SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan...

19
© 2013 IBM Corporation Missbrauch und Betrugserkennung Sozialversicherung, Healthcare, Tax & Revenue Erfahrungen mit dem Einsatz analytischer Werkzeuge Richard Dobis Stefan Heimrich Bern, 6 März 2013

description

Swiss eGovernment Forum | 6. März 2013 | Referat Richard Dobis & Stefan Heimrich Der Einsatz analytischer Instrumente und Werkzeuge zum Erkennen von Missbrauchs- und Betrugsfällen im Bereich Sozialversicherungen wurde in Pilotprojekten in Tschechien und Slovakien erfolgreich erprobt. Erste Ergebnisse und Erfahrungen sowie mögliche erfolgsabhängige Vergütungsmodelle werden im Rahmen dieses Vortrages dargestellt.

Transcript of SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan...

Page 1: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation

Missbrauch und Betrugserkennung Sozialversicherung, Healthcare, Tax & Revenue

Erfahrungen mit dem Einsatz analytischer Werkzeuge

Richard Dobis Stefan Heimrich Bern, 6 März 2013

Page 2: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 3

TAXATION

Why talking about Fraud & Abuse Management?

Loose estimates say that the EU loses €100 billion annually in value added tax (VAT) revenues to fraud.

Government Healthcare

Business Outcomes

State government stopped fraud prior to payment and saved on average $220 million annually

Healthcare provider significantly reduced false positive rate from 79 percent to 29 percent

State Government denied more than $60 million in fraudulent refunds in 6 months.

TAX & Medicare

Damage due to fraudulent claims in TAX and Medicare estimated at 23bn $ annualy

TAXATION

Jährlich verlieren die EU-Staaten Milliarden Euro durch MwSt-Betrug. Jetzt hält die Kommission dagegen […] mit neuen Verfahren bei grenzüberschreitendem Handel

IDC Industry Analysis 2012

www.impulse.de 08/2012

Over €1 trillion is spent on healthcare every year across Europe - the potential losses to fraud and corruption are at least €30 billion each year, and may be as high as €100 billion.

HEALTHCARE

SOCIAL

„Die Missbrauchsquote liegt mehrheitlich zwischen 2 und 10 Prozent. Bei jährlich ca. 1,9 Milliarden Franken an bedarfsabhängigen Sozialhilfeleistungen entspricht das 39 bis 190 Millionen” Beobachter, Juni 2012

IDC Industry Analysis 2012

EHFCN – EU-Healthcare Fraud & Corruption Network, 2012

Page 3: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 4

Objectives of this presentation

In the next 28 minutes we would like to: Show the key principles of Fraud & Abuse Management

Show an example of how it works (Healthcare)

Share project experiences and lessons learned (Other PUB)

Answer your questions

Page 4: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 5

Agenda

Why „Fraud and Abuse“ Managment in the Public Sector Our objectives for the next 30 minutes

Fraud & Abuse Management Solution – Key Principles

Case Study – Healthcare Provider Slovakia

Learnings from other engagements in the PUBLIC Sector

Summary

Q&A

Page 5: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 6

Fight against fraud and errors can only be only successful when performed everywhere, by everybody at each level all the time

6 06.03.2013

Moments of truth …

Impede breaking internal regulatory policies

Organiza-tion Compli-ance

Member-ship Granting

Fraud Detection

Fraud Identifi-cation

Fraud Investi-gation

Fraud Discovery Monitoring

Disarm intruders at the door, reject or assign special treatment

Flag & route fraudulent cases at the submission (Pre-select cases)

Help ‘fraud auditor’ during case clarification and adju-dication (Decision)

Investigate, prosecute & recover fraud (Proof)

Perpetually identify new fraud patterns and learn from past frauds

Report and support decision making on adjustment of the processes

… or opportunities for errors

Page 6: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 7

Complexity of fraud plus challenges of social and tax systems requires intelligent use of existing technology

7 06.03.2013

Complex law and process rules, which are difficult to comprehend even for experts

A lot of data - „Needle in a haystack“

Hard to distinguish between „OK case“ and „undetected fraud“

Data privacy requirements often blocks simple checks of decisions (4-eyes principle)

Need for flexibility within the legal frame

Information and motivational imbalances (need for just, transparent and auditable process and decision rules)

Challenges of Social and Tax Systems

Fraud and Abuse Management Solution

Impede breaking internal regulatory policies

Organiza-tionCompli-ance

Member-ship Granting

Fraud Detection

Fraud Identifi-cation

Fraud Investi-gation

Fraud Discovery Monitoring

Disarm intruders at the door, reject or assign special treatment

Flag & route fraudulent cases at the submission(Pre-select cases)

Help ‘fraud auditor’during case clarification and adju-dication(Decision)

Investigate, prosecute & recover fraud(Proof)

Perpetually identify new fraud patterns and learn from past frauds

Report and support decision making on adjustment of the processes

Complex Fraud Management Process Advanced Analytics Anomalies in behavior

signal for thorough investigation

Detection models mimic medical diagnostic approach easy to localizable fuzzy logic segmentation

Peer group analytics self adaptation to new situation

Prioritization and scoring focus to most suspect cases

Embedded in overall solution

Information Governance

Integration and Collaboration

Productivity Tools (Case Management, Document Management)

Page 7: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 8

Integrated IBM Fraud and Abuse Management Solution – Reference Architecture

FAMS Drill Down Component

IBM Drift

Scoring Component

iLOG Event & Rules Engine

Case Manager

DB

i2

Live & Simulation

Data W

arehouse &

Source System

Claim Execution

Claim Cancellation

Other Options

FAMS

real time messaging

batch

real rime messaging

real time messaging

Messaging /batch

real time messaging batch

Drill Down GUI

real time messaging

Reporting

real time messaging

real time messaging

JDBC

JDBC

batch

Historical Data

Scores

Profiles

Industry specific solutions - software plus mathematical method plus detection models (content) - exist for eg tax, social, healthcare, procurement, marketing, IT security, expense management,

Page 8: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 10

Slovakia Health Care Insurance Company - Situation

Client wanted to improve effectiveness of processing claims - Final decision: "pay" or "investigate“

Client heard from IBM sellers how technology is able to provide quick results

We were given data and really short deadline to apply IBM technology and methodology on the client data

- one month for delivering results,

- from sending zipped 5GB data.

Page 9: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 11

Other comparable physician

Physician in small surgeon office in a small town

Physician in the small local hospital

Example 1: Analytical view detection model on the abuse of opiates

11

Page 10: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 12 12

Possible “bad” explanation: Physicians wants to increase income to the average income level, therefore they overacting – more treatments, more drugs.

Example 2: Discovery of new fraudulent or erroneous pattern

Page 11: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 13

Slovakia Health Care Insurance Company - Results

Project delivered results in the given one month timeframe

Client-acknowledged abnormalities (observations) have been raised in data that passed auditing process (client did consider “fraud-free”)

Based on the one month proof of concept, the financial effects have been extrapolated to the full scope

- +50% increase in success rate of fraud investigation accepted as a realistic target

Client asked to submit offer for implementation project

13 06.03.2013

Page 12: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 14

Lessons learned from other engagments

Client Situation Result Lessons Learned Health- care Provider (Slovenia)

Understand potential of technology in small market (less than 30 player)

If peer group is small, the search for anomalies is hard.

New dedicated detection model is difficult to invent

Detection models developed from foreign countries and different law systems was successfully applied.

Client used observation as a trigger for further investigation

“Models and experience" from foreign countries can be successfully transferred and leveraged

Ministry Office Procurement (Czech Republic)

Client wanted to empower controlling department - who oversees the procurement of the small items and services

Applied full cycle: development of detection models, investigation and interpretation of the results, suggestion of the correction action, refinement of the detection models (POC)

Identified people that used the procurement system wrongly, identified persons that are suspect of “cooperation with specific companies”

People are able to acquire the knowledge to use the FAMS-solution quickly

Collaborative approach to model building is important (industry experts & statistics / tool experts)

Ministry Office Taxation (Hungary)

Client wanted to improve results against tax evasion.

For the PoC, tax type “company income tax” has been selected

Team identified set of suspect companies; a few these companies were already suspected and investigated by officials

Officials adopted project recommendations to investigation schedule

Project ongoing, discussions for full implementation and VAT Tax PoC under way

FAMS detection successful even with basic data

Collaborative approach to model building is important

Page 13: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 15

Sharing Risks and Reward for Czech Ministry and Social Services Administration Situation and Objective

- Improve targeting and effectiveness of benefits paid within the social system for decisions of medical assessment service

- Ministry wanted to support medical assessors which asses health status of applicant for social benefits for disabled persons.

- Not just fair, but also perceived as fair system (Equal assessment for everybody - nobody receives more than others, but nobody receive less then ought to receive)

Solution: - End-to-end services (hosted infrastructure and software)

- Risk & reward model • No one-time implementation costs charged

• IBM rewarded a percentage of the savings achieved

• Mechanism ensures that savings have to be acknowledged by the client expert

- Independent auditor

Lessons Learned - Extrapolated (agreed) annual savings: millions of EUR

- Risk and Reward Model needs to be clearly defined – the devil is in the detail, examples • How to distinguish between “savings” client identifies vs. project team identifies

• How to measured?

• What happens if “increase savings” are in conflict with other goals (eg. imperative to close the case)?

Current state: Mutual agreement to proceed - however, process currently on hold - Competition appealed against the selection process; will be restarted

15

Page 14: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 16

Summary

It is possible to achieve results in a short time frame Concepts and FAMS solution are applicable for different public sector

situations - Social Security - Tax - Healthcare

There are different and flexible ways to engage - Retrospective analytics with historical data (Proof of Concept) - Embedded in the processes before payment - Risk and Reward vs. T&M vs. FP

We can leverage models and experience from other projects

Page 15: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 17

Which questions may we answer?

Page 16: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 18

Thank you for your attention

Stefan Heimrich IBM Global Business Services Client Executive Federal Government, Postal and UN (+41) 79 600 24 02 [email protected]

Richard Dobis IBM Global Business Services Consultant Business Analytics and Optimization (+420) 731- 435 876 [email protected]

Please don’t hesitate to contact us for further information. We are here this afternoon at the IBM booth for a live demo of the solution

Page 17: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 19

Missbrauch und Betrugserkennung Sozialversicherung, Healthcare, Tax & Revenue

Erfahrungbericht über den Einsatz analytischer Werkzeuge am Beispiel CZ / SLO

Richard Dobis Stefan Heimrich Bern, 6 März 2013

Page 18: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 20

Fraud Reference Architecture

Decisioning Rule Execution

Simulation Predictive Clustering Scoring

Anomalies Detection

Geospatial Analysis Textual Analysis

Observation Space Internal

CRM

Policy Claims

Commissions Payments

Products

Underwriting

Billing

Reinsurance

Action Case Management

Workflow Alerts

Events

Content Analysis

Reporting & Awareness

Reporting

Scorecards

Dashboard

Users

Corp. Security

Risk Analyst

SIU

Case Manager

Claims Manager

Info

rmat

ion

Gov

erna

nce

Secu

rity

, Pri

vacy

, and

Com

plia

nce

Serv

ice

Man

agem

ent

Dat

a In

tegr

atio

n an

d D

eliv

ery

App

licat

ion

& B

usin

ess

Proc

ess

Inte

grat

ion

Foundation

Evolving

Social Media Unstructured Data

Internal and External

Streaming Data

Context Unique Entities Relationship

Awareness Unstructured

Text Extraction

External

Governmental Records

Industry Data

Managed Lists

Visualization

Identity Visualization

Geospatial Mapping

RE

SP

ON

D

AN

ALYTIC

S

SE

NS

E

Operational Process Integration

Real-time Invocation

Recommenda-tions

Source Feedback

Event Notification

FAMS

Page 19: SeGF 2013 | Missbrauchs- und Betrugserkennung im Bereich Sozialversicherung (Richard Dobis & Stefan Heimrich)

© 2013 IBM Corporation 21

FAMS Solution Overview (3 of 3) 4 basic types of support tools integrated in FAMS

1 Algorithmic Data Mining FAMS uses two statistical techniques to perform

risk scoring. The first technique is fuzzy logic. This technique identifies the similarities and

differences in a group of entities, identifying those entities that are outliers from the median.

The second technique is segmentation. Using this technique, FAMS can identify groups of entities based on their behavior.

2 Visual Analytics IBM Researchers have found that the

combination of algorithmic data mining and visualization provide the strongest type of fraud analysis.

As a result, FAMS provides for multiple types of data visualization.

Data Visualization is an interactive facility for displaying and manipulating graphical representations of behavioral data and attributes of the entities.

3 Behavior Modeling This technique will allow users to identify a

group of similar entities, then compare and contrast them to identify outliers.

Behavior modeling focuses on the behavior of an entity (based on data) instead of on entity characteristics.

In this manner, Behavior Modeling provides a highly objective method for selecting entities for review.

4 Predictive Modeling This type of analysis will allow users to predict

which entities are likely to be anomalous based on their past behavior. The Data Discovery function within FAMS uses Prediction and Segmentation advanced data mining techniques. The complementary modeling approach generates abnormality scores for entities.

The segmentation capability places the entities into clusters, based on similar performance characteristics. The application defines the features that are the basis for membership in a particular cluster.