Blog

We’ve just scratched the surface on machine learning in telco

Artificial intelligence (AI) and machine learning (ML). I’ve been tossing these terms around a lot lately, but haven’t spent much time talking about what they are and why the time is NOW for telcos to start using this technology. With all the buzz in the industry this week about how it’s so hard to come by revenue growth, it’s the single best idea I have for the industry to take back what’s ours from the internet giants. I believe telcos should use their data with AI and ML to improve experiences, reduce churn, boost customer satisfaction, and increase average revenue per user (ARPU). But in order to be successful —SPOILER ALERT!—you’re going to have to use the public cloud.

What is machine learning?

Machine learning is what enables computer systems to learn from data, rather than being explicitly programmed. It helps computers solve problems that are too complex for an algorithm alone.

A subfield of AI, ML dates back to the 1950s, and it evolved and matured alongside advances in computing over the last 70 years. In the 1990s and 2000s, people started using it for fraud detection and spam filtering. But it was hard for ML to reach its full potential because it demands a lot of data and processing power, and running it on-premise was a beast even for organizations with business warehouses, data lakes, and analytical models. Plus, each vendor had its own data model, and pulling them together across hundreds of applications was a herculean task, and the processing and compute needs of doing analytics on-premise made it hugely expensive.

Enter the public cloud

The advent of the public cloud changed the game. Twenty years ago, the hyperscalers pulled ML and massive computing resources together to design tools that moved their own businesses forward (think, Amazon’s retail recommendation software, and search and Ads for Google). They honed and battled tested their solutions at internet scale. That was great for their own companies, but the game was turned on its head when they made them available to all kinds of businesses, including telco, by offering them as a service on the public cloud.

ML requires three key things: tons of data, tons of computing resources, and tons of capacity. Telco definitely has the first item in spades, but the last two are best served by the public cloud. While it is possible to run ML models on premise, it’s definitely easier with public cloud resources. For example:

  1. Scalability. Running ML models requires a lot of memory, processing power, and storage. Public cloud hyperscalers provide it all, scaling up or down on-demand to handle workloads and large data sets quickly and efficiently.
  2. Specialized hardware. Public cloud chips, graphics processing units (GPUs), and tensor processing units (TPUs) are purpose-built for specific jobs, including ML workloads, and can speed the training of complex models.
  3. Cutting-edge tools and services. Hyperscalers offer all kinds of ML tools and managed services for training and deploying models, as well as data prep and visualization, all built and battle-tested by the best technologists in the world.
  4. Data storage and management. ML needs big, big data (more on that below). Public clouds offer a range of databases and storage services that can make it easier to organize, share, and access it all.
  5. Cost-efficiency. Running ML workloads on-premise would cost a fortune in upfront hardware investments and ongoing maintenance. The public cloud’s elasticity and pay-as-you-go model not only puts ML within reach of more organizations, it also ensures less waste and more efficient use of the required infrastructure.

Enter Totogi, bringing it all together

In general, ML algorithms work by analyzing patterns and making predictions or decisions based on those patterns. The more data the ML model has to learn from, the more accurate and robust its predictions or decisions are likely to be. Like that episode in the TV series Silicon Valley, a machine learning model built to determine if an item is Not Hotdog, it will do better once it’s seen a million hot dogs, rather than just a hundred.

Luckily, telcos are in a GREAT position here, with shit-tons of their own subscriber data. While an operator could build an ML model in the public cloud training on just its own data set, you can imagine how pulling in data sets from a DIVERSE group of telcos from around the world would make it even better, stronger, and more valuable.

This is why we’re so excited about the future at Totogi, where I’m acting CEO. We’re building ML models for telco based on data from *all* our customers. No one has been able to do this before, because historically vendors have been installing software on-premise in silos, in largely inaccessible walled gardens. With Totogi’s software as a service (SaaS), multi-tenant, public cloud software, we are able to tap into an extremely large AND diverse dataset to build all sorts of ML models. So you see, we aren’t building a SaaS product *just* to be able to say it’s a SaaS product; there’s an actual method to our madness.

We just launched our first ML data service a few weeks ago: Totogi Churn Prediction Service, which uses ML and AI to predict, in real-time, individual subscribers’ likelihood of walking out the door. The Churn Prediction Service returns a value between 0-100%, which is the probability that a subscriber is likely to churn. Using this new service gives telcos the ability to launch real-time, targeted retention campaigns to discourage subscribers from leaving. Since it’s available as an API call, and because it’s a service of the public cloud, telcos of any size can use it without buying, installing, and maintaining additional hardware and software. Call the service, get the answer, and then direct the software to take action, as appropriate. It’s super cool.

There will be more data science services we debut over the next several months. Tomorrow at MWC at the MVNO Summit, I’m going to talk about Totogi’s Auto Plan product, which not only uses Churn Prediction, but also has another ML model in development that helps identify ways to increase subscriber spend. Want to learn more? Come to my talk tomorrow, March 1, at 4 pm CET in Hall 7, or reach out to the Totogi team for a demo.

Recent Posts

  1. Cloud burst for disaster recovery
  2. How to transform your customer support with AI
  3. 5 things you gotta do at MWC24
  4. 4 ways to spot #fakeAI
  5. The perfect BSS

Telco in 20 Podcast - Tune in

Get my FREE insider newsletter, delivered every two weeks, with curated content to help telco execs across the globe move to the public cloud.

"*" indicates required fields

This field is for validation purposes and should be left unchanged.

More from TelcoDR