Future Anthem is a market leader in game data science and uses Artificial Intelligence to provide actionable intelligence on safer play and improved game experience within the gambling industry. It aims to connect human behaviour with data science to build a player centric universe that is enjoyable and sustainable and is trusted by teams at Paddypower, betfair and Betfred. Chris Conroy, Chief Data Officer at Future Anthem, spoke to Intelligent SME.tech about how the company has leveraged Databricks to deliver Data-Science-as-a-Service to the global gambling industry at scale, and its role in the debate around gambling regulation.
When did Future Anthem start using the Databricks platform and why did you choose it?
We actually started using Databricks near the beginning of the company. For us, it was part of our first MVP, which is great. When we were building a platform, we went through a few decision points. I think the first one for most new tech businesses is which cloud you’re going to go to.
At that point, we chose Azure. With the kind of platform we wanted to build, there were some obvious elements to it. We knew we were going to be dealing with massive data. We knew that Spark would have to be a core component of the platform.
There were some decisions that were almost already made. We knew we were going into Azure, we knew we would have some sort of platform based around the data lake for storage, we knew we would be using Spark as a core framework. And ultimately, we knew that we had to deliver an end-to-end platform for data science.
We were really clear in terms of what we needed for a platform, which was something to facilitate a data product business with data science at its core.
I became very aware of Databricks at a previous company because we developed a platform within Azure. When we went through the process at Anthem, I spoke to a few different experts in terms of ‘here’s the reference architecture we would like to deliver’, and pretty much everybody said Databricks should be at the core of this.
There was one more fundamental for me, which was, given that data science is much more than models, and given that we need end-to-end processes, and I’m really passionate about delivering data science models as a single end-to-end entity rather than a whole set of different packages, one of the things that Databricks delivers almost inherently is that it forces your organisation to think more end-to-end. It inherently breaks down a lot of the barriers between different functional teams.
How does Future Anthem use the platform to help clients?
We deliver basically a Software-as-a-Service model. We call it Data Science-as-a-Service. I’m pretty sure we didn’t actually coin that term but until anyone tells us any differently, we are claiming it.
It’s an interesting concept because it’s basically data as a product. We build products in three core areas. One big area is all around safer play and responsible gambling. The second one is around personalisation. And the third one is around data driven design.
Essentially, our job for a client is we tell them about a great product, they happily sign up and say that’s great, but our job is to make a product experience.
For a product experience, all our clients want to do or the reason they buy technology like Future Anthem is they just want to give us some data or they want to receive outputs from us that they can automate into their delivery systems, they just want it to be as simple as possible.
What we use Databricks for, from an internal perspective, is to build those product pipelines. For us, Databricks is the train track that goes through it all.
How does Databricks enable Future Anthem to capture fraud and risk more efficiently?
Risk is a massive area for us and in terms of what we do and what products we build, for us risk is all around player risk. Obviously, we’re working within a gambling industry and it’s well documented that there are a proportion of people who gamble and who are going to develop problems with their gambling. The product we have around safer play is all about identifying those players before the problem happens. We built a number of algorithms which we put into a product, all using Databricks technology at its core, which identify potential risk in players.
It really is a big area for us, and we know that it can be deployed as a batch because many companies provide these things in batch, but it can also be deployed within the play of session as well. As the players go spend by spend, we can deploy that algorithm to see what is happening in this player session and if they are now starting to exhibit potential sense of harm. And that’s the key, it’s the potential. You want to be able to intervene. For example, in a situation where we’re flagging in real-time within a session, initially, maybe a pop-up message to say do you know how much you have spent recently? Within those models, explainability is very important. Within that type of system, explainability to say, ‘this is why we’ve identified this player’ becomes really important and that same message may be triggered to the operator’s customer service team so a human in the loop system.
What is Future Anthem’s role in the debate around gambling regulation?
In terms of debate, that’s a really relevant question at the moment – just that there’s quite a large debate around gambling regulation.
It’s been something that various parliamentary groups have been discussing over the last few years, there’s been quite a lot of public proclamations from different parliamentary groups about gambling regulation. And I think ultimately that’s culminated in a formal review of the UK gambling legislation. which is actually on-going. For us at Anthem, we don’t tend to get involved in a debate in a very public way, but I would like to think we’re very involved more in the background.
We keep a really close eye on the regulation and there were recently two significant pieces of regulation released separately within the last 18 months. What we would do with something like that is make sure that the products we are building speak very closely to the problems that the legislation is asking us to look at. Secondly, we spend a lot of time with clients in this area but again in the background, so we work really closely with the teams within our clients, especially within the safer play area.
Finally, with something like the review of the UK Gambling Act, for example, we spent a lot of time internally thinking about, what’s our view on that because we spend a lot of time in this area, we’ve got a lot of experience in this area so we formulated quite a detailed response to the call for evidence on the legislation and we’ve submitted that as part of the call for evidence into the review
Why is prioritisation of culture the key enabler for cutting edge data products and solutions?
This is probably one of the most used and potentially overused words in any business and given how the whole agile methodology transitioned into most mainstream businesses over the last 10 years, I’m sure there is a business that doesn’t even have a computer at the moment – they are still talking about prioritisation. Clearly, with great power comes great responsibility but with great technology comes great opportunity.
When I’m thinking of a really important prioritisation for me, I just go back to my first-year economics textbook. And one of the biggest principles in economics is opportunity cost.
With a business like ours, we could build loads of new features, loads of new models. We’ve been talking about product in the business at the moment and a mantra of product market fit – you only build things that people actually want. With any tech business, prioritisation is probably the most important thing to do. Again, just always go back to the opportunity cost, every decision to do X is a decision not to do Y.
The other big opportunity cost you have is people’s time. It’s not just about prioritising which product feature you focus on. When I think about prioritisation, it genuinely goes down to ‘does that data scientist need to be in that meeting’ because an hour in that meeting is an hour not spent coding.