Skip to content
All insights

Industry operations

In logistics, the same question arrives a thousand times a day.

Every shipment generates questions. Where is it, when will it arrive, send me the paperwork, something has gone wrong. Three of those four are lookups, and they consume the people who should be handling the fourth.

Published July 14, 20265 min read

What this covers

  • Status and document requests are high-volume lookups that arrive in unstructured form, which is exactly the shape agents handle.
  • Routing accurately is often the largest single improvement, and it is lower risk than answering.
  • Exceptions are where the service actually differentiates, and they should be escalated with context rather than answered.
  • Answers must come from live shipment data, because a stale status answer is worse than no answer.

A logistics provider's service desk is a volume business inside a volume business. Every shipment generates inbound traffic, most of it asking for something the system already knows, and none of it arriving in a form the system can read.

The structural problem is that the questions arrive as email, in several languages, addressed to a shared mailbox, referencing a shipment by whatever number the sender happens to have: your reference, their reference, a container number, a purchase order, or a description of what was in the truck.

The four categories

CategoryShare of trafficWhat it needs
Status and ETAThe largest single blockA lookup across the shipment record and the carrier
Document requestsConstantRetrieval from the document store, and sending it
Change and booking requestsSteadyPreparation, then a person's approval
Exceptions, complaints, claimsThe smallest, and the most importantA person, with the full picture assembled

The first two categories are almost entirely mechanical and they are consuming the capacity that should be going into the fourth. That is the argument in a sentence: automating the lookups is not primarily a cost saving, it is how the exceptions get handled properly.

Start with routing

The safest and often largest first step is not answering. It is classifying: identifying what each request is about, which shipment it refers to, and which desk owns it, then putting it there with the record attached.

This removes the triage hour and the overnight queue without the agent sending anything to a customer. The one published figure we have on this site is from exactly this workflow: a European logistics company measured 40% faster ticket routing, where the agent identifies the intent of each incoming request and routes it into the right operational workflow. The customer name is confidential, and the figure describes their operation rather than being a benchmark for anyone else.

Answering status properly

When the agent does answer, the quality difference is entirely about whether it described the real state or filled a template.

A template answer says the shipment is in transit with an ETA of Thursday. A real answer says the shipment cleared customs on Tuesday, two of the three pallets are with the line-haul carrier under this reference with an ETA of Thursday morning, and the third was held for a documentation query that was resolved yesterday and is now following on Friday.

The second answer takes the same three seconds to produce and prevents the two follow-up emails the first one guarantees.

Documents, which are pure retrieval

Proof of delivery, bill of lading, customs paperwork, packing lists, certificates. These requests have a factual answer, the document exists, and the entire cost is somebody finding it and attaching it to a reply.

This is the least controversial automation in the category and usually the fastest to show value, because there is no judgment involved at any point and the customer experience improvement is immediate.

What must reach a person

  • Anything that is a claim, or that could become one.
  • Damage, loss, or a delivery that did not happen.
  • Any request that involves a cost concession or a credit.
  • A change to a booking that is already in motion, which should be prepared and then approved rather than executed.
  • Any message where the tone indicates a customer is escalating, whatever it is about.

The last one is worth designing deliberately. An agent that recognizes frustration and hands the message to a person immediately, with the full shipment history assembled, is providing better service than one that answers the literal question accurately.

Measuring it

  1. 01First response time from the customer's timestamp, not from when a ticket was opened.
  2. 02Share resolved on the first reply, which predicts follow-up volume better than any other number.
  3. 03Share of traffic that never required a person, split between routed-only and answered.
  4. 04Reopen rate, as the check on whether speed came at the cost of completeness.
  5. 05Time the exceptions desk spends on exceptions rather than on lookups.

The fifth is the one that justifies the project internally. AI agents for logistics sets out the workflows, and customer operations automation covers the general pattern across industries.

Questions

Questions we get about this.

How does it know which shipment a message is about?

By resolving whatever reference the sender used against the shipment record: your reference, their reference, a container or purchase order number, or a description of the goods. That resolution is the step that makes automated handling possible, and it is what keyword matching cannot do.

Can it answer in the customer's language?

Yes. Reading and replying across languages is normal rather than a separate capability, which matters for shared mailboxes receiving mail from several markets. Your escalation rules and tone guidance apply the same way in each.

What happens with a damage claim?

It routes to a person, with the shipment history, the documents and the correspondence already assembled. Claims, damage, loss and anything involving a concession or credit are routed by rule rather than by confidence.

Does it send documents automatically?

Where you allow it to. Document retrieval is the least contentious case because there is no judgment involved, but whether the agent sends directly or prepares the reply for approval is your decision and can differ per document type.

Is routing alone worth doing on its own?

Often, yes. Classifying requests, identifying the shipment and putting the message on the right desk with the record attached removes the triage step and the overnight queue without the agent sending anything to a customer, which makes it the lowest-risk place to start.

See what this would look like in your operation.

A Workflow Review traces one of your real workflows end to end and names the first step an agent could take over. No platform to evaluate first.

Start with one workflow

Fixed scope for the first workflow, agreed before we start. Expansion is your decision once it proves value.

How it works

Security, stated plainly

Where the agents run, what they can and cannot do, and exactly what we do and do not claim.

Security