Banking fraud detection and prevention system, based on intelligent data-driven decisions

Transcription

1 Banking fraud detection and prevention system, based on intelligent data-driven decisions George Damianov SEUGI' - Paris, France 1

2 CONTENT INTRODUCTION PRE-DEVELOPMENT CONSIDRATIONS DEVELOPMENT MANDATE APPROACH OUT-OF-LINE INFERENCE FURTHER ANALYTIC CONSIDERATIONS TECHNICAL CONSIDERATIONS LOGISTICS FUTURE SUMMARY SEUGI' - Paris, France 2

3 INTRODUCTION Company profile as at end of 2001: Total assets R395.8bln; Banking assets R306.4bln; Represented in 37 countries; Combined Funds under management R150bln; Total return on equity 39%; Cost-to-income ratio 57.3%. SEUGI' - Paris, France 3

4 INTRODUCTION Knowledge Discovery in Standard Bank Empirical insight generation group-wide, and facilitating the use of it for real business advantage Make the information able to predict and drive the future success Done in a distributed multi-platform SAS environment SEUGI' - Paris, France 4

5 INTRODUCTION SAS ANALYTICAL AND TECHNICAL ENVIRONMENT Econ.Div.(remote host) - Macroecon. Data Base; - ECOSERV. NT Servers (remote host) - SAS (certain modules); - SAS Enterprise Data Miner; - Platform connections. Network EDW (SUN remote host) - Base SAS; - Access to ORACLE; - Platform connections. Mainframe (remote host) - Mainframe SAS; - Access to DB2; - Access to mainfr. systems; - Platform connections. PC Client (Process driver) Peoplesoft (remote host) -PC SAS; SAP (remote host) - Access to PC systems; - Platform connections. SEUGI' - Paris, France 5

6 INTRODUCTION The fraud dimension in banking business Part of empirical insight generation and intelligence, ultimately at a customer level Pathological part of the business, causing significant loss SEUGI' - Paris, France 6

7 PRE-DEVELOPMENT Pre-development considerations Current status-quo (do we need to change it) Business case and functionality Off-the-shelf buy vs. in-house development Decisions and sponsorship SEUGI' - Paris, France 7

8 MANDATE Development mandate Intelligent system with near-real-time reaction, based on datadriven decisions Maintain account, instrument and eventually a customer view, covering any possible fraud dimension Output/reaction Incremental SEUGI' - Paris, France 8

9 APPROACH Approach Agree on a proper typology of banking fraud Explore and implement the cause-effect relationship normalabnormal-fraud Develop best-of-breed out-of-line inference on all levels Develop fuzzy rules engines with what-if capabilities Develop predictive models, based on known fraud cases, to generate potential fraud cases based on propensity Define first steps, first deliverables, exclusions Use the established SAS environment for the intelligence generation SEUGI' - Paris, France 9

10 OUT-OF-LINE Out-of-line inference Data assessment Establish normal behaviour (signatures, bin vectors, Benford s Law) Establish distance measure and decision logic for abnormality generation with tuning parameters Decide how to initialise new and non-established accounts Create initial out-of-line distance and threshold parameters per established account, instrument and customer Establish the updating procedure SEUGI' - Paris, France 10

11 ANALYTIC Further analytic considerations Signatures and the binning procedure Benford s Law Predictive modelling SEUGI' - Paris, France 11

12 TECHNICAL Technical considerations Multi-tiered architecture, realised in Visual Age for Java, Websphere, DB2, XML Updating outside the production system, in SAS Production operations and fraud case generation for viewing by a fraud analyst Performance assessment-, effectiveness- and security subengines SEUGI' - Paris, France 12

13 LOGISTICS Logistics Exclusions and limitations for the first phase The time sequence of events The 4bln. records question The event loader SEUGI' - Paris, France 13

14 FUTURE Future Add products (go wide) Add models (go deep) Go towards real-time (go high) Gradually automate reaction into the legacy systems (go long) Achieve global customer-level real-time automated prevention (go everywhere) SEUGI' - Paris, France 14

15 SUMMARY Summary Fragmented anatomy of a banking intelligent fraud prevention system Implementation deadline for the first phase Cost of first phase R3.5mln., expected revenue from the first phase R25mln. p.a., based on an assumed increase in prevention rate of 25% more than now SEUGI' - Paris, France 15