US Tanker Enquiry Extraction: Preserve the Quantity Option Before Searching
The option is part of the cargo instruction
A tanker enquiry can contain a firm quantity, an optional addition, and a deadline for deciding whether to exercise that option. An extraction tool that returns only one quantity has already made a commercial choice. For a Houston desk using AI to prepare enquiries, preserving the relationship among those terms matters more than producing a clean looking row in a spreadsheet.
The correct objective is to represent what the sender actually instructed, including what remains undecided. The extracted record should distinguish the committed parcel from the possible addition and identify who may decide. If the message does not establish that authority or deadline, the output should flag the omission. It should not complete the missing commercial terms from a familiar pattern in earlier enquiries.
NIST's generative AI profile identifies confident but erroneous outputs as a risk. In cargo extraction, a plausible invented option term can be more dangerous than a visible blank. NIST generative AI profile.
Follow a quantity through a realistic message sequence
Consider an invented US Gulf product enquiry. The first message asks about a firm 30,000 metric tons with a possible additional 5,000 metric tons. A later message says the customer no longer needs the addition, but the sender wants to know the freight consequence of reducing the minimum. A third message asks whether a proposal would still work if the supplier restored part of the optional volume.
These messages do not necessarily establish a new final cargo instruction. They may describe an original enquiry, an amendment, and a subsequent exploratory question. Treating the newest number as the current committed quantity could misrepresent the principal's position. Treating every number as cumulative could produce an even larger error.
A useful extraction view would preserve the firm instruction separately from the scenario being explored. It would show the source text for the change and ask whether the third message revises the enquiry or merely requests a sensitivity. That question has a direct commercial purpose: the broker needs to know what quantity to represent when seeking or negotiating an offer.
Extract the relationship, not just the number
The quantity field needs its unit, status, and condition. An amount in metric tons should not silently become barrels. A supplier estimate should not become a charterer's firm commitment. A range should not become its midpoint simply because the downstream matching tool accepts one number.
The same principle applies to dates. An option decision deadline, cargo readiness date, and preferred loading interval are different concepts. A system may recognize all three dates perfectly and still produce a commercially wrong enquiry if it assigns them to the wrong fields. Evaluate the relationship between the extracted facts, rather than checking whether the characters were copied correctly.
When a message refers to an attachment, preserve that dependency. If the attachment is unavailable, the extractor should say that the relevant condition could not be read. A polished summary that omits the missing attachment can create false confidence precisely because nothing appears visibly incomplete.
Show how an extraction error changes the comparison
Use a hypothetical freight proposal with a minimum payment of $600,000 covering the first 30,000 metric tons and an additional charge of $18 per metric ton above that quantity. If the optional 5,000 metric tons is exercised, the extra freight is $90,000 and the total is $690,000. These are invented terms for illustrating the extraction problem.
If a tool incorrectly labels all 35,000 metric tons as firm, the desk may seek proposals for a commitment it has not made. If it discards the option, the desk may compare the $600,000 base payment with another offer that includes capacity and terms for the larger parcel. Neither error is merely cosmetic. Each changes which commercial alternatives appear relevant.
The correct record should support two clearly labeled cases while preserving the principal's authority to choose. It should also state whether the proposed incremental terms have actually been offered or are only an analyst's assumption. A calculation can be arithmetically correct and still mislead when its contractual basis is invented.
Design review around the changed meaning
A reviewer should see the source passage beside the extracted relationship. Highlight a change from optional to firm, a change in who can exercise the option, or a change in the deadline. Those differences deserve more attention than a corrected spelling or a reformatted date.
Ask reviewers to resolve specific questions. Does this message replace the previous instruction? Does the smaller quantity remain acceptable to the buyer? Is the extra amount still subject to supplier confirmation? A generic request to approve the summary encourages a quick reading without identifying the commercial ambiguity.
ASBA's ethics code emphasizes accurate handling of proposals by member brokers. An AI-assisted enquiry should preserve that accuracy when transmitting the principal's quantity instructions. ASBA Code of Ethics.
Test before connecting extraction to external action
Build a test set with authentic patterns that the desk is authorized to use, with confidential details removed where appropriate. Include withdrawn options, forwarded historical messages, revised attachments, and a sender asking a question without changing the instruction. Record the expected interpretation before reviewing the model's answer.
Measure false changes to commercial meaning separately from missing fields. A tool that returns many complete fields but repeatedly converts questions into instructions is not ready for unattended use. Also test whether the interface lets a person correct the relationship without destroying the link to the original message.
Zeaclub's agentic shipbroking ambition can be grounded in this kind of narrow, valuable improvement: preserving the cargo's commercial meaning while reducing repetitive entry. That is a development direction, not a claim of a deployed extraction feature. Bring a difficult enquiry sequence to a Zeaclub discussion and define the quantities, options, and amendments that a useful system must keep distinct before any external enquiry is sent.
Sources
- NIST, Generative Artificial Intelligence Profile. Narrow confabulation risk principle; checked October 23, 2026.
- ASBA, Code of Ethics. Accuracy principle for member brokers; checked October 23, 2026.
Related articles
- 0469: US LNG Enquiry Automation: Keep Conditional Cargoes Out of the Firm Supply Column
- 0430: AI Tanker Matching in the US Gulf: Define What a Match Actually Means
- 0470: US LNG Contract Data: Extract the Option Holder, Not Just the Word Option


