13 times better hitrate with Big "Enough" Data

Pål Sundsøy, Johannes Bjelland Telenor Group Research

Transcript

1. 13 times better hitrate with Big “Enough” Data Pål Sundsøy, Johannes Bjelland Telenor Group Research IT-Tinget, 26.09.2013
2. Among the major mobile operators in the world 148 million mobile subscriptions 32 000 employees Present in markets with 1.6 billion people Telenor Group holds an owning stake of 33 percent (economic) and 43 per cent (voting rights) in Vimpelcom Ltd, operating in 17 markets
3. Our customers generate an increasing amount of information in our systems What’s in it for Telenor?
4. Bringing the Big Data Buzz down to earth • Telecom Data • Big Data Analytics in Telecom - state of the art today • Prediction is money Big Data Use Case Predicting the next mobile internet users in Asia AGENDA
5. We live in a world of Big Data… Number of Tweets • 400 million tweets per day (12TB) YouTube • 100 hours of video are uploaded every minute Facebook • 300 million pictures uploaded - per day • Deals with about 105 terabytes of data – at any given hour Telenor Pakistan • Over 700 million voice and sms sessions collected every day • 1 billion social relationships stored every month
6. 6 A number - Caller IMSI: SIM card Cell_ID: Location TAC: Handset Type: Call, SMS, Data, etc For each call, sms and data session: hundreds of data points are stored Date & time B number – Receiving party Data volume
7. Inventory & Sales Usage Top up/Payment Network Performance Call Center Customer Interactions DWH & BI Each BU generates huge amounts of data Other data sources Today we are not leveraging sufficently on our customer data
8. A technical definition of Big Data according to Gartner What do we learn from the data? Volume Velocity Variety GB batch Table MB TB PB Database Photo Web Audio Social Video Unstructured Mobile Periodic Near real-time Realtime
9. 9 08/10/2013 Learning from data: Mobile Phone Data Research Researchers are realizing that mobile phone data can be a valuable data source Everyone has a mobile phone – Suddenly we have a tool to measure whole populations. Study mobile phone data can give us new insight into human sociology Telenor Research collaborates with MiT, Harvard and NorthEastern University
10. Research on human interactions: By analyzing anonymized CDR-data we can map out a proxy for the social network among our customers Social connection 10 (built from traffic data)
11. Measuring social influence and ‘word of mouth’ The adoption network tells us about the viral spreading 11 Social network Adoption network «who influence whom» Social network + smartphone adoption history 1 2 3
12. Q307 Q407 Q108 Q208 Q308 iPhone is spreading like wild fire! iPhone not yet available in specific market: “Cracked iPhones” bought in the US. 2G release in US We found a core of Apple users that were loyal to the brand and well connected to other Apple users: The ‘Apple Tribe’ Comparing and visualizing the social spreading of products on a large-scale social network P.Sundsøy, J.Bjelland, K.Engø Monsen, G. Canright, R. Ling. The influence on Technology on Social Network Analysis and Mining, Volume 6, 2013, pp 201-225 Presented at NetMob conference, Massachusetts Institute of Technology
13. = Android customers = Apple customers Bergen Norway Apple was quickly winning the cities Equally many Apple & Adroid customers 80-100% more Apple customers than Android 80-100% more Android customers than Apple Oslo
14. Bringing the Big Data Buzz down to earth • A useful definition of Big Data • Big Data Analytics in Telecom - state of the art today • Prediction is money Big Data Use Case Predicting the next mobile internet users in Asia AGENDA
15. Boosting Mobile Internet uptake in Asia with prediction and SMS marketing This Pilot is a collaboration with the Mobile Internet Asia project in Digital Services Motivation • For many in Asia, the mobile phone is their only gateway to the Web. • Many customers have internet capable phones, but do not use them • The business unit is using state of the art Below The Line marketing process
16. Selecting the right campaign target groups is key to maximize Campaign revenue • 6000 yearly SMS campaigns effectively boost customer revenues • Number of campaigns cannot be pushed further • Contact rules: Max 1 offer each 14 days • Efficiency of campaigns can be improved with a big data approach Risk of spam: Customer attention is valuable 16 08/10/2013
17. Machine Learning assists us in selecting optimal target customers from huge data sets Data sources • Traffic usage data • Subscription data • Handset Features • Location • Handset switching • VAS usage 300 variables 40 000 000 customers ? Who are most profitable targets for SMS campaign
18. Its impossible for a human to relate to all these data (!)
19. In addition to 300 ‘ego-based ‘ customer features we test the effect of using aggregated data from the customer calling circle Our earlier studies have shown that friend’s behavior can be used to predict actions by ego Social connection (built from calling data)
20. The predictive model learns from existing cases of data conversion Non-convertors ‘Negatives’ Natural Data Convertors ‘Positives’ 2-6 months back: Use Historical data Non Data Customers today Create model Find patterns identifying the data convertors based on historic data Model deployment Use the patterns to identify likely adopters Identify and Run Campaign on 200k most likely adopters Today: Present time data *Offers are 15 MB & 99 MB data packages offered for half-price
21. Validation Scoring The Predictive Model is not a ‘black box’, but algorithms put together and tuned by the analyst Final Output Complex historic data input • This is the actual model for this pilot • All the boxes are model interaction points • 80% of the work is data preparation Model Training
22. The prediction model outperforms existing best practice approach – 13 times better than best practice 6,42 3,76 0,50 0,70 0,00 1,00 2,00 3,00 4,00 5,00 6,00 7,00 P7 data pack P9 data pack Hitrate PSPM Microsegmentation 15 MB Data Package 99 MB Data Package Actual Campaign Hit Rate 99% Renewal– the algorithm is optimized to avoid ‘freeriders’ Current best practice Microsegmentation approach Prediction Model
23. Calling circle behavior • High internet usage among friends SMS usage • High usage and spending on SMS Handset features • Edge enabled phones • Recent handset switch Value Added Services • Have paid for incoming SMS • E.g. News 23 08/10/2013 The Model gives new insight by identifying the ‘likely data convertor’- segments and attributes The model combines the attributes to create the final target segment Human Learning would have required a lot of trial-and-error to get similar performance!
24. A C B D We give another offer to 70k customer and ask them to tell friends – no incentive Testing viral marketing in Asia: letting the customers spread the offer Part of ongoing research collaboration with David Lazer and team at Northeastern We have implemented a unique tracable offer mechanisms
25. 6,05 53,37 0,00 10,00 20,00 30,00 40,00 50,00 60,00 % Direct Hitrate % Total Hitrate (direct+viral hits) On average one offer spreads to 8 customers • The offer is sent to 70 000 customers. • 4 300 accepted the offer and forwarded to their friends. • 33 500 friends were also recruited. • Good sustainability: 50% renewal rate
26. The extreme case 1 customer recruits 360 other customers Offer is only sent to the yellow customer. We know that all red adopters originate from this customer – all offers have a unique code This customer is targeted and adopts – and recruits 360 other people No mobile communication! E.g Spreading via facebook! The social network among the mobile internet adopters after 5 days Most important influencers are based in the social core
27. Using Big Data for Social Good Goal: Explore the use of telecom data in the fighting of socio-economic problems Partners : UN Global Pulse Harvard University - Infectious diseases - Predict and control, build real-time early- warning systems - Migration patterns - Improvement of disaster relief in flood or earthquake situations - Macro-economic trends - Predict unemployment & economic crisis Handling privacy is vital! • Telenor Group’s privacy principles are followed • Transparency of data handling is important • Personal data is anonymized before utilized in these projects
28. A ‘Big Data’ company is distinguished, not by how many terabytes it sits on, but by the way the company exploits the data in Business! • Answering business questions via data mining and ad hoc analysis • Using pilots and data driven marketing to let the customers tell us what they want • Collaborating with world leading research environments within data science • Petabytes is not a prerequisite - What we need is ‘BIG ENOUGH’ Data for business Telenor is taking steps toward becoming a Big Data company