We run a car supermarket group in the Midlands. We've been in the trade a long time. Today that means four sites, around 1,900 cars in stock and the same problems every dealer in Britain has: too many cars to price properly, too many invoices to read, too much admin, and never quite enough good people to do all of it well.
We are not a software company. We didn't set out to become one. What follows is what happened when a dealer started working with an AI agent and kept asking it to do one more job.
17 March 2026
The first conversation
George Manning, our dealer principal, started working with an AI agent over a web chat, then over Telegram so it could go wherever he did. The question was simple: could an agent do the reports a manager did every Monday morning? It could. The reports arrived before anyone had got to the office, and they were right.
Three days later, on 20 March, we had the first agent that acted rather than reported. It worked our trade sources to George's own buying rules and came back each morning with what it had seen, what it had done and what it had passed on. That was the moment the idea changed from "a clever assistant" to "a member of staff who never sleeps".
By the end of the month we had a review-handling app running. The agents were now dealing with customers' words, not just our numbers.
April
Dashboards, then the spares bill
April was about seeing the whole business in one place. Live dashboards for every department, an overheads tracker, and a key-cabinet integration so that every set of keys was tracked automatically instead of being asked after.
Then the invoices. On 14 April an agent read hundreds of supplier invoices in a single day, line by line, across more than 40 suppliers. It caught wrong VAT treatments, duplicate references and mis-read dates that had been paid without a second look for years. Nobody had ever read those lines, because nobody had the time. That one day was the start of what became a 48% cut in our spares bill.
May
Pricing
This is the month that changed the numbers. We started with a small tool we called "Profit Sanity", a check on whether each deal made sense. It grew into a full forecourt pricing system: every car priced against the live market, every day, with a reason written in plain English for each hold, raise or cut.
Alongside it came a Price Changes app so the team could see and challenge every move, and a test-sales dashboard so we could measure what happened next. Stock turn started climbing here, and it has kept climbing. It is up 48% since agent-led pricing went live.
It didn't all go right. The first pricing agent was too aggressive on slow stock and cost us margin on a handful of cars before a manager spotted the pattern. We fixed the rules, added sign-off for big moves, and re-score every price change the next night. The pricing agent is our best performer today because it was our worst to begin with.
June
Prep and the workshop
Prep is where cars sit and margin leaks. In June we built a prep tracker with the pre-delivery inspection done in the app itself, logins and roles for every technician, and jobs written back to the DMS so the workshop and the sales floor were finally looking at the same facts. Every car through prep is now tracked to the hour.
July
Sales, customers and the cost of it all
July turned the agents towards the people who buy from us. A CRM board and inbox. An aftercare desk where AI classifies every case before a person picks it up. A sales board with per-rep coaching, so every salesperson knows where they stand and what to do next. An "ask the business anything" agent for managers. And a self-learning skill library: when an agent works out how to do something well, it writes it down for the others.
We also looked hard at what all this was costing. We reviewed every model and every job, cut what wasn't earning its place, and settled the whole stack at an average of £2,400 a month. That is less than one salary, for everything.
September
The pricing agent gets its own control room
The Pricing Command Centre is the pricing agent's own dashboard: the daily scan, the patterns it has learned, and a straight answer to the question "are we getting smarter?". By now the head agent had run tens of thousands of self-improvement loops, re-scoring each day and rewriting rules overnight.
And we made a decision. Other dealers had seen the boards and asked if they could have them. OmegAi is the answer to that question.
Today
Where it stands
26 live agent apps run the day-to-day of our dealership group. The head agent has run more than 42,000 self-improvement loops. Over a million pricing decisions have been logged, each with its reason. 21,000 competitor adverts are watched. In one month the agents read 1,879 invoices and tracked 1,290 finance proposals. On one day the sales board showed 67 cars invoiced at £158,000 margin, £942 a unit. The whole stack still costs about £2,400 a month to run.
Every one of those figures is from our own operation. We haven't sold this to anyone yet. That is the point of the next chapter.
OmegAi is a trading name for now; company registration is to follow. What matters is this: we are the customer as well as the builder. We know what a bad month feels like, and why nobody wants a six-month IT project. So there isn't one.
Omega is the last letter. We liked the idea that this is the end of doing it by hand: the last spreadsheet, the last Monday morning spent re-pricing, the last parts invoice that nobody read. We didn't spend long on the name. We spent it on the agents.
