documents.
Over time, practitioners begin to recognise things they would not have noticed when they started. A document attracts attention before they can necessarily articulate why, or wording within a credit causes them to consider how it might operate once documents are presented. Information which looks innocuous to somebody less experienced may prompt them to look elsewhere in the presentation before reaching a conclusion.
Some of this knowledge is documented, but much of it is acquired through practice.
That distinction is becoming increasingly relevant to the wider debate about AI and employment. Research from the Stanford Digital Economy Lab, using payroll data covering millions of US workers through June 2026, found no evidence of widespread economy-wide displacement but did identify weaker employment outcomes among younger workers in occupations with high exposure to AI. Particularly interesting for trade finance is the researchers’ discussion of codified knowledge, which can readily be recorded and communicated, compared with tacit knowledge acquired through experience and repeated exposure to real situations.
Anyone involved in trade finance education should recognise the difference.
UCP can be recorded and taught, as can ISBP and internal procedures. Professional judgement develops differently because it depends upon understanding how that knowledge behaves when confronted with documents that were not created to provide a convenient training exercise.
Successful automation therefore creates an interesting challenge. If technology becomes increasingly capable of performing the routine work traditionally allocated to people at the beginning of their careers, today’s operations may become more efficient while one of the mechanisms through which tomorrow’s experienced practitioners develop is gradually weakened.
The answer cannot sensibly be to preserve unnecessary manual work merely because previous generations learnt that way. Technology has always changed how people work, and there is no particular educational virtue in requiring somebody to perform manually an activity which can be completed reliably by a machine. What needs to be preserved is not the work itself but the learning that occurred while undertaking it.
That requires us to think differently about training.
Trade finance education has understandably concentrated upon transferring knowledge. Practitioners learn the rules and work through examples showing how those rules apply. Qualifications provide an essential foundation, but judgement begins to develop when the carefully structured training example gives way to a real presentation in which the relevant issue is not conveniently identified in advance.
AI could weaken that process if it simply provides the answer before the practitioner has had any reason to think about the question.
Imagine a junior practitioner receiving a presentation which has already been examined by an automated system. Several potential discrepancies appear on the screen and the practitioner’s role is largely to review them before approving the result. The process may be considerably faster, but the person has not necessarily had to identify the relevant information or determine which rule applies. Nor have they necessarily considered why an apparent difference between documents is significant while another is not.
The danger is not that the person has used AI. It is that the technology may have removed the reasoning through which understanding would otherwise have developed.
Whether that happens depends largely upon how the system is designed. Instead of simply presenting a conclusion, technology can expose the reasoning behind it. The relevant document information can be shown alongside the credit requirement, with the applicable UCP provision or ISBP paragraph available for the practitioner to consider. Where an ICC Opinion or Briefing appears relevant, the user can examine it and decide whether the circumstances are sufficiently comparable.
Used in that way, AI begins to contribute to the learning process rather than bypassing it.
There is an important difference between this and suggesting that a system can replace an experienced mentor. The colleague who has spent twenty or thirty years examining documents possesses judgement developed through practice, much of which will never be completely represented within a database or model. What technology can do is make some of the accumulated knowledge surrounding that judgement easier for somebody else to access while developing their own.
In some respects, this could improve upon the traditional route into trade finance because access to good mentoring has never been uniform. Someone fortunate enough to join a team containing experienced practitioners willing and able to explain their decisions may develop very differently from somebody whose early work consists largely of processing transactions under time pressure.
AI-assisted systems could make explanatory material available much more consistently, provided they are designed to help the practitioner understand rather than simply to process more transactions.
A potential discrepancy, for example, need not immediately be accompanied by the system’s final answer. A training environment could require the practitioner to reach a conclusion first and then compare the reasoning with the automated analysis or with a previous decision. More complicated cases could become the basis for discussion with experienced colleagues, allowing technology to identify learning opportunities which might otherwise disappear into the normal flow of work.
This would move training closer to the operational environment rather than treating education as something that occurs separately from it.
There is already wider research interest in this question. Work associated with the Stanford Digital Economy Lab is examining whether generative AI enables apprentices to undertake more advanced tasks earlier in their development and whether those apparent gains affect the underlying understanding being acquired. The question has obvious relevance to any occupation in which professional capability depends upon experience rather than knowledge alone.
Being able to complete a task with AI is not necessarily evidence that somebody understands the task sufficiently well to perform it without AI or to recognise when the technology has made a mistake.
That distinction could become particularly important in documentary trade because an answer often looks straightforward once somebody else has identified the problem. A discrepancy can appear obvious after it has been highlighted, while the expertise lies partly in recognising that there was something requiring attention before anyone pointed towards it.
As automation develops, the role of experienced practitioners may therefore need to evolve alongside that of their junior colleagues. If less senior staff spend less time performing routine examination, experienced colleagues may need to become more deliberate about creating opportunities through which judgement can develop. Mentoring becomes part of the operating model rather than something that happens naturally because two people happen to be examining the same presentation.
This also raises questions about how organisations measure the success of automation. If the principal measure is the number of transactions processed with progressively less human involvement, there will be a powerful incentive to remove people from as much of the workflow as possible. The same organisation may nevertheless expect to have experienced specialists available years later when a difficult transaction requires judgement that cannot readily be automated.
Those objectives need to be considered together. A bank can acquire technology relatively quickly, whereas replacing decades of accumulated experience is considerably more difficult. As experienced practitioners leave the industry, organisations need to consider not only how their knowledge can be retained but how somebody else develops the ability to interpret it.
AI may actually help with both aspects of the problem. Decisions and commentary accumulated over many years can become more readily searchable, allowing previous experience to remain accessible after the individuals involved have moved elsewhere or retired. A practitioner encountering an unusual issue years later may be able to retrieve relevant internal reasoning alongside the applicable ICC material, giving them access to knowledge which might otherwise have disappeared.
There will inevitably be limits to how much can be captured because professional experience includes intuition developed through repeated exposure to situations that were never formally recorded. This makes preserving human expertise more important as automation increases, while also giving us an opportunity to think more deliberately about how that expertise is passed on.
The objective should not be to protect junior practitioners from AI, any more than it should be to retain manual processing for sentimental reasons. It should be to redesign professional development for an environment in which technology performs more of the routine work through which people once learnt.
That requires systems that allow practitioners to understand how conclusions were reached and training that deliberately exposes them to difficult cases rather than relying upon those cases appearing naturally during daily processing. Experienced practitioners will have an important role in that process because technology can retrieve knowledge without necessarily possessing the judgement required to understand its significance.
The wider employment evidence suggests that questions of this nature are already emerging outside trade finance. Stanford’s research found the clearest employment weakness among younger workers in occupations exposed to AI, with much of the adjustment appearing through reduced hiring rather than widespread dismissal. More experienced workers did not show the same pattern, while outcomes differed according to whether AI was principally being used to automate human activity or complement it.
It would be premature to translate those findings directly into predictions for trade finance, and the researchers themselves are careful not to claim that the patterns establish causation. They nevertheless highlight something our industry should consider before automation becomes sufficiently embedded for the consequences to be difficult to reverse.
If fewer people need to perform the work traditionally undertaken at the beginning of a trade finance career, we need another route through which they can acquire the experience required later in that career.
There is no reason for that to become an argument against AI. It is instead an argument for thinking about the development of people at the same time as we redesign the process around technology.
The best outcome would be an environment in which routine activity no longer consumes time unnecessarily, whilst the technology itself helps practitioners understand the more difficult work that remains. If we can achieve that, AI may not only improve the efficiency of trade finance but also help preserve the expertise upon which the industry will continue to depend.
