Large-scale and continuous customer feedback analysis at Avira (Manuel Eugster, Avira)

At Avira, the Customer Insights Research group takes responsibility to make the user’s voice heard within the company. With a holistic data-driven view of the customer lifecycle – based on internal and external human- and machine-generated data – the challenge is to transform these massive amounts of (real-time) data into descriptive knowledge about our customers and products.

This talk will present the pipeline for Avira's continuous customer feedback system. In every product, customers are able to give structured and unstructured feedback on different time points during their user experience. The data is processed and analyzed in near real-time, and insights are distributed to relevant stakeholders within the company. For example, an automatic key driver analysis based on linear models continuously monitors the positive and negative customer drivers. Behind the scenes, R is used as analysis tool as well as ETL tool operating on different internal services and data bases.

Presentation given at the 4ländereck Data Science Meetup: https://www.meetup.com/4laendereck-Data-Science-Meetup/events/234041619/

Transcript

1. Large-scale and continuous customer feedback analysis at Avira 4ländereck Data Science Meetup Manuel Eugster – 05.10.2015
2. Customer Insights Research Mission: • Make the user’s voice heard in the company. • Create a data-driven view along the customer lifecycle by combining machine performance and events with human behavior. Data source: Internal vs. External Data format: Structured vs. Unstructured Data generation: Human vs. Machine
3. We continuously collect customer feedback at different touch points during the customer lifecycle
4. 1 2 3 Net Promoter Score (NPS) Customer Satisfaction (CSAT) Open Text
5. Why “Very Good”? CSAT Open text • viruses • system • upgrade • problem/update ? • Mac/MacOS 10.121 Tags Natural language processing 2 Score
6. Why “Very Good”? CSAT Open text • viruses • system • upgrade • problem/update ? • Mac/MacOS 10.121 Tags Natural language processing Influence 2 Score
7. Why “Very Good”? CSAT Open text • viruses • system • upgrade • problem/update ? • Mac/MacOS 10.121 Tags Natural language processing Influence 2 Score Statistical modeling
8. Customer Centric & Data Driven
13. Behind the scenes
14. Google Translate API Data Warehouse Translator Getter Tagger Modeller IBM Text Analytics
15. Statistical model for key driver analysis
16. “Essentially, all models are wrong, but some are useful.” -- Box, George (1987)
17. Data
18. Linear Regression Model or Old but Gold Gauss, 1809 Legendre, 1805 Data “Influence” of tags on customer satisfaction
19. Bootstrapping or How can we (at least somehow) trust our customers? Confidence intervals based on 100 bootstrapped statistical models
20. Model evaluation
21. Model evaluation
22. Model evaluation or Is the customer happy and do we make more Money?
23. Large-scale and continuous customer feedback analysis at Avira Data Science @ Avira http://avira.github.com/dsblog Manuel J. A. Eugster http://mjae.net