Roadmap to Realizing the Value of Telco Data – Opportunities, Challenges, Use Cases- By Srinivasa Ravi, COO, Flytxt at Big Data Monetization Event Singapore 2015
Value Drivers: Rich Data, Deeper Insights, Faster Decisioning, Diverse internal and external monetization use cases
Data integrated and analyzed
Advance analytics to derive deeper insights
Ability to take decisions faster and execute actions on time
Ability to support diverse use cases for value generation – internal and external monetization
Realizing optimal value of data is a journey undertaken from data streams to value streams. It passes through
Integrate Big Data, Fast Data from multiple sources – customer, usage, network, location, device etc
Performing advance analytics to derive deeper hidden actionable insights
Generating incremental revenue through faster decisioning – within internal workflows
Sharing these insights to external workflows, to drive adjacent services and generating revenue streams
Measure the economic impact, and optimize iteratively with better data sources, reinforced learning, iterative decision making, etc
It is important to have Big Data processing capabilities, through taking care of high volumes, high variety of data and ensuring high data veracity (quality)
It is equally important to ensure real time data processing through Fast Data capabilities – through billions of events per day – reduces time to marketing action significantly
Operators need a diverse set of technical capabilities to ensure this, including technologies such as Hadoop (for Big Data processing), In-memory database (for real-time trigger generation) and Spark (for running machine learning on large data sets).
Advanced analytics involves employment of various types of analytics, different packaged analytical models, and smart visualizations. How quickly insights can be served and used by business users holds the key in faster decisioning and value creation.
Analytical models are created with multi-dimensional analysis using advanced techniques ranging from descriptive and exploratory to heuristic, predictive and prescriptive. Different types of data like profiles, usage, product, location, social, handset, network, campaign, RoI, etc. are integrated and analysed with a wide range of algorithms to build reusable analytical models. Packaged analytics library has a collection of models available for business users to drive various internal and external data monetization use cases.
The integrated analytics framework can generate interactive smart visualizations that help business users to analyse and interpret complex data clusters, patterns and correlations with ease. These visualizations are also built into specific packaged analytical models consumed by multiple business applications to enable faster decisioning.
Value of data can be derived from internal use cases and external use cases.
Leveraging insights for decisioning across business workflows to enhance customer experience, increase revenues and reduce churn through internal workflows – Internal Monetization
Few examples – Marketing Campaigns, Customer Care Touch Point Personalization, Sales and Distribution Optimization etc.
Applying right analytics , transforming data to deeper consumer insights, and sharing these insights to third parties to drive adjacent services and use cases – External Monetization
Few Examples – Mobile Advertising, Footfall Analysis to retailers, Handset market share analysis etc
Let us look at how business users can make use of insights quickly to make decisions faster with few examples.
Indian CSP faced the challenge of ensuring right communication reaches right subscribers – as many subscribers opted out of communication due to irrelevant offers
Micro-segmentation using multi-dimensional data analytics was the solution – here drilled down usage analysis is done and different usage leg wise details are available to slice and dice the segment base. Here from the UI users can choose any of these granular usage criteria so as to segment it. There could be even machine learned heuristic persona like ‘Heavy Data Users’ which can also be used while segmenting the base. The more granular you can go, the more accurate would be your understanding of each segment.
Such micro-segmentation helps in having an in-depth understanding of the usage preferences and usage patterns so as to decide right offers for each segment
Inactive subscribers were approached more aggressively and active subscribers were given less aggressive offers
8 million unique subscribers were covered in a month
Rejection rates at customer care touch points was high for the Indian CSP
Every agent could see many offers mapped to a single subscriber – hence no personalization was possible
Prescriptive analytics based offer recommendation and recent touch point behavior based offer prioritization was done with machine learning algorithms. These offers were then served to customer care touch point for individual subscribers visiting customer care touch points. Here Agents could get the right offers with a single click, reducing the call hold time by 30%. Machine learning driven offer recommendation and prioritization made the life easy for agents.
Here the conversions improved to 11% over the customer care touch point.
In this use case, the CSP wanted to win-back subscribers who has high propensity to churn
The CSP could do a 3-step filtering process based on propensity to churn, propensity to win-back and uplift score. Now these scores are made available as segmentation criteria for business users when they plan out a retention campaign.
Propensity values and scores were derived through multi-dimensional analytics based on profile, usage, recharge and behavioral data
Such a campaign not only reduces churn, but also increases revenue and margin as not the same offer and discount is given to all. It is varied as per the uplift propensity.
Case of external monetization on multi-dimensional analytics and insights
Brand was looking to target a niche segment quickly so as to promote their new models before competitors could react with counter offers. They wanted to target young population who uses data services frequently and who are still using relatively lower end models from competitors.
They could also include location criteria so as to restrict campaigns to those regions where brand has retail outlets.
Any persona can be created through natural clustering and heuristic analytics and these persona are in non PII format, hence ensuring subscriber privacy. These persona models can be readily used by brands and advertisers
The brand was able to target the right niche of subscribers with the right promotional offers with in 2 weeks time, much quicker and less costly when compared to ATL campaigns
Once the use cases are established, it is important to measure the impact of use cases effectively, and use the measurement to iteratively improve the workflows
Accurate impact measurement is achieved through techniques like control group
Feedback is given back to analytics which in turn impacts actions/decisions in the cyclical process. This is important as we can iteratively optimize with
Better/different data sources
New insight models/reinforced machine learning
Iterative decision making – changing the product, channel, price etc
Value of data depends on how much economic value can be derived using internal and external monetization
Internal monetization can potentially generate 2-7% economic value for CSPs
External monetization can generate 2-5% economic value