A freight enquiry arrives as an email, a spreadsheet and a forwarded message with a different weight. The commercial team spends time determining what the customer actually wants before it can consider a rate. AI can help organise those inputs, but a neat table is not proof that the shipment details are correct.
For a Dubai freight forwarder, a sensible first assistant prepares an RFQ record for staff. It extracts stated facts, preserves units, flags contradictions and identifies missing information. It does not choose a carrier, commit a rate, classify goods or promise customs clearance. This is preparation for a quotation, not autonomous freight execution.
Define the quotation-ready record
Ask the operations and commercial teams which fields they genuinely need for the selected service. Origin, destination, mode, cargo description, package information and requested timing may all matter, but the exact checklist should come from your workflow.
Distinguish customer-supplied information from an assumption. If the sender did not specify an incoterm, the assistant must not select one. If a destination is only a city, do not turn it into a precise port or delivery address without verification.
Start with one class of RFQ handled by one team. Mixing air, sea, road, project cargo and specialist goods into the first pilot makes error review much harder. A narrow checklist is easier to test than a universal freight expert prompt.
Preserve units instead of normalising silently
A model may convert a dimension or weight to make a table consistent. That can conceal a mistake. Keep the original stated value and unit alongside any calculated value, and require staff review before using a conversion commercially.
If a source says gross weight and another says chargeable weight, do not treat the numbers as interchangeable. The assistant should flag that distinction for the freight team rather than decide the pricing basis. It must also avoid filling missing quantities with a plausible default.
Use field-level source references. Staff should be able to see which attachment or message supplied a critical figure without re-reading the entire thread. An unsupported value in an otherwise tidy record is an error, not a convenience.
Make contradictions a useful output
The assistant's most valuable contribution may be a short list of questions. If the email and spreadsheet disagree, mark both values and ask staff to resolve them. Do not prefer the most recent-looking file unless the workflow has a reliable rule for identifying the authorised version.
Keep attachment versions and customer clarifications connected to the same RFQ reference. A follow-up message should update the reviewed record or raise a task, not create an unrelated opportunity that another salesperson also quotes.
A useful draft says “package count differs between the email and attachment” rather than “validated shipment details.” The latter implies a level of checking the assistant has not performed.
Separate extraction from routing and compliance
A cargo description is not a customs classification. A destination country is not proof that the proposed service is permitted. Dangerous goods, controlled items, sanctions, documentation and customs questions need the forwarder's qualified processes and current authoritative sources.
The assistant may flag a phrase for review, but it should not certify that goods are safe, unrestricted or properly classified. Do not have a general model invent a commodity code or approve a declaration from an incomplete description.
Likewise, an enquiry mentioning Jebel Ali does not establish every local port or handling requirement. Your team must verify the actual route and service. This guide deliberately does not prescribe customs rules or carrier acceptance conditions.
Keep rates and commitments with authorised staff
Even a complete RFQ does not establish a valid rate. Carrier terms, surcharges, capacity and validity may change. The first assistant should prepare the information needed for a commercial review, not choose a margin or promise a transit time.
If a later phase retrieves rates from an approved system, show the source, validity period and exclusions. A stale rate should produce a review flag, not a confident quotation. The person approving the offer remains responsible for the final commercial content.
Do not grant booking or payment authority as an incidental part of an RFQ pilot. Creating a shipment, accepting terms or issuing a purchase instruction is a separate externally consequential action with its own controls.
Protect customer and trade information
An RFQ may contain personal contact details, delivery addresses, invoices or commercially sensitive descriptions. The UAE government's data-protection overview explains personal-data controls and cross-border sharing. Commercial contracts may require additional restrictions.
Map where email content, attachments, extracted fields and logs are processed. Give the assistant only the access needed for the agreed task. A customer-intake workflow should not have unrestricted access to financial records or every company's shipment history.
Use synthetic RFQs for demonstrations, including realistic formatting errors without real consignee information. Do not upload a confidential bill of lading simply because a vendor's demo performs better on an authentic document.
Draft clarification questions without changing the offer
A staff-reviewed clarification can name the missing field and explain what the team needs next. It should not imply a quotation has been accepted or a shipment slot reserved. Keep the question concise and grounded in the record.
For Arabic-English requests, verify units, location names and technical terms with competent staff. The bilingual-reply guide describes a review process. Language fluency is not freight expertise.
The WhatsApp Business Messaging Policy and local marketing requirements matter if the chosen channel is WhatsApp. A freight enquiry is not automatic permission for unrelated promotional campaigns. WH Digital's initial approach excludes cold WhatsApp, SMS and calls, with people approving outward messages.
Run a difficult document test
Include mixed units, a scanned attachment, conflicting package counts, an unknown destination abbreviation and a revised RFQ after a draft was prepared. Check whether the assistant retains uncertainty and links every important field to its source.
Measure extraction corrections, duplicate records, unresolved questions and staff review time. Do not turn a reduction in retyping into a guaranteed margin improvement. If a structured RFQ form would solve most of the problem, consider that simpler route first.
The 2,500 AED AI Ops Audit maps one intake workflow and its dependencies. The free check requires no shipment documents. A useful first assistant produces a reviewable set of facts and questions—not a freight commitment that nobody authorised.
The AI Ops Audit is a 2,500 AED review of your workflow, with a prioritised plan. No guaranteed outcomes. Implementation is separate.
Explore the audit