What you need to know:

• Kayo is four years into the development of an AI decisioning-powered customer intelligence project that has seen sports streaming platform tap into global sporting schedules and customer data to predict behaviour at the individual level and automate 1.2 million potential actions across its 1.5 million subscriber base.
• Built on an ecosystem of 10 in-house machine learning models, with integrations from AI decisioning firm OfferFit and customer engagement platform Braze, the Customer Cortex has enabled Kayo to forecast quarterly churn and reactivation with 98-99 per cent accuracy – all the while “maximising individual customer metrics across the customer life cycle”.
• Foxtel Group’s executive director of customer and revenue Anthony O’Byrne reckons the setup is unmatched in the local market and is leading globally in terms of sports streaming.
• He wouldn't disclose specifics on churn data, but says reactivation offers are now working at 90 per cent efficiency – meaning 90 per cent of churned users who reactivate their subscription off the back of an automated action (like an email) would not have done so otherwise.
• It’s translating commercially, with Kayo subscriptions up 14 per cent in FY2024 and continuing to lift ahead of the Dazn acquisition.
• Foxtel is looking to replicate the same success for Binge – though tweaking the system to work for entertainment has been a more complex task than mapping sports schedules.

Strong subscriber growth in recent years has put Kayo Sports chief among Foxtel Group’s streaming assets – a key draw for new owner Dazn.

Before the deal went through Kayo had ended the first quarter of the 2025 financial year with 1.511 million subscribers – seven per cent up year on year. It followed on from a strong end to fiscal 2024, in which the platform grew subscribers 14 per cent.

Speaking to Mi3 last August, then-CEO of Kayo and Binge Julian Ogrin said the sports platform’s success was in part due to its new status as a “utility service” alongside Netflix and Spotify. Ogrin, who has since stepped into a new a new role as chief business officer, told Mi3 that sport – and Kayo – had not just weathered cuts to discretionary spend but had seen the “lowest pausing rates since the launch of the company” and previously lapsed customers returning in droves.

Per Ogrin, no small factor in the equation was personalisation at scale. Foxtel has spent the last four years building out what it touts as a "world-leading" customer engagement and retention capability dubbed the Customer Cortex. Built in partnership with customer engagement platform Braze and AI decisioning firm OfferFit, it feeds off 10 in-house machine learning models that track and predict customer engagement at the individual level and deploys AI decisioning agents to drive thousands of daily actions designed to keep subscribers active – or to bring them back in.

Four-year build

Foxtel Group executive director of customer and revenue Anthony O’Byrne says the project first came about after Covid decimated local and global sports competitions and the bulk of Kayo’s programming temporarily fell away.

In response, the customer team sent out an email to 10,000 customers highlighting Kayo’s on-demand offering to see if they could customers engaged without live sport – and it fell flat. Per O’Byrne: “500 people canceled in the first five minutes after receiving that email”.

“Then over the next 18 months, we built this rapid A/B and multi variant testing system and did 300 tests… We found a bunch of value in that but then realised pretty quickly that they [the tests] had picked up a lot of low hanging fruit – and the way to get to the next level of the tree is through AI.”

So in July 2021 Foxtel kicked off a pilot with AI decisioning start-up OfferFit – becoming their “second or third-ever customer”, per O’Byrne. From there, the business went onto build out 10 in-house machine learning models that digest customer data and global sports schedules to model and predict behaviour “at the individual level”.

For each individual customer, a “strength of schedule” is created, taking each user’s historical affinity with different sports, series, and teams and mapping it to “the next two years of upcoming sport”, which is used to model customer predictions.

The data produced by the ML models, now each in their fourth or fifth iteration, sits underneath what O’Byrne describes as the “command centre” of the Customer Cortex – encompassing three core operational functions: customer decisioning, business forecasting and valuations.

“It's a really mature capability now,” says O’Byrne. “It's actually how we run the business day-to-day, and with customer marketing. We still occasionally have some bespoke stuff over the top, but the bulk of our day-to-day running is now in this automated daily cycle and run through AI decisioning.”

Beyond next best action

Customer decisioning is handled by OfferFit’s AI-decisioning agents, which receive around one billion incremental data points from Kayo’s ML models each day. The bots use the data to generate around 50 thousand recommended actions per day into Braze – Foxtel’s customer engagement platform – which executes and then feeds back customer responses into Foxtel’s CDP.

“If they're a currently churned customer, it might be, 'what is the best way to reactivate that customer? Is it a piece of content? Is it a promo? And if it is, what's the what's the channel, what's the message, what's the what's the time, what's the event we should be talking about?'” says O’Byrne. “But also, if you're an active customer, it's looking at trying to retain you. It's looking at trying to get you watching more content.”

Foxtel made the switch to Braze in February, broadening the Customer Cortex system’s capabilities to over 1.2 million different potential action canvases – up from only 300 with the previous provider.

Conveniently, the new tech has a native integration with OfferFit after spending US$325 million to acquire the AI decisioning tech six weeks ago – a move O’Byrne says was partly influenced by proven use cases demonstrated by Kayo’s Customer Cortex.  

“They tell me that we were one of the blueprint use cases for justifying that purchase –because we showed them the power of the two partners working together.”

The latest iteration of the system is driving significant incremental revenue gains for the Kayo platform, per O’Byrne, though he won’t give specifics on how it’s impacted churn rates directly, instead using reactivation efficiency as an example.

“When we first launched the agentic AI on our reactivation use case, it was running at 50 per cent efficiency,” he says. Meaning: “half of the customers who were coming back on a specific promotion would not have come back otherwise”.

Within three months, as the AI learned, efficiency hit 80 per cent and now its consistently over 90 per cent – and often in the high 90s.

O’Byrne says high hit rates comes down to the models’ predictive capabilities – unlike traditional next best action models, which are “trained only on historic information”.

“When we launched, we had the equivalent to a next best action model, and so that 50 per cent is where we would have plateaued. But the benefit of going to agentic AI for this decision making, is that growth through 50 to 80 to 90 per cent.”

Forecasting and valuation

The second function of the Customer Cortex, per O'Byrne, is as a forecasting tool – with Kayo having been able to predict quarterly churn and retention rates with 98-99 per cent accuracy for the last eight quarters.

“All of our business decision-making, our budgeting and forecasting and tracking, now happens at a one-to-one level. So if you were a seasonal customer, as an example, we might know that you're most likely to reactivate at the start of the footy season in March, and then we have a percentage likelihood that you're going to leave in October, and then a similar percentage likelihood that you as an individual, are going to come back the following February, March, April – whatever it is.”

The third part of it is valuations – using the forecasting deal to inform business decisioning with modelling.  

For example, Foxtel could look at hypothetically “what would happen if we were to go and make some improvements to the tech on one of our devices” per O'Byrne.

“Say, if bit rate [the rate at which data is processed or transferred] is a bit slow on the devices, we could go and tweak that in the forecasting tool… Flow that through the forecast, and it gives you a new forecast with a new total revenue off the back of it, and the delta between the two then gives us the value of making that change.”

Binge next

With Kayo providing proof of concept, work has been underway to replicate the success of the Customer Cortex solution with Binge. The entertainment platform will be under pressure to perform after the loss of the Warner Bros slate – and it's growth has been slower than Kayo's – Binge subscribers rose only 1 per cent in FY24, and at half the rate of Kayo in the first quarter of F25.

O’Byrne says both platforms have always been in the blueprint, and were initially set up in tandem, but the bulk of the sophistication went into Kayo and has only recently been migrated into Binge.

To get the entertainment streaming service up to scratch, the Foxtel team had to build out a completely different strength of schedule based on how customers’ tastes align to the existing back catalogue of shows, he says.

“Then we layer in returning shows, like second and third seasons of shows you've already watched, and then we use the editorial team to help us get evaluation for new shows.”

So far so good: Binge has already seen customer retention efficiency lift from 60 per cent to above 90 per cent, per O'Byrne, who says the grunt work is paying off.

“It's been four years of work to get it to the level it is, and it's just working really well – it's a slam dunk commercially for us.”

He reckons the self-built tech puts the group ahead of competitors.

“I can say with very high confidence that certainly in our space we’re a few years ahead of the Australian market, and in sport, we’re probably leading this globally in terms of understanding of sports customers.

“We go to a lot of conferences, we hear the presentations, and we have people pitching next best action models to us," per O'Byrne. "They are the old way of doing things. We're at the next evolution of where this is going.”