Racing writer and betting-ring chronicler Simon Nott has been running his Zooms In interview series for a while now, and for episode 43 he turned the camera on our founder. Forty minutes on where RaceMetrics came from, how it gets built, and where it's going — including a few things we've never really explained in public before. The full conversation is above; here are the highlights, lightly edited for clarity from the recording.
It started as a free game
RaceMetrics didn't begin life as an analytics platform. It began as a competition.
"My initial plan was really simple. I just wanted to create a competition that was free to enter, based on ratings of horses or trainers or jockeys… like Elo ratings, the system from the chess world. I managed to create a rating for all seven connections in horse racing — updated daily. The idea was a bit like fantasy football: pick seven participants each week, they score points as the week goes on, and the winners get a prize."
That competition still runs every week — it's RaceMetrics 7, and it's still free. But once the ratings existed for every horse, trainer, jockey, owner, sire, dam and damsire in the database, putting them on a racecard was the obvious next step. Everything since has grown from there.
Twenty-six years of data underneath
The part that doesn't show on screen is the part that matters most.
"The key to how RaceMetrics works is you've got to have a good data set behind it. I've created my data set over 26 years now, so it's a very clean horse racing database with data coming in from various angles."
That's the same database that has powered Proform since the nineties — and it's why features can now appear at a pace that surprised even the man building them:
"I'll get an email from a customer who says 'it'd be great if you could do this'… and I can have a feature live within an hour. The other day someone asked for something, and I created the feature, it went live on the site — a video, a blog post, a news article, social media posts and a mailshot — in an hour. I could do that before, but it might have taken me a week."
Ask the database, not the internet
A lot of AI racing products are appearing at the moment. The interview gets into why most of them worry us — and what we did differently with Ask RaceMetrics.
"There's quite a lot of people doing AI stuff in racing now… but they're often just asking AI to come up with something, not knowing where it's got the information from. It can just completely hallucinate and give you false information. When you've got domain knowledge, you know what's false — that's the key."
"If you just go on an AI chat and ask the same question, it has to pull context from what it's learned from the internet. It hasn't got access to a database in the right format. Ask RaceMetrics is linked to the database — it knows the schema, so it knows where to look. It astounds you sometimes."
The answers come from our results database, not from whatever the wider internet half-remembers.
"You might have just worked out the past"
The most important exchange in the interview is about back-testing — and it's the objection every experienced punter raises, so it's worth quoting at length. Build a system, test it against history, and the danger is always the same:
"You might have just worked out the past rather than worked out the future. So you can do what's called a hold-out test: it doesn't test against the current year until the end, and then it tests against that — and they've both got to be working. Or you can do a batch test, where you might use 20% or 50% of the data as the initial test."
That's the design philosophy of the System Builder in one answer. It will happily show you a beautiful back-test — and then tell you whether the same rules held up on a year of racing they never saw, whether the result is statistically significant, and how the angle performed against the market's own prices. Most ideas fail those checks. That's the tool doing its job.
The tips, warts and all
The daily tips came up too — including the honest bit most services would leave out.
"They're published at quarter to eight in the morning and that's it, they're set… it's warts and all. They're profitable to Betfair SP, which is often quite difficult to do long-term. The site's been live since March and they're still making a profit — it's had one losing month, but that's going to happen."
And on the hot streak earlier this month:
"At the beginning of this month there was a real run — up about 35 points by mid-month with a strike rate around 45%. And I was thinking: this is not sustainable, it's not going to continue… even at four or five to one as an average price you're going to get losing runs. That's what's happened in the second half of the month."
Every selection, every result and every price is published on the tips page — the good months and the bad ones at several price bases. If you're the sort of person who checks a record before trusting it, that's exactly the record to check.
Your racecard, your layout
"When the Racing Post changed their layout, everybody was complaining about it. My first thought was: why doesn't anybody offer a layout you decide yourself? So there's My Card now — you can choose what you want on the racecard and save up to ten of them. A different one for flat racing, a different one for maidens with more breeding data, a different one for the jumps."
What's next
The interview ends on where this is heading: linking into the exchanges for live prices, automated betting on saved systems, and a tool that lets Proform subscribers plug the racing database directly into the AI assistant they already use. As Simon Nott put it: "It's like a beast — you'll never be able to stop it."
Our thanks to Simon Nott for the conversation — you'll find the full episode above, and plenty more racing interviews on his channel and at simonnott.co.uk.
RaceMetrics has a free tier if you'd like a look around, with paid plans from £10 a month — the System Builder discussed in the interview is part of RaceMetrics Pro. Details on the pricing page.
