Creating an Omnichannel Banking Experience with Machine Learning on Azure Databricks
3 likes903 views
The document outlines a presentation on creating an omnichannel banking experience using machine learning on Azure Databricks, featuring insights from experts at Česká Spořitelna and Datasentics. It discusses the use of data analytics to optimize digital marketing spending and enhance customer engagement through personalized user experiences. Key takeaways include the importance of connecting data at the individual customer level and utilizing AI models to drive efficiency.
Creating an Omnichannel Banking Experience with Machine Learning on Azure Databricks
- 1. WIFI SSID:Spark+AISummit | Password: UnifiedDataAnalytics
- 2. Creating an Omnichannel Banking Experience with Machine Learning on Azure Databricks Petr Pluhacek (Ceska Sporitelna) Jakub Stech (DataSentics)
- 3. Who is presenting today Petr Pluháček, Česká spořitelna Product owner: • Digital engagement and acquisition squad Responsibilities: • Digital sales • Web platforms and chats • Digital marketing data and analytics • Online customer experience Contacts: • [email protected] • www.linkedin.com/in/pluhacek
- 4. Česká Spořitelna About us: • Almost 200 years history • 4,6 millions customers • 10 000 of employees • Part of ERSTE group • Driver of innovation in the group • Undergoing agile transformation ČS Mission: “We are your lifelong guide on the path to prosperity, and in this way we contribute together to the prosperity of the whole country. When someone believes in you, you achieve more."
- 5. Who is presenting today Jakub Stech, DataSentics Data Science architect in: • DataSentics and Digi data team in CSAS Responsibilities: • Translate business problems for data science team • Personalizing user experience using data and machine learning approaches • Building and employing the analytical platform in cloud Contacts: • [email protected] • www.linkedin.com/in/jakubstech
- 6. DataSentics – European Data Science Center of Excellence based in Prague • Machine learning and cloud data engineering boutique • 50 data specialists (data science, data/software engineering) • Helping customers build end-to-end data solutions in cloud • Incubator of ML-based products • Partner of Databricks & Microsoft • Make data science and machine learning have a real impact on organizations across the world • Bring to life transparent production-level data science.
- 7. Digital marketing spend optimization Case study #1
- 8. ADVERTISMENT WWW.CSAS.CZ INTERNET BANKING OTHER SOURCES BRAND AWARENESS SALES (ONLINE AND OFFLINE) CARE Digital marketing interactions
- 9. How to improve digital marketing spend effectiveness Business challenge:
- 10. 10
- 11. ADVERTISMENT WWW.CSAS.CZ INTERNET BANKING OTHER SOURCES BRAND AWARENESS SALES (ONLINE AND OFFLINE) CARE Digital marketing interactions
- 12. Adform is one of the world's largest private and independent advertising technology companies and is best known for its seamlessly integrated DSP, DMP, and Ad Server. ADVERTISMENT
- 16. A viewable impression is a standard measure of ad viewability defined by the International Advertising Bureau (IAB) to be an ad which appears at least 50% on screen for more than one second.
- 17. Decrease costs for visible seconds Better specification of business challenge
- 19. Robot score (Adform definition) Soft fraud score (own definition) Multiple impressions CTR (too low/high) Visibility % Visibility time SEZNAM.CZ; BID 0.2 ; NOVINKY.CZ; BID 0.6 ; IDNES.CZ ; BID 0.1 ; BLESK.CZ ; BID 0.4 ; … Metrics of domain model
- 20. • API • BigQuery • SFTP • CSV • Packages • JSON • Database dump! • Web Pages • … Daily download, transformation and scoring jobs 4+ BLNs rows in 12months! Data sources
- 21. Automated download MS Azure CSAS Storing the data in Data Lake Results stored into Data Lake Scoring new domains Whitelists, bid multipliers, cookies lists, blaclicks, … Automated update SEZNAM.CZ; BID 0.2 ; NOVINKY.CZ; BID 0.6 ; IDNES.CZ ; BID 0.1 ; BLESK.CZ ; BID 0.4 ; … Fully automated pipeline
- 22. 23% Desktop 28% Mobile Increased effectivity of 1 EUR:
- 23. Omnichannel Banking Experience Case study #2 23
- 24. Leading every client to prosperity = Data-driven advisory based on clients needs and real-time situations 24
- 25. … is not easy in bank Customer- centricity…. Low frequency of interactions between a client in offline channels
- 26. 100 Things, We touch our phones 2,617 times a day, says study Around 100 sessions every day…
- 27. Offline vs. Online Typical CRM data • Age/sex/address, policy history, policy configuration, claim history, sales channel, … • Static, mostly long-term behaviour • Facts and transactions • Well structured, easy to process with traditional tech Digital „footprints“ • Ad interactions (wider internet behaviour), web interaction (own sites), mobile apps, external/partner data, … • Dynamically changing, reflecting short and long-term needs • Uncertainty, fragments about interests, behaviour, lifestyle • Enormous data (B+ ads, M+ visits of website…), messy, unstructured, changing interfaces
- 28. 1 Ad Interactions (what the person is interested in across the internet) Own website interactions Emailing / SMS / Push Siloed customer behaviour data Classic client profiles 4) Limited customer experience 2) Missing environment for data analytics and machine learning Classic CRM / data processes Branches & sales networks Transactional data / product data Callcentrum data / call logs Digital campaign management tools Classic campaign management tools 1) Missing connection between digital and CRM 3) On-premise environment is lacking customer data from digital Mobile app interactions Client portal / Internetbanking interactions 3rd party data, voice, text, image, geo data, etc. Digital engagement (3rd party) Offline vs. Online
- 29. 1 Ad Interactions (what the person is interested in across the internet) Own website interactions Emailing / SMS / Push Non-client & client behavior Classic client profiles Automatic optimization, personalization of customer journeys Machine learning Your Customer Engagement 360° Platform (CSAS) Classic CRM / data processes Automatic signals for classic channels Branches & sales networks Transactional data / product data Callcentrum data / call logs Connecting the data on individual customer level Digital campaign management tools Classic campaign management tools New opportunities Higher efficiency Mobile app interactions Client portal / Internetbanking interactions 3rd party data, voice, text, image, geo data, etc. Digital engagement (3rd party) AI-augmented Customer Engagement 360°
- 30. CASH LOAN MORTGAGESAVING ACCOUNT Non-client & client behavior Classic client profiles Connecting the data on individual customer level
- 31. Digital data based sales signals Mortgage / Loan / Saving account / Investments … What product?
- 32. OK, Loan… but what message? Digital data based sales signals
- 33. Digital data based sales signals
- 34. MACHINE LEARNING Digital data based sales signals
- 35. Call script: “Client is interested in taking loan. He is planning wedding according to his activity on internet.” Digital data based sales signals
- 36. Standard classification models (GLMs as benchmark, GBTs in production) ML(lib) part TEXT REPRESENTATIONS CUSTOMER PATHS NUMERIC REPRESENTATIONS
- 37. Loans predicted by Adform data 4x Higher probability Words with predictive power: loan, car, moto, wedding, …
- 38. ADFORM WEB DMP CRM CALL CENTRES … ADFORM GA CRM Automated downloadMS Azure ČS Storing the data in Data Lake Results stored into Data Lake AI monitoring dashboard Model (AI) re-training Scoring new data using existing model (AI) Whitelists, bid multipliers, cookies lists, blaclicks, … Automated update Transforma tion Results (bids, audiences, …) Fully automated pipeline
- 40. 50% Improvement Benchmark: Offline sales signals for mortages sales: call centre 10% sucess rate Current: Digi data enhanced signals for mortages sales: call centre 15% sucess rate
- 41. Take aways • No extra tech, just extend platforms with AI models • Connecting the data on individual customer level is crucial • Clear business specifications and convincing results are essential
- 42. Thank you
- 43. DON’T FORGET TO RATE AND REVIEW THE SESSIONS SEARCH SPARK + AI SUMMIT