US AI Freight Drafts: Review Numbers Before Sending Counteroffers
Check the meaning of the number before the multiplication
A US chartering team reviewing an AI-drafted counteroffer should verify each number's meaning before checking its arithmetic. A correctly calculated total can still be wrong for the deal if it uses the wrong quantity, unit, currency or service scope. The review should connect the proposed commercial position to the source instruction and the authority to communicate it.
NIST's generative-AI profile identifies confidently presented erroneous content as a risk. NIST: Generative AI Profile.
A plausible explanation from an AI tool therefore does not establish that the underlying freight calculation is supported. The system may have selected an obsolete quantity or treated an internal target as an approved counteroffer. The reviewer needs a traceable basis and a reproducible calculation, not reassurance that the draft sounds commercially fluent.
Attach units and scope to every commercial input
NIST's SI guide distinguishes a quantity's numerical value from the unit used to express it. NIST: Expressing Values of Quantities.
For a freight review, the original practical application is to keep the quantity basis explicit. A cargo weight, package count, volume and contractual freight quantity are not automatically interchangeable. If a quote uses a particular charging basis, the draft must preserve that basis rather than selecting whichever cargo number is easiest to extract.
Currency belongs beside the amount, and a rate needs its denominator. A figure entered as 40 without context could be a freight rate, a percentage, a time allowance or a quantity. The draft should not infer its role from proximity to another number when the source is ambiguous. Ask the responsible person to clarify before the figure becomes part of an outward proposal.
Identify the included services separately from the rate. If the counteroffer changes a handling obligation, a price comparison must reflect that change. The review should also distinguish a proposed concession from a confirmed amendment. Labeling a new number as agreed because it appears in the latest internal message can misrepresent the state of the negotiation.
Reproduce a proposed Gulf Coast counteroffer
Consider a hypothetical Gulf Coast export enquiry for 20,000 metric tonnes. An owner proposes USD 40 per tonne, giving USD 800,000 freight. The cargo interest authorizes the broker to explore USD 39 per tonne on the same cargo and service basis. That proposed figure produces USD 780,000, a USD 20,000 reduction. These are fictional inputs.
Suppose an AI draft instead uses a later internal planning quantity of 22,000 tonnes and states that USD 39 per tonne totals USD 858,000. The multiplication is correct, but the quantity is not the authorized basis for this counteroffer. The reviewer should not fix the problem merely by changing the displayed total. The entire draft must be reconciled with the intended quantity and instruction.
Now suppose the draft retains 20,000 tonnes but omits an excluded service estimated at USD 30,000. The comparable planning cost becomes USD 810,000 when that service remains the cargo interest's responsibility. Describing the proposal as a saving against an inclusive USD 800,000 alternative would be misleading. The rate is lower, but the complete scope is more expensive by USD 10,000.
Finally, imagine the draft calls the USD 20,000 rate-only reduction a ten-percent saving. The correct percentage relative to USD 800,000 is 2.5 percent. This is a separate error from the scope issue. A good review checks the calculation and the commercial comparison rather than assuming that correcting one makes the entire recommendation reliable.
Review the source instruction and the outward action
The reviewer should be able to identify which instruction authorizes the proposed rate, which cargo description it applies to and which conditions remain attached. A numerical limit may depend on a particular window or service allocation. If the draft removes that dependency, the counteroffer can exceed authority even when its price remains within an approved range.
Check whether the document is an internal analysis or a message intended for the counterparty. An internal sensitivity table may contain negotiating limits that should not be included in an external proposal. The workflow should not send the entire analysis merely because the generated response refers to it. Recipient selection and attachment review remain part of the commercial approval.
A corrected draft should be reviewed in its final form. If another user or automated step changes the quantity, rate or conditions afterward, the earlier approval should not be treated as covering the new text without the organization's required review. The system needs to make that change apparent to the people responsible for sending it.
Test the workflow with deliberately difficult cases
A useful pilot includes mismatched units, an obsolete quantity, two currencies, a conditional concession and a lower rate with a new exclusion. Ask reviewers to identify the problem and explain its effect. This tests whether the workflow supports commercial understanding rather than merely producing correct calculations when the input is perfectly clean.
Keep numerical verification independent enough to expose an error. Repeating the same generated explanation is not a strong check. The organization can use an approved calculation method and source review appropriate to its work. The aim is to reproduce the result from the actual inputs and determine whether those inputs belong in the proposed transaction.
The YC-inspired startup approach described by Zeaclub is useful here as a product-development attitude: focus on a specific user problem and improve from observed failures. It is not evidence of YC affiliation or a claim that a proposed AI workflow is already reliable. A narrow counteroffer-review pilot can produce valuable learning without granting software negotiation authority.
Bring representative freight drafts and the errors your team most needs to catch to Zeaclub to discuss a proposed review workflow. For US brokers and cargo interests, the practical objective is an externally approved counteroffer whose numbers, conditions and recipients have been checked by an authorized person who understands the deal.
Sources
- NIST: Generative AI Profile, checked October 24, 2026; confabulation-risk scope only, aggregated with other AI600-1 uses.
- NIST: Expressing Values of Quantities, checked October 24, 2026; number-and-unit distinction only, aggregated with underlying SP811 if reused.
Related articles
- 0762: US Freight Counteroffers: Show Every Change in the Commercial Package
- 0780: US Brokerage Digital Adoption: Keep Human Negotiation Authority Visible
- 0995: US Voyage Estimate Audit: Reproduce the Number Before Approving


