MultiChoice boosts user satisfaction with Azure Machine Learning and Azure Databricks

With 23.5 million households in 50 markets across sub-Saharan Africa, MultiChoice aspires to be the entertainment platform of choice for African users. "We want to do this by providing world-class entertainment and customer service,” shares Adolfo Almeida, Senior AI Engineer at MultiChoice. “And we need technology to achieve this at scale."
The global pandemic saw a surge in platform use, as people confined to their homes spent more time watching television or browsing the news. For MultiChoice, it was important to adapt to these changes. "There were a lot more people using our platforms and relying on our content recommendations," remarks Nishen Naicker, Principal AI Product Manager at MultiChoice. This shift encouraged MultiChoice to make its content more accessible and refine its recommendations engine.

Personalizing content suggestions

MultiChoice launched its AI Center of Excellence in 2020, focused on projects crucial for enhancing consumer experience through technology. This included initiatives like the subtitles program, aimed at capturing more users by localizing content into various languages. MultiChoice chose to use Microsoft Azure advanced AI services to grow its business. “With Microsoft’s support, we fast-tracked the subtitles program, which paved the way for more ambitious projects,” highlights Almeida.

With the success of the subtitling project, MultiChoice then went about updating its SuperSport recommendations engine. Initially, recommendations were generated based on a set of explicit rules—relying on factors like user preferences, item metadata, and the freshness of the item. “For example, if I like Manchester United, I select it as my favorite,” explains Almeida. “Then the recommendation engine will suggest Manchester United–related news articles and videos when I come to the platform.”

However, this rule-based method struggled to adapt when users' favorite content was not available, often suggesting unrelated items instead. Almeida shares, "If we didn’t have news about Manchester United, the system would suggest items that weren’t interesting to the user, like tennis and golf." In theory, the company could tweak the rules, but the process is highly manual and adds extra work for its staff. The rule-based system is also inefficient—it needs multiple rules to make sure the right content surfaces for its different users. Rules-based recommendations also didn’t take user behavior into account, making it impossible for the content suggestions to evolve.

MultiChoice decided to upgrade SuperSport’s recommendations engine using Azure Machine Learning and Azure Databricks. With a couple of months’ historical data, MultiChoice trained an AI model on proprietary information like user metadata, item metadata, and even contextual information around user interactions, such as the time of day. With this rich training data, the AI engine can map user preferences and serve the most relevant and fresh content to each customer. “The AI-based recommendations enable us to provide our sports fans with improved personalized experiences that keep learning from how each fan consumes content. Machine Learning algorithms analyze the user data and preferences and provide them with tailored content suggestions,” adds Johan Huyser, Senior Manager: Product for SuperSport Applications.

More engagement, greater satisfaction

To see if this upgrade was successful, MultiChoice performed an A/B test on the SuperSport mobile platform by splitting the traffic between the old and the new recommendation systems. “With Azure Machine Learning and Azure Databricks, we observed a 20% increase in daily users reading recommended articles and a 15% increase in viewership of recommended video highlights,” shares Almeida. “I don’t believe we could ever have gained a 15% increase with the rule-based model, even with more rules."

Unsurprisingly, the AI model improved user experience and satisfaction. “Since implementation, we have seen an improved user experience, which led to a 15% increase in customer engagement,” notes Huyser. Before, complaints around unwanted content were common. "Since we launched the AI model, we haven’t seen any complaints about recommendations," Naicker mentions. “People like what they’re being offered, and they click on the recommendations. They’re consuming more content and spending more time on our platform.”

After the upgrade, the system is also easier to maintain and protect. “It takes up to 12 hours to train the AI model, so even when we do a major update, we can do it overnight without disrupting the users’ experience,” remarks Almeida. “And we know our users’ data is secure with encryption for stored data and data in transit. As our data moves between Azure Data Lake Storage, Azure Databricks, and Azure Machine Learning, we know it’s not compromised.” Huyser adds, “We have seen a 20% reduction in operational costs.”

What’s more, the AI-based model is more efficient and timelier. Instead of a batch recommendation system that generates precomputed recommendations once a day, MultiChoice opted for real-time recommendations. Offers are generated only when a user accesses the app. This reduces the staleness of the information and saves computing resources.

Shaping an AI future

MultiChoice’s collaboration with Microsoft has been transformative, accelerating its AI journey and reshaping operational paradigms. "From the very beginning, Microsoft placed us in the AI Accelerate Program. This gave us access to a number of technical resources, helping us accelerate our adoption of the Azure AI services," remarks Naicker.

This support was pivotal in developing a high-performance AI recommendation engine from scratch. “We had to make sure that the latency is within the 200 milliseconds at 99 percentile and that it can handle 1,000 queries per second. The Microsoft team helped us to meet all our latency and load requirements and enhance the performance of the system,” Naicker adds.

MultiChoice also embraced other services from Microsoft while streamlining its operations. "Apart from Azure, Microsoft Teams became the most used platform across MultiChoice for internal team communication and collaboration. What’s more, Copilot in Microsoft Teams is transforming our workflows, from handling meeting recordings to accessing information more efficiently," adds Naicker. MultiChoice also deployed an employee chatbot based on Azure OpenAI Service, aiming to help its staff access information and understand internal processes or policies more easily.

Looking ahead, MultiChoice is set to continue its AI innovation journey. With Microsoft's continued support, it aims to further integrate AI into operations, enhancing customer experience, retention, and internal workflows. “With entertainment, we have only one shot; if we don't capture or engage the customer when they open the app, they won't be interested. So, it’s very important that we personalize our customers’ experiences and that we can do it on a scale. Microsoft AI services enable us to do just that,” sums up Naicker.