What an AI agent actually does at 3am

People ask me what the agents "do" and I think they expect something dramatic. So here is a plain account of what happens in our business at three in the morning, when the building is empty.

The pricing agent has finished its market scan. It has looked at every car we have against every comparable advert it could find and written a hold, raise or cut recommendation for each, with a reason. Those are waiting for the morning.

The admin agent has read every advert against the DMS record and the manufacturer spec. It has found a few where the service history is in the file but not on the advert, and a couple where the photos are out of order. The routine ones it has fixed. The rest it has listed for a person.

The spares agent has read yesterday's invoices. Two lines look wrong. Credit requests are drafted.

The buying agent has rebuilt tomorrow's buying brief from what actually sold this month, and is watching the trade sources for cars that match it.

And the head agent is doing the thing I think matters most. It is going back over the day: every deal, every price move, every action each agent took, and scoring it. Where a rule produced a poor result, it changes the rule. Where a person overrode an agent, it tries to work out why. Then it writes the morning report. We've done this 42,000-odd times now, and the agents are noticeably better than they were.

What it isn't doing: changing anything in the DMS without permission, sending anything to a customer that we haven't approved the shape of, or moving data out of the country. Those are hard rules and they don't relax at night.

At 07:58 my phone tells me how we did. That's the whole thing. It isn't dramatic. It's just work that used to not get done.

George Manning is the founder of OmegAi and a dealer principal who runs a UK car supermarket group. About →