Driving digital engagement: Claro uses AI to prioritize quality over quantity
01 Aug 2019
Driving digital engagement: Claro uses AI to prioritize quality over quantity
A TM Forum Catalyst is demonstrating how telcos can increase digital engagement with customers by providing more targeted products and offers through a smart machine learning engine.
As customers increasingly shop via digital channels and technology gets smarter, there is a risk that consumers will be swamped with offers and recommendations. This can have the opposite of the desired effect. Highly personalized targeting is crucial in the digital world.
A TM Forum proof-of-concept Catalyst project is exploring how machine learning (ML) and artificial intelligence (AI) could make this a reality for telcos. In the Machine Learning Engine for Product and Sales Management through Digital Channel project, the champion is the communications service provider (CSP) Claro Colombia. Champions set a real-world business challenge and work alongside participants – in this case, BluePrism, everis, IBM and Salesforce – who collaborate to find a rapid solution.
Claro offers several pre-pay services where customers can buy minutes, data packages and bundles for streaming content such as video and music. The CSP wants to ensure customer engagement and improve customer experience in the pre-pay segment.
Pre-pay customers typically belong to and switch between three main propensity profiles: potential churn, potential growth and stable. According to these, actions and offers can be recommended directly and automatically to customers to generate value or avoid negative impact.
An ‘action’ could be activating a CSP’s music app or unused benefit, or changing the product mix. Examples of offers or products could be to upsell more data, migrate to a specific post-paid plan or buy a TV or gaming console.
A machine learning algorithm can quickly ascertain and predict which profile group a customer belongs to, what products could be offered to maximize customer value, or what the next best action is to avoid churn.
Further, the team’s solution doesn’t just provide a next-best action/next-best offer engine based on predefined business rules. Rather, it shows how decision and recommendation policies can be “dynamically and autonomously” optimized as conditions changed, in real-time.
The Machine Learning Engine for Product and Sales Management through Digital Channel project team demonstrated their proposed solution at the TM Forum’s Digital Transformation World event in Nice in May.
The team started out using value-based algorithms like Q learning, as well as policy-based algorithms like multi-bandit, contextual bandit and full reinforcement learning (RL). However, the project then decided on a contextual-bandit game algorithm because this takes into account the context of the subscriber as an input to select and maximize recommended offers.
Customer interaction takes place through an app which uses a cognitive agent that interacts with a machine learning engine (MLe) to orchestrate the customer’s often-complex interaction with: ML results, a marketing management service, product and customer catalogs plus a robotic process automation (RPA) service that helps to execute the actions.
Through the app, a pop-up indicates when new recommendations and offers are available to the customer, inviting them to learn more via a virtual assistant. If a customer likes the suggestion, they can buy it, but if not they can explain why, which in turn feeds back into the machine learning and a further recommendation is offered.
Vergara describes the solution as “an intensive technological landscape”. It comprises eight platforms including a real-time ML engine, ML repository, virtual assistant, mobile application, product and sales catalog, campaign management, API manager and RPA, all integrated with industry-standard Open APIs from TM Forum.
The team also relied on the Open Digital Framework, which includes the TM Forum Business Process Framework (eTOM) Nine business processes were used to map sales, marketing and product interactions, while TM Forum’s CurateFX digital ecosystem management tool was used to author the business/administrative infrastructure of the solution, exploring issues such as how to approach a customer, which stakeholders are involved and what their roles are.
In all, the solution incorporates six ML/AI algorithms including churn propensity, post-paid migration propensity, upsell/cross-sell propensity, unsupervised clustering, contextual bandit game, and natural language processing.
The solution also allowed the Catalyst to create four TM Forum use cases to feed back to the collaborative community, including encouraging pre-paid to post-paid conversion, product performance optimization, churn risk prediction for customer retention, and personalized offers – again, to safeguard customer loyalty.
The project made use of software from Salesforce for the sales, marketing and customer relationship management (CRM) activity, while RPA specialist Blue Prism was responsible for turning manual processes into fully automated ones.
everis was responsible for handling all customer communication. “The everis machine learning engine is at the heart of the whole recommendation process, where all events come through IBM’s Event Streams, and using big data tools such as Apache Flink and Redis, we can have streaming analytics tools to handle all events, the processing of all the recommended actions and exposing this to the customer,” Vergara said. “Each recommendation affects the next one. It’s all real-time.”
In the future, thanks to IBM’s API Connect, CSPs will be able to improve the customer experience further by recommending not only their own products and services to customers but those from elsewhere too.
For more details, watch the video presentation:
As customers increasingly shop via digital channels and technology gets smarter, there is a risk that consumers will be swamped with offers and recommendations. This can have the opposite of the desired effect. Highly personalized targeting is crucial in the digital world.
A TM Forum proof-of-concept Catalyst project is exploring how machine learning (ML) and artificial intelligence (AI) could make this a reality for telcos. In the Machine Learning Engine for Product and Sales Management through Digital Channel project, the champion is the communications service provider (CSP) Claro Colombia. Champions set a real-world business challenge and work alongside participants – in this case, BluePrism, everis, IBM and Salesforce – who collaborate to find a rapid solution.
Digital engagement
Claro offers several pre-pay services where customers can buy minutes, data packages and bundles for streaming content such as video and music. The CSP wants to ensure customer engagement and improve customer experience in the pre-pay segment.
“Customers could easily receive around 30, 40 or even 50 recommendations a month,” explains Dahyr Vergara Suárez, Telecom Manager, everis. “What we are trying to do is get much higher customer acceptance with [fewer] recommendations.”
As well as improving the customer experience, this would help to drive up the CSP’s revenue, he said.
“Digital engagement is the name of the game,” adds Gustavo Diaz of Claro Colombia. “And this is what we are dealing with here. We can identify an offer that is specific for our customer.”
The solution
Pre-pay customers typically belong to and switch between three main propensity profiles: potential churn, potential growth and stable. According to these, actions and offers can be recommended directly and automatically to customers to generate value or avoid negative impact.
An ‘action’ could be activating a CSP’s music app or unused benefit, or changing the product mix. Examples of offers or products could be to upsell more data, migrate to a specific post-paid plan or buy a TV or gaming console.
A machine learning algorithm can quickly ascertain and predict which profile group a customer belongs to, what products could be offered to maximize customer value, or what the next best action is to avoid churn.
Further, the team’s solution doesn’t just provide a next-best action/next-best offer engine based on predefined business rules. Rather, it shows how decision and recommendation policies can be “dynamically and autonomously” optimized as conditions changed, in real-time.
The Machine Learning Engine for Product and Sales Management through Digital Channel project team demonstrated their proposed solution at the TM Forum’s Digital Transformation World event in Nice in May.
The team started out using value-based algorithms like Q learning, as well as policy-based algorithms like multi-bandit, contextual bandit and full reinforcement learning (RL). However, the project then decided on a contextual-bandit game algorithm because this takes into account the context of the subscriber as an input to select and maximize recommended offers.
“This had to be implemented in real time using open source,” explains Andrés Estupiñan, Advanced Analytics Manager, Claro Colombia. “It was a really complex technical task to build an algorithm like this.”
It’s all in the app
Customer interaction takes place through an app which uses a cognitive agent that interacts with a machine learning engine (MLe) to orchestrate the customer’s often-complex interaction with: ML results, a marketing management service, product and customer catalogs plus a robotic process automation (RPA) service that helps to execute the actions.
“Maximizing value is not the only benefit we can get from a solution like this,” comments Estupiñan. “Actually, the best benefit is knowledge – knowledge about the way the market reacts to an offer; knowledge of different solutions and products that we have for different customer segments; and knowledge about how to maximize the value of our product mix.”
Through the app, a pop-up indicates when new recommendations and offers are available to the customer, inviting them to learn more via a virtual assistant. If a customer likes the suggestion, they can buy it, but if not they can explain why, which in turn feeds back into the machine learning and a further recommendation is offered.
Sum of its parts
Vergara describes the solution as “an intensive technological landscape”. It comprises eight platforms including a real-time ML engine, ML repository, virtual assistant, mobile application, product and sales catalog, campaign management, API manager and RPA, all integrated with industry-standard Open APIs from TM Forum.
The team also relied on the Open Digital Framework, which includes the TM Forum Business Process Framework (eTOM) Nine business processes were used to map sales, marketing and product interactions, while TM Forum’s CurateFX digital ecosystem management tool was used to author the business/administrative infrastructure of the solution, exploring issues such as how to approach a customer, which stakeholders are involved and what their roles are.
In all, the solution incorporates six ML/AI algorithms including churn propensity, post-paid migration propensity, upsell/cross-sell propensity, unsupervised clustering, contextual bandit game, and natural language processing.
The solution also allowed the Catalyst to create four TM Forum use cases to feed back to the collaborative community, including encouraging pre-paid to post-paid conversion, product performance optimization, churn risk prediction for customer retention, and personalized offers – again, to safeguard customer loyalty.
Participants’ roles
The project made use of software from Salesforce for the sales, marketing and customer relationship management (CRM) activity, while RPA specialist Blue Prism was responsible for turning manual processes into fully automated ones.
everis was responsible for handling all customer communication. “The everis machine learning engine is at the heart of the whole recommendation process, where all events come through IBM’s Event Streams, and using big data tools such as Apache Flink and Redis, we can have streaming analytics tools to handle all events, the processing of all the recommended actions and exposing this to the customer,” Vergara said. “Each recommendation affects the next one. It’s all real-time.”
In the future, thanks to IBM’s API Connect, CSPs will be able to improve the customer experience further by recommending not only their own products and services to customers but those from elsewhere too.
For more details, watch the video presentation:
