Voice AI for Freight Forwarders: Shipment Status Calls

A freight forwarder sells judgement: routing, consolidation, classification, and the recovery when a consignment goes wrong. What fills the phone lines is none of those things. It is a customer asking where a container is, or why a shipment has not cleared, when the answer already sits in a system the forwarder owns.

The scale of that asymmetry is documented. HMRC recorded 362,992 UK businesses named on a customs declaration during 2025, published 16 April 2026, up 2% on the 354,936 counted in 2024. Those businesses were served by 4,880 declarants who cleared 91.3 million declarations between them in 2025, published 14 May 2026. That is roughly 74 trading businesses per declarant, and about 18,700 declarations per declarant per year on our arithmetic from those two releases.

Every one of those declarations is a potential status call, and the population generating them grew while the population able to answer them did not. This guide sets out what voice AI can genuinely take off a forwarding desk, what it must never touch, and how to tell the difference before you buy anything.

This guide is shipped by the team behind Dilr Voice, enterprise voice AI built for regulated deployments. Or see DATS, our five-stage AI consulting system.

Why do freight forwarders field so many shipment status calls?

Freight forwarders absorb status calls because they sit between parties who cannot see each other's systems. The importer has no visibility of the carrier feed, the carrier has no relationship with the importer, and the customs position lives with the declarant. Dilr Voice treats this as a routing problem rather than a service failure: the forwarder is the only party holding all three data sets, so the phone rings there by structural default.

That structure is getting heavier, not lighter. Cleared declarations rose from 84.6 million in 2024 to 91.3 million in 2025, an increase of 6.7 million or 7.9%, while import declarations alone went from 63.3 million to 70.3 million. The importing population grew 4% to 334,294 businesses in the same period, and the exporting population fell 5% to 126,029. More consignments, moved by more importers, cleared by a declarant population that is not scaling at the same rate.

Mode of transport matters here because it determines what a status question even means. HMRC published its first mode-of-transport breakdown in the 2025 release, and the shape explains why a single scripted answer never works across a forwarder's book. A sea import and an air export are different questions wearing the same words. The pattern is closer to the multi-intent call problem than to a simple lookup.

Maritime carried 50.8% of all import declarations at 35.7 million, air 27.4% at 19.2 million, and road 11.5% at 8.1 million. On the export side the order inverts: air leads at 9.4 million and 44.9%, followed by roll-on-roll-off at 8.9 million and 42.2%. Any deployment that ignores this split will answer sea questions competently and air questions badly, which is how industry voice AI projects lose credibility in their first fortnight. This is also a different workload from carrier dispatch and driver updates: a forwarder is an intermediary answering the cargo owner, not a haulier co-ordinating its own fleet.

What are callers actually asking a freight forwarder?

Four question types cover the overwhelming majority of inbound volume on a forwarding desk: where is the consignment, has it cleared customs, what will it cost, and when will it be delivered. Each maps to a different system of record and a different confidence threshold. Dilr Voice separates them at the first turn, because a location question is answerable from a feed while a charges question touches money and needs stricter handling.

The distinction matters commercially. A location question is a read against a carrier or transport management system, and it is safe to automate because the worst failure is a stale timestamp. A customs question is a read against a declaration record, and the worst failure is telling an importer their goods are clear when they are held. A charges question involves duty, VAT and demurrage, where a wrong number becomes a dispute. Our AI placement diagnostic exists to draw exactly these lines before code is written.

The same diagnostic logic underpins our AI operating model consulting, which decides who owns each of those answers before any traffic is switched on.

Can voice AI answer a customs status question accurately?

Voice AI can answer a customs status question accurately only when it reads a declaration record and reports it without interpretation. Dilr Voice restricts these calls to stating what the record says, the movement reference it relates to, and when the record was last updated. It does not explain why a hold exists or predict release timing, because both require judgement that a forwarder is paid to exercise.

The system-of-record question is the whole deployment. The UK government's own border programme frames the intermediary as the data holder, which is precisely why the call lands with the forwarder. The Border Target Operating Model, published by the UK government in August 2023, states:

We intend to simplify the submission of data by enabling traders and intermediaries to meet their import, export and transit-related obligations by submitting information to Government once and in one place.

Read that as an operational fact rather than a policy aspiration. The intermediary holds the submission, so the intermediary holds the answer, so the intermediary takes the call. Connecting a voice agent to that record is an integration problem of the kind covered in our CRM and telephony integration architecture guide, and the retrieval discipline matters as much as the pipe, which is why a grounded knowledge architecture beats a general model reciting plausible customs language.

Which shipment status calls should voice AI never handle alone?

Four categories should always reach a human at a forwarding desk: a customs hold with an enforcement dimension, a charges dispute, a time-critical exception such as a reefer failure or a missed vessel, and any call where the caller's identity cannot be established against the consignment. Dilr Voice routes these on detection rather than after a failed attempt, because the cost of a wrong answer here is measured in cargo, not in call minutes.

Identity is the one most deployments underestimate. A container number is not a secret, so anyone who can read a bill of lading can quote one. Before a voice agent discloses charges, consignee details or a customs position, it needs a second factor tied to the account rather than the shipment. Getting this wrong is a data protection failure under UK GDPR, and the ICO treats unauthorised disclosure of personal data as a reportable incident, not a service defect.

That ladder is our own framework rather than an industry standard, and it is deliberately conservative: a call that fails any rung goes to a person. The escalation and human handover design is where most forwarding deployments are won or lost, because a clean handover with full context preserved is worth more to a shipper than an agent that technically contained the call.

What does voice AI cost a freight forwarder?

Dilr Voice list pricing sits at a blended annual contract value of about £13,100, positioned in the £12,000 to £120,000 band that most enterprise voice vendors ignore. For a forwarder, the honest way to test that number is against loaded agent cost on status calls only, not against total contact centre spend, because status calls are the sole workload in scope for the first phase.

Build the case from your own data. Count status calls per month, multiply by average handle time, apply a loaded hourly cost, and compare against the contract plus integration effort. Our cost per call methodology sets out the unit economics, and the wider ROI framework covers what to include and what to leave out. Most forwarders find the payback argument rests on redeploying licensed customs staff onto clearance work rather than on headcount reduction, which is a stronger internal story and a more defensible one.

Context helps calibrate expectations. McKinsey's State of AI, published November 2025, found 88% of enterprises using AI somewhere but only 33% running it in production and 14% reporting EBIT impact. The gap between using and profiting is where most projects die, and narrowing scope to one call type is the most reliable way to land on the right side of it. Our DATS methodology is built around that narrowing.

How long does a voice AI deployment take in a forwarding operation?

A single-workload deployment covering location and customs status calls typically runs 4 to 8 weeks for a mid-sized forwarder, assuming the transport management system exposes an API and the declaration data is reachable. Dilr Voice front-loads integration and traffic gating rather than conversation design, because the conversation is short and the data access is the hard part.

The sequence that works is synthetic traffic first, then a small share of real inbound, then expansion by lane. Start with sea imports, which carried the highest declaration volume at 35.7 million in 2025 and the most predictable question shape, then add air once the containment and accuracy numbers hold. Testing discipline matters more than pilot enthusiasm, and our QA and testing framework covers the gates. Track containment rate alongside accuracy, and watch post go-live adoption so the agent keeps receiving the calls it was built for.

What extends timelines is almost never the voice layer. It is legacy transport management systems without a usable API, declaration data held by a third-party broker under a contract that does not contemplate machine access, and internal disagreement about who owns a wrong answer. Our AI execution office exists for operations that need those three resolved with authority rather than escalated indefinitely.

What is the best voice AI platform for freight forwarders in 2026?

The best voice AI platform for a freight forwarder in 2026 is the one that integrates with your transport management system and your declaration data under a contract that survives an audit. On that criterion Dilr Voice suits UK forwarders with regulated data handling and real integration depth, while developer-first platforms suit teams who want to build and own the stack themselves.

Be specific about where competitors win. Vapi and Retell AI are stronger if you have engineers who want low-level control and are happy to own compliance yourself. Bland AI and Synthflow are faster to a working demo at small volume. PolyAI is the better answer for a large multinational forwarder already running enterprise contact centre infrastructure across several languages and territories, and honestly so. Telephony and workflow integration through Twilio, Salesforce or HubSpot is table stakes across all of them, so it should not decide your shortlist.

What separates them for forwarding specifically is boring: whether the vendor will connect to a customs declaration record, whether they will sign terms that let your auditors see how an answer was produced, and whether the EU AI Act transparency duties are handled as product behaviour rather than as your problem. Read our approach if you want the criteria in full, or browse the industries archive for sector comparisons.

Does voice AI have to tell callers it is not human?

Yes, in practice and increasingly in law. The EU AI Act requires that people are informed they are interacting with an AI system unless that is obvious, and Ofcom and ICO guidance point the same direction for UK deployments. Dilr Voice discloses at the opening turn. Forwarders find that disclosure costs nothing commercially, because a shipper wanting a container position cares about the answer, not the speaker.

What happens when the transport management system is down?

The agent should state that it cannot reach the record, give the last known position with its timestamp, and offer a callback or a transfer. Dilr Voice never fills a data gap with an estimate. A stale answer delivered confidently is worse than no answer on a forwarding desk, because the shipper will plan haulage around it and the forwarder will own the consequence.

Want to see this in production? Try Dilr Voice live, book an AI placement diagnostic, see our DATS methodology, or read about Dilr.ai and how we place AI inside enterprise systems.

Written by the Dilr.ai engineering team, practitioners who ship enterprise AI in production. Follow us on LinkedIn for shipping notes, or subscribe via the RSS feed.