We caught up with Andrew, our data scientist and the man behind DataBase Bets.
Good to chat with you. I know you’re fairly keen to fly under the radar personally, but did you want to start by giving us a bit of an idea about your background?
Sure. My earlier background in betting was fairly typical, I guess… nothing too special! I was a keen punter on racing. I certainly wasn’t a professional by any stretch of the imagination, but I gradually got more serious as the years went by… “went on the journey” like many punters do, I guess.
So you’ve always had an interest in racing. When did it become a career for you?
Yes, very much so. I’ve been a keen punter since I was old enough to do it legally (and a little before that, of course!). Professionally, my background was largely in IT and data science, and the two never crossed over… racing was very much an after-hours pursuit for me.
Eventually however, a work opportunity came up with one of the corporate bookmakers, which was the first time I was working on anything racing-related from a career point of view.
What did you do there?
My role varied a little, but ultimately it became about looking at the pricing of racing markets and whether there was a way to price racing accurately using data models, which could then be integrated into how the bookmaker did things. It really was not much different to how some data- minded punters build models to try to find an edge. I was doing it for the bookmaker however, using my background in data science to see if there was an edge to be had that could make our pricing better.
As well as that, I spent quite a lot of time looking at the trading operations and also working as a trader myself. Which was really good experience as it gave me a good idea of how it all works from that end.
Did you enjoy that?
I did, yeah. I guess you can call the bookies “the dark side” like many do, but it was a great opportunity for me to use my professional background and skills on something I had a great interest in. I was getting paid to find winners, in essence… that can’t be a bad thing!
For sure. So what happened next?
Well as I’d spent a few years doing this, when I left the bookmaking industry it made sense that I’d have a go at putting together a similar model and process that I could use for my own betting. So I got to work on that and have been refining it over a couple of years since.
At the same time, through some contacts I’d made I also hooked up a large-volume racing punter and was able to spend quite a bit of time working with him. He effectively mentored me and showed me how his whole operation works, from data gathering to form analysis to bet selection to the betting itself. I know he’s also keen to keep a low profile so I won’t talk too much about him… but safe say he’s one of the biggest racing players in the country.
So you’ve seen things from both sides of the fence, bookie and punter?
I guess so, yeah. I had my previous professional skills I could apply, but I’ve now been able to see how the punting industry works as well.
The DataBase Ratings – are they a product of all the work you’ve done on your own system?
Yes, basically. Once we got the system to the point where we were confident in the prices, we started publishing the ratings. But it’s been a continual process of evolving and refining since to ensure the highest possible accuracy.
So how does it work?
Well without getting technical – because it’s not that interesting – basically the models take around forty different variables which cover horse, track and race factors. These are crunched by 25 different models, all looking at them from different angles, and the results are then blended to create a single final probability, which gives us a price.
Are they accurate?
Yep, we’ve achieved really good strike rates for our top raters in terms of win, place and exotic results. They’re constantly evolving, too.
A major tactic we’ve employed is machine learning. The models produce our ratings, but every day new results are fed in, and the models re-train themselves based on the latest data. So while things are always changing – and we know the betting market is constantly evolving – machine learning ensures the model stays on top of that as much as possible.
So when did DataBase Bets come about?
The ultimate goal is to produce a profitable betting approach of course, so once we had the DataBase Ratings to a sufficient level, we started betting to test the results.
So how do they work?
We’ve refined it to target the bracket in the market where our ratings have the most success, and from there we take the best value selections on our ratings.
In 2019, we’ve now clocked up 53 units profit at 6.4% PoT. And pre-launch testing showed us it’s better than that.
And what does the package cover?
We have an average of around 20 bets per week spread across any day or state. The model analyses everything: any race at any TAB meeting in Australia could qualify for these bets.
So what does a punter get, and how should they bet?
They get the selections each day. We’re going to record official results at both Best of the Best (BoB) and Betfair SP, which are the two best tote-style betting products. Everybody has access to them so it’s easy to get on and it’s not time-sensitive.
Dead simple: one-unit stake per bet. As I said, we’re targeting the bracket in the market where our ratings have the most success, so we don’t need to stake variably on the rated or expected price. If we were going to be hitting low strike-rate plays – say, $40 shots or more – then of course we’d have to stake differently because you’d burn through the bank waiting for winners. But we won’t be here.