The pitch arrives before the plan
If you run ocean freight operations, the AI pitches have already reached you. Vendors describe platforms that automate documents, manage bookings and send milestone alerts. The demonstrations look convincing. What they rarely answer is the question your team actually faces: of the work we do all day, which parts are worth handing to software, and which parts should never leave our people?
Getting this wrong in either direction is expensive. Automate a judgement call and you ship wrong amendments at machine speed. Automate nothing and your operators stay buried in rekeying while competitors move faster.
A useful starting point comes from a vendor’s own commentary. When WiseTech Global’s chief executive Zubin Appoo discussed AI in supply chain, he argued that the winners will not be the companies with the most advanced models, but those with the deepest operational data, the strongest domain expertise and the most connected ecosystems. Interface-level tools, he said, can reduce friction, but transformative AI asks whether the workflow should exist in its current form at all. Whether or not you use that vendor’s platform, the test is a good one: evaluate the process and its data before the technology.
Four questions to score a candidate workflow
Take one recurring piece of work and score it against four questions.
Is the input available as data? AI cannot compare a returned booking confirmation against an approval if both live in scattered emails and screenshots. A workflow with a clear record, like the amendment record described in our article on booking amendment record tracking, is a candidate. A workflow whose inputs are half-remembered phone calls is not, yet.
Does it repeat with the same shape? A cargo-ready date change follows the same pattern every time: new ready date, equipment position, sailing options, approval, revised confirmation. Repetition with a stable shape is exactly where assistance pays. One-off commercial negotiations are the opposite.
What does an error cost? A mis-sorted internal summary costs minutes. A booking amended with the wrong container count, or a bill of lading corrected against the wrong revision, costs money and customer trust. Cost rises with the step, so the expensive steps need the most human control, not the least.
Can a human check the result quickly? Assistance works when reviewing the output takes less time than doing the work. If the operator must re-verify every field from scratch, nothing was saved.
Where AI assists, and where it must not decide
For ocean freight booking changes, the honest split looks like this. Software can gather the request, the approval and the returned confirmation into one place, flag field-level mismatches, and keep the record searchable. Platform vendors document exactly this kind of capability: CargoWise, for example, describes workflow templates configured by mode or customer, automated task assignment with clear ownership, milestone alerts and exception logs for responding quickly to deviations. Those are real, documented features of a real platform.
What stays with people is the decision layer. Approving a revised sailing with its cost basis. Deciding whether a mismatch on the returned confirmation is acceptable or must be disputed. Deciding whether an equipment substitution is equivalent. Our amendment request checklist exists precisely because the human check before submission is what makes the rest of the workflow safe. The pattern that works is assembly and comparison by software, approval and submission by a person, and a returned-confirmation check that names the mismatch rather than silently fixing it.
Example
A ten-person forwarding office handles around forty booking amendments a week across three carriers. Each amendment currently takes an operator roughly twenty minutes of hunting: finding the original request, the approval message, and the returned confirmation, then comparing them line by line. Amendments that arrive while the file owner is away sit longer, and mismatches surface weeks later as charge queries.
Before buying anything, the operations manager scores one workflow — the standard container count change — against the four questions. The inputs are available: requests and confirmations arrive as emails and portal PDFs with stable fields. It repeats constantly. Errors are costly downstream, so the approval and submission steps stay human. And the comparison step, the ten minutes of line-by-line checking, is exactly where a reviewed output is faster than manual checking.
The resulting design: software assembles the three documents, extracts the fields and highlights every difference against the approved change; an operator confirms or rejects each difference in a couple of minutes and submits through the carrier’s own amendment flow, following the carrier’s documented process. Nothing is submitted automatically. The office’s measure of success is simple: amendments closed with a checked confirmation, and charge queries that can be answered from the record in minutes instead of an afternoon.
Start with one workflow, not the whole office
The vendor commentary is right on one point even sceptics should accept: the value sits in the operational process and its data, not the model. So begin with a single workflow your team already performs consistently, make its record explicit, and add assistance where the four questions score well. The same sequence shapes our proposed ocean booking rescheduling workflow: start with one recurring booking-change process, review where the record and the checks actually consume the team’s time, and build the assistance around the approvals that must stay human.
If you have a workflow like this — a booking change, a document correction or a record-keeping task that eats your team’s week — Let’s talk. Bring one recent booking change, its request, approval and returned confirmation, and we will review where the work actually goes and where the useful line between software and your people sits.