From a text message to a planned route

Proctor is the system we built with a waste-services company to turn customer texts into booked, routed jobs. Customers keep messaging the way they always have: “Crush the bins Friday.” “Can we get a swap at the yard this week?” Proctor works out who is asking, which site and which service they mean, records the booking in the company’s existing system, and re-plans the day’s routes around it. The dispatcher approves every new plan.

Before Proctor, a person did all of that by hand: recognise the sender, remember the site, enter the job, and find room for it in a driver’s day. This post walks through how each step works now, and which requests still go to a person.

How does an AI agent turn a short text into a booking?

By filling in everything the customer left out from the company’s own records. Regular customers write in shorthand, and a few words carry a lot of meaning. Proctor reads each message and resolves:

• Who is asking, from the sender.
• Which site, from the customer’s past jobs.
• Which service, from the words they used.
• When. A day with no time means any time that day. No day at all means today.
• The price, which is always the customer’s agreed rate.

It also applies the trade’s own rules. A replacement container is the same size as the one already on site, so Proctor does not ask. The driver who last visited is assigned again where possible. If a customer has several sites and has not said which one, Proctor asks, because a booking at the wrong address is worse than a question.

How can a dispatcher check what the agent booked?

Every booking shows the words it was based on. When Proctor decides a message is a request for a particular service, it points to the phrase in the customer’s message that says so, and the system confirms that phrase is in the message. A booking cannot cite words the customer did not write.

That makes every job checkable at a glance, in the customer’s own words.

Which requests does the agent book itself, and which go to the dispatcher?

Proctor books a request only when it is routine and two independent checks agree. Plain rule-based checks run alongside the agent’s own reading of the message. When both agree and the request is routine, it is booked. When they disagree on anything, a person reviews it. If a request is waiting and nobody has looked at it, the owner gets a reminder.

Requests that arrive in a fixed format, such as a standard form, are read by ordinary code, not by the agent. Each kind of input gets the simplest method that is reliable for it.

How are customer and address details checked?

Against the company’s records, before anything is saved. Each customer and site reference is matched to the company’s data. Addresses are looked up within the company’s service area, and any address implausibly far outside it is questioned rather than booked.

The dispatcher can trust what appears on the board because nothing reaches the board unchecked.

What happens to the route when a new job is booked?

Proctor accepts the booking and re-plans the route. It does not ask whether there is room first. That is how a good dispatcher thinks: a regular customer gets served, and the plan adjusts.

The route is planned from where each truck is at that moment. Visits with a fixed time stay where they are. Everything else can move, so a new stop can fit in. If a day becomes very full, the plan adds extra time at the end of it.

The dispatcher stays in charge throughout. They can fix a visit’s time, move a job to another driver, and they review and approve each new plan.

How is the system kept dependable?

Two checks run all the time. The live map depends on location data from the trucks, so Proctor regularly confirms that every truck is visible on the map, which is what the dispatcher relies on. And the system is tested against a collection of realistic customer messages, so every change is checked against the requests customers send every day.

What changes for the business?

For the owner, bookings are no longer retyped and the day’s route is no longer rebuilt by hand every time the phone buzzes. For customers, nothing changes, which is the point. They send the same short message they always have, and the truck arrives.

More on Proctor and our work with field services operators.