Ambiata.com Personalisation with Predictive Analytics Dr Rami Mukhtar National ICT Australia May 2013
Personalisation in Enterprise
Commoditisation
Customers Today Real-time influence External unpredictable
The Latent Truth Who do they deal with? What do they buy? Where do they withdraw money? Do they keep a promise? Measurements Customer Why they call the bank? When/how they use digital channels? What do they need? What do they want? Would we insure them? Who do they work with? Are they a credit risk? Will they repay debt? Latent Truth Business Value 5
Personalised messaging 6
Inferring Latent Truth ATM withdrawals Ave. Balance CC Spend Num. Products CC Spend Salary Amt. Segmentation Predictive Model 7
Income protection insurance 32 yo Best Response Rate 34 yo 8
Income protection insurance What went wrong? 9
Income protection insurance Felt secure Concerned Could we have interfered this? Better Targeting 10
Income protection insurance $ $ Felt secure Concerned The right signals were in the data! 11
FSI Data Assets Small Volume 10 MB/day Customer Datamarts CRM logs Data Warehouse Voice to Text/ IVR Payment Journal Raw WWW logs Smartphone application telemetry Authorization Logs Big Volume 10 GB/day Machine-Machine transaction logs Human Scale Transactional Scale
Feature Generation Free Text Transactional Mobile Web Bi-grams Topics Sentiment Statistics Category aggregates Entity relationships Spending behaviour Location entity statistics Behavior vs. location Time of day activity Browsing behavior Dwell times Conversion statistics 13
Machine Learning Signal Extraction Customer Modeling 100,000 s + variables Source Assets Customer Context/ Behaviour Segment of one 14
The right applications Marketing Black Box Immediate Accurate Volatile Sales Product Propensity Churn Recommendations Customer Influence Satisfaction Personalisation Profitability Pricing/ Product Design Pricing Bands Business Insight Usage Event Risk Finance Default Risk Interpretable Mature Valid Stable Risk Underwriting Liability Ingest Everything Breadth Agility Big NICTA Data Copyright Storage 2013 Data Lineage Governance Reliability Data Warehouse 15
Personalisation for FSIs Recommendations/ Search Next Best Offer Continuing Conversation Targeted Marketing Loyalty Pricing Problem Escalation a relevant customer experience automated decision making Customer 16
Traditional Analytics vs. Personalisation 1990 s Data Warehousing Insights Human Scale Segmentation Personalization Transaction Scale Insights 2010 s Big Data
Making it a success 18
Step 1: Plumbing, Platform, People CRM People IVR Payments CC Auth Collections Plumbing Plumbing 19 Source Systems Platform Channels 19
Step 2: Right business problems Business Support Privacy Clear business need Business process supports automated decisioning Can implement in time frame and cost of relevance Clear understanding of how to measure success Privacy policy permits it Unlikely to have negative sentiment Does not undermine brand Data Assets Have access to the right data assets Permission or rights to fuse assets Permission to fuse with 3 rd party assets 20