Policies · AI Limitations
AI Voice Assistant — Limitations Policy
Our AI is helpful but it has limits. Anything it says about prices, timings, or rules is INDICATIVE — not a binding promise. Quotes, contract terms, claims decisions, and refusals always come from a human in writing. If the AI got something wrong, we fix it; we don't try to argue the AI's word was final.
What the AI can do
Answer general enquiries (services, routes, lead times). Provide tracking status from your booked shipments. Issue INDICATIVE rate ranges based on lane and mode. Take initial booking details and create a draft booking. Send you the customer portal link, quote PDF, or document. Transfer you to a human at any point. Recognise complaint / claim / legal / payment dispute and escalate. Schedule a callback with a named human. Confirm receipt of documents (e.g. customs paperwork). Read out a written quote a human has already issued.
What the AI explicitly cannot do
Issue a binding quote (pricing-final = human + written). Approve or commit to a booking (= human + written). Vary our standard T&Cs (= Director signature). Accept liability or settle a claim (= human, recorded). Decline service on KYC / sanctions grounds (= MLRO only). Make credit / refusal decisions (Article 22 + commercial sensitivity). Discuss other customers (confidentiality). Provide legal, tax, or customs-classification advice (routes to qualified human). Promise specific transit times beyond published indicatives (carrier-dependent — see /policies/service-level-agreement). Confirm whether a hazardous good is accepted on a specific carrier sailing (DGSA review). Authorise refunds (Accounts team, written). Process payments by phone (no card capture by AI — PCI scope avoidance). Issue a release of cargo against payment (Operations team, written). Discuss litigation or legal threats (Director, legal channel).
Indicative pricing — what it means
A figure the AI mentions is ALWAYS indicative. The AI uses the word 'indicative' before any price and follows up with: 'A Taurex specialist will confirm the binding quote in writing before we book.' If the human-issued written quote differs from the AI indicative, the WRITTEN QUOTE PREVAILS. Customer not bound to AI's number; nor is Taurex. This is not a disclaimer trick — it's how every freight quote works in practice (surcharges and carrier rates move daily).
Accuracy — best efforts, not guarantees
The AI is trained on Taurex's published pricing data, lane catalogue, surcharge schedule, and operations playbook. Reviewed regularly. HOWEVER the AI may cite an out-of-date surcharge, miss a recent rate change between data refresh and the call, not know about a one-off carrier disruption, misinterpret an ambiguous question, or 'hallucinate' a service or feature we don't offer. We accept this risk: every AI output is indicative; every call logged for audit; 1% of calls monthly sampled for accuracy review; errors fed back to vendor for improvement; we apologise and remedy when the AI got something wrong. Where AI gave clearly inaccurate information that a customer relied on to their detriment, we DO NOT hide behind the AI — we assess impact and remedy properly through /policies/complaints.
UK GDPR Article 22 — no solely automated decisions
The AI does not make a decision producing legal effects or significantly affecting the caller. Pricing: indicative; binding quote = human. Credit: refusal / approval = Accounts (human). KYC / sanctions: refusal = MLRO (human). Service refusal: human after review. Booking acceptance: human confirms in writing. Any extension to a decision engaging Art. 22 requires: lawful basis update, disclosure of meaningful logic, route to human review, fresh DPIA, customer notification.
Right to human review and contest
Right to human review of any AI output at any time. Acknowledge within 1 BD. Human review within 5 BD. Confirm outcome in writing. Apply human's decision retroactively if AI was wrong. Contact [email protected] or any service email. Right to contest: if AI's output influenced a Taurex decision and you believe it's wrong because of AI error, within 30 days email [email protected] with the call reference and reasons. We pull the call recording and transcript, review, respond within 14 BD. If we agree, we correct (honour corrected price, refund difference). If we disagree, we explain in writing — escalate per /policies/complaints.
Out-of-scope automatic transfer, no roleplay
Out of scope: specific legal advice, tax / customs classification beyond simple HS lookup, insurance coverage questions for individual policies (refer to broker), investment / financial advice, health, safety, or emergency situations (transfers, may dial 999), personal opinions, politics, contentious topics, other companies' offerings (we do not disparage competitors), counterparty disputes between you and your supplier or buyer. AI does NOT roleplay: pretend human, pretend named Taurex employee, pretend authority it doesn't have, roleplay scenarios caller proposes, generate unrelated content on demand. Caller's prompt-injection attempts → polite decline, repeated → human transfer.
Last reviewed 2026-06-06.