delivery and logistics

An AI voice agent for delivery operations, on the calls that decide whether the van comes back

A delivery window nobody confirmed, an address that is nearly right, an access instruction that never reached the driver. Each of those is a failed drop, a phone call, and a second journey. The agent confirms windows before the run, reads addresses back until they are right, and rebooks the failures the same day.

what your phone line actually carries

Three calls, all of them about the same van

In logistics the phone is not customer service. It is operations.

01

Confirming or moving the window

Outbound and high volume: is somebody in, does the slot still work, can we come earlier. Every confirmed window is a drop that succeeds; every unconfirmed one is a coin toss with a van attached.

02

The failed delivery

Nobody home, wrong address, no access. The recovery call has to happen fast, because tomorrow's route is being built this afternoon.

03

Where is it, and what happened

Inbound, repetitive, and answerable from a live read of your own system. It becomes a complaint only when the answer given does not match what the system says.

where the line leaks

The cost of a failed drop is paid twice

Both of these are call-handling problems wearing an operations costume.

The address error is discovered at the door

A wrong house number, a missing flat, a building code nobody wrote down. It costs nothing to catch on a phone call and a whole second journey to catch on the doorstep — and the driver's time is the most expensive minute in the operation.

Confirmation calls do not scale on people

Ringing every drop before the run is obviously worth doing and nobody staffs it, because it is hundreds of short calls in a narrow window every single evening. So it gets replaced by a text message that a third of recipients never read.

what we would build for a delivery operation

Confirm, verify, rebook

Short calls, high volume, and every outcome written back as a field rather than a note.

01

Confirms the window before the van leaves

Hundreds of short outbound calls in the same evening slot, each ending in a structured outcome: confirmed, moved, cancelled, or nobody reached. Billing is talk time only, so an unanswered dial costs you almost nothing — which is what makes calling every drop affordable at all.

02

Reads the address back until it is right

Addresses, building codes, floor and access instructions are confirmed on the call rather than assumed, and spellings are verified. This is the single feature with the most obvious payback in this industry, because the alternative is finding out at the door.

03

Recovers the failed drop the same day

When a delivery fails, the agent calls, establishes what went wrong, and rebooks into a window your operation has actually released — capturing the access detail that was missing the first time so the second attempt is not a repeat.

where it stops

What it will not do in logistics

The limits here are about not writing cheques the operation cannot cash.

It will not promise a window dispatch has not released

It offers what your system actually holds. Inventing a convenient slot to end a call politely creates a second failed delivery and a customer who was told a specific thing that did not happen.

It will not rule on customs, duty or liability

Duty, clearance, damage claims and anything with money or law attached go to a person with the case captured. These are determinations, not answers, and a recorded guess becomes your problem.

It will not call a list you cannot account for

Confirmation and recovery calls are outbound calls. There has to be a lawful basis for the number, opt-out registries and calling-hour restrictions apply, and we will not run a campaign against a list whose provenance you cannot evidence.

Consent, lists and calling hours →
what it plugs into

Your transport system and your route plan

The outcome of the call has to reach the route before the route is frozen.

Transport management and delivery platform

The call list comes from your system and the outcome goes back into it: confirmed, rescheduled, address corrected, access instruction added, with the transcript and recording attached. Timing matters here more than in any other vertical, because a result that lands after the route is built is a result nobody used.

Volume when you need it and none when you do not

Concurrency is not a rota, so a peak week is answered like a quiet one. With talk-time-only billing, a run of unanswered dials costs a fraction of a run of connected conversations.

other industries

The same machinery, a different script

What changes between these pages is which calls matter and what the agent has to be stopped from doing.

Clinics, dental and medical practices

Reception is already serving the person at the desk. The phone is the patient you never met.

What we would build →

Real estate and property

The agent who answers first gets the viewing. Everyone else gets a callback nobody returns.

What we would build →

Home services and trades

You are under a sink. The phone is a job you have not quoted yet.

What we would build →

Retail and e-commerce

Three questions are most of your call volume, and all three have an answer in a system.

What we would build →

Automotive dealers and service centres

The service desk is with a customer at the counter. The sales enquiry arrives at nine at night.

What we would build →

Not on the list?

These six are where inbound call volume is heaviest and the integrations are best understood, not the limit of what the agent handles. If your calls are repetitive and the answers live in a system, the shape of the work is the same.

Every industry page →
questions buyers ask us

Frequently asked questions

Can it call every delivery to confirm the window?
Yes, and the economics are what make it worth doing: the calls are short, billing is talk time only, and an unanswered dial costs very little. Each call ends in a structured outcome written back into your system rather than a note for somebody to read. You need a lawful basis for the numbers, and calling-hour rules apply.
How does address verification actually work?
The agent reads the address, the building or door code and the access instruction back to the recipient and confirms them on the call, and verifies spellings rather than assuming them. It is the same mechanism used for registrations and dates in other industries, applied to the field where an error costs a whole second journey.
Can it handle the recovery call after a failed delivery?
Yes: it establishes what went wrong, captures the detail that was missing, and rebooks into a window your operation has released. It will not invent a slot to make the call end pleasantly, because that produces a second failure and a customer who was promised something specific.
What about peak season?
Concurrency is not a rota, so the volume that breaks a human phone room is the volume this handles unchanged, and reply latency does not degrade as calls pile up. Peak is the period where the difference is easiest to measure, because it is when your current answer rate is at its worst.
Will it work in more than one language on the same operation?
The speech stack is ours, so the agent can hold the conversation in the languages we configure. The real work is that your access instructions, your window rules and your escalation policy have to exist in each of those languages, which is scoping rather than a setting.
How do we audit what it told our recipients?
Every call is transcribed and scored against a checklist you agree with us, not sampled. Locator reviews 100% of conversations; a human quality team reaches 3–5%, a figure from our own experience of running real call operations. For a delivery operation the useful report is the pattern across a route, not one call.
What call scoring produces →