Off the beaten path

Seemingly the whole world is talking about data, machine learning and AI. The tech giants are all investing heavily into these areas; for some players – step forward Google and Baidu – these domains are at the very heart of their business models.

Data and machine learning can add value across many elements of the gambling sector from smarter forecasting, through better targeted sales and marketing, to enhanced customer experiences and service.

Andy Walton

But what are the elements needed to successfully transform a business and drive real value from machine learning?

The first step is the choice of use case; there are so many potential opportunities to exploit and the temptation is to get too complex too early. You’ll be getting peppered with bold claims from numerous third parties, and the bar moves higher with each headline about self-driving cars.

You should aim for first use cases that are valuable, realistic and understandable. Use them to gain buy-in, learn the tech and prove the value. Secondly, the quality of the internal data ecosystem is crucial to success; you’ll ideally have access to a wide variety of quality datasets from transactional to behavioural.

Plus there’s a plethora of third party and open source that can act as extra valuable model features. Then you’ll need the compute power to store, transform and process this data at scale. Many operators have focused on this area recently, some building on premise platforms, others using the cloud.

All this requires investment, time and skills, not to mention the complexity of compliance with important regulations such as GDPR. Next is the data science team who require a balanced mix of advanced maths skills, cutting edge engineering talent and deep business domain knowledge.

Designing, training and deploying machine learning models across distributed platforms is stimulating work; the right team will thrive on the challenge. However, assembling this team will be challenging and you’ll likely experience numerous false dawns before you see any productive output.

The next step

With your model built, the hard work starts. A model running in isolation on a cluster is academically interesting, but can only add value when used to drive positive change for customers or internal teams. You now need to herd the cats and mobilise teams across the business.

Each link in the chain, whether an internal system, external system or business process, needs to be reviewed and potentially refactored to maximise the model output at the customer facing layer. You may well use a prototype to assess the value, but when running at scale operational efficiency and automation are crucial.

Manual interventions are almost certain to severely inhibit the potential upside. The last step is the hardest of all but delivers the biggest benefit: the right culture. Your organisation, from top to bottom, will need to buy-in to the benefits of machine learning.

They’ll need to trust the blackbox; it’s imperative this doesn’t become a case of “computer says no” with distrust over the model outputs. Established processes will be ripped up. Manual flexibility may need to be sacrificed for improved accuracy and automation.

People will need to work collaboratively across many teams and departments to make things happen. They’ll need to truly believe in failing fast, as this is a complex area with many opportunities to learn from mistakes.

Does this all sound tough? Yes, but the potential upside is huge. In the short term, there are numerous low-hanging fruits for operators that stride into the machine learning jungle. Medium term, data rich businesses that don’t embrace AI could get picked-off by faster and more artificially intelligent competitors.