Multi-Agent Systems for Insurance Companies Chris van Aart (chris@acklin.nl) agent based support
B.V. Founded in 2000, due to merge of two companies which were not interested in agents Use agent concepts/technology for business cases: Insurance, Energy, Travel, Telecom, Retail Applied research University of Amsterdam, National projects, Agentlink, Agentcities, IBROW project, Combined, Ontogrid
Commercial Constrains Strong Business case What is the problem (i.e. costs) Solution directions Return on investment Why use something new? Why hire small company? Proven Technology Open Source is stolen technology?
Insurance Business cases Interpolis R+V - KBC International insurance traffic Interpolis SOS International External call center Interpolis Insurance (real) agents Insurance Intermediates for claim handling
Business Case Green Card Traffic Car accidents involving international parties Settlement via European Insurance Network: Euphoria network: 17 partners 4th guideline by European Commission Settle within 3 months Penalty as high as the total amount claim
Setting communication by phone: Policynumber Licenceplate Coverage BOSO application b Policy clients Claims clients Policy companies Claims companies Interpolis, the Netherlands Back office country B
International Claim Settlement Settlement time per case: 6 months with 4 to 6 contacts for party identification Internal bureaucratic processes Information processing by humans Heterogeneous systems Costs: +/-6,000 EURO for only material damage +/-100,000 EURO for physical injury 3500 cases / year
Alternatives (1) 1. Central database in Brussels Mapping from and to own systems Transparency! 2. Web-based access to every system 1. What language? 2. 17 different procedures, login, etc 3. One single technical implementation? 4. Still man in the loop
Alternatives (2) Distributed solution Network of information exchangers: agents Work in the background Interoperability by wrapping locally Easy scaled: 2 agents -> n agents Secure: information access is locally Locally maintainable
Constraints Functional: Arms length relationship Start-up / shut-down procedures Not to much intelligence Technical: No direct access to databases No direct access the Web Organizational: Cooperate with IT departments: Private Insurance ICT Department Company Insurance ICT Department Company ICT Department Concern ICT Department
Architecture agent communication application a transducer a transducer b application b Back office Back office country A country B
Protocol snapshot
Construction of Ontology
Request Identification <fipa-message act="request" > <sender> <agent-identifier> <name id="agent@kbc.be"/> </agent-identifier> </sender> <receiver> <agent-identifier> <name id="agent@interpolis.nl"/> </agent-identifier> </receiver> <content>((action (agent-identifier :name "agent@interpolis.nl") (IdentifyParty :licenseplate "AA-10" :policynumber "7890489")))) </content> <language>fipa SL</language> <ontology>http://www.interpolis.nl/greencardontology </ontology> <protocol>fipa-request</protocol> <conversation-id>req1008770622742</conversation-id> </fipa-message>
Inform Party <fipa-message act="inform" >... <content> ((result ((action (agent-identifier :name "agent@interpolis.nl") (IdentifyParty :licenseplate "AA-10" :policynumber "7890489"))) (Party :name "Maxima" :address "Dam 1" :place "Amsterdam" )))) </content> <language>fipa SL</language> <ontology>http://www.interpolis.nl/greencardontology </ontology> <protocol>fipa-request</protocol> <conversation-id>req1008770622742</conversation-id> </fipa-message>
KIR System KBC Belgium Interpolis, the Netherlands R+V, Germany
Operationalization Transducer built in 30 days by IT department Coupling between competence model and B.O Interpolis Agent built in 60 days Models as asynchronous threads with mailbox semantics Intermodel communication via internal database R+V adjusted in 10 days Network of 3 agents running from April 2001
Interpolis SOS International Case: 8FTE Solution SOS takes over call center during non-office hours Need to know if person is insured by exchanging faxes History of communication Couple heterogeneous companies and systems using agents Let agents log interaction ROI: costs 2FTE, revenue 7.5 FTE
Interpolis Insurance (real) agents Case: 4FTE Solution Intermediates are unaware of claims of (big) clients Queries go to callcenter Copy part of database to DMZ One agent guard DMZ database One user agent with (real) agent profiles (real)agent login via extranet of insurance company ROI: costs 1.5FTE, revenue 3.5 FTE
Why own implementation Constraints/demands of customers Work with (traditional) systems and IT people Internal policies for development and deployment Not invented here syndrome Success stories? Academia and R&D departments produce tools to support articles (not industry strength products?) Understand/trust software Dedicated Why use AP for p2p communication? Which tools have rule engines? No solution is pure agent technology
Green Cat Transparent coupling between JADE and COUGAAR (MTP) Agent-Middleware independent communication Best tool for specific type of agent More information: http://www.decis.nl
Discussion Do not build General Problem Solvers Separation of concerns: Agents are software (components) A lot of infrastructure/middleware available (agent) system boundaries Competition with software vendors? Killer cocktail Methodology taking into account: Traditional software paradigms Legacy systems Alternative components Maintenance ROIs