Design of Risk Management Strategies in Business Process ...

Design of Risk Management Strategies in Business Process Information FlowXue BaiOperations and Information ManagementSchool of BusinessUniversity of ConnecticutRisk Workshop SAMSI2Outline•Motivation and problem definition•Methodology•Experimental study•Real world application•Future researchRisk Workshop SAMSI3Motivation•Impact of errors in corporate business processes –“10percent to 30percent of the data flowing through corporate systems is bad…” (CFOmagazine 2003)•Impact of errors in healthcare processes–More than 8.8 millionADE’s occur each year in ambulatory care, cost at least $5,000per ADE. Medication errors account for 1 out of 131ambulatory care deaths (Washington: eHealth Initiative2004).–Health care data quality: accuracy 67%,completeness: 30.7%(Stein et al. 2000) •Legal mandates–Sarbanes Oxley Act (2002)–HIPAA(1996), Medical malpractice lawsRisk Workshop SAMSI4PhysicianPatientPharmacistInsurerPrescriptionPrescriptionBillMedicationBillAn Example of BP with Errors and RisksCall back and complainCall back and complainCall back and complainFormulary mismatchWrong dosageAdverse Drug EventCall back efforts; Administrative cost; Patient satisfaction & loyalty & litigation issues. Patient ends up in ER. An example: a medication processBills for ER visit. Manual checkE-prescribing systemsPerformance reviewE-order systemsRisk Workshop SAMSI5PhysicianPatientPharmacyInsurerPrescriptionPrescriptionBillMedicationBillAn Example of BP with Errors and RisksA medication processRisk Workshop SAMSI6•A Business Process (BP)–Tasks–Information flow–Errors•Accuracy•Completeness•Occurrence–Risk exposure–Design of Control structure for risk managemententer order info.check alert info.update medication info.orderdatabaseinaccurate dosagemissing information The order management process at the pharmacyManual checkE-order mgt.check alertinfo.update medication info.databaseenter order info.Elements of the ModelRisk Workshop SAMSI7Outline•Motivation and problem definition•Methodology•Experimental study•Real world application•Future researchRisk Workshop SAMSI8Check alert info.Update medication informationEnter order informationDatabaseinfo.Update informationinformationwrong dosagewrong dosageA Simple Sequential ProcessProcess Structure Affects Error ImpactRisk Workshop SAMSI9A Simple Parallel ProcessOrder of medicationShipping invoicePayment voucherPreparing voucher packagecontrolcontrolProcess Structure Affects Control FunctionRisk Workshop SAMSI10BP as a Grapht1t2 t3 Theprecedencematrix:otherwise.,0 , precedesdirectly task if ,1][jitTijij00010011033TRisk Workshop SAMSI11t1t2 t3 Thevolume transitionmatrix:p(T): the ratio between the volume output by task iand the volume fed to task j, giventij=1.000100110)(33Tp0001000)(131233rrTpt1:Non-Condition node:t1:Condition node:BP as a GraphRisk Workshop SAMSI12•The propagation impact (PI)matrix:–K: the length of the longest path in a process. –p(T): the volume transitionmatrix kKkTp1)(jiji][000100210000100110T000100110)(Tpt1:Non-condition node:00010002321000100110T00010010)(21Tpt1:Condition node:Impact of ErrorRisk Workshop SAMSI13Error Generation•Error correlation structure•Models for error generation processes –hierarchical sampling schema•Controlling for dependence/independence due to the homogeneity/heterogeneity of operations and resources involved–Within a task–Across tasks,~,|MiormaldLogisticNGeneralizep)( ~|imimimpBernoullipeMm,,1Ni,,1Risk Workshop SAMSI14•The number of errors of type mat task i:–: number of errors of type mthat show upat task i–: number of errors of type mthat arriveat task i–: occurrence of errors of type mgeneratedby task i–: average number of eimimNjjmjiimimimeeeee~~~1imeimeimeime~Error PropagationRisk Workshop SAMSI15Loss and Risk Measurement•Loss:cim: cost of an error of type mat task i.•Risk Measures–Expected Loss (EL), Value-at-Risk (VaR), Conditional Value-at-Risk (CVaR)imimNjjmjiimceel~~1ELCVaR VaRlossβRisk Workshop SAMSI16Risk Measures–Expected Loss–Value-at-Risk–Conditional Value-at-Risk),(:min )(llxlxrdlxlfxxr)(),(-11 )(dlxlfx0),()(EELCVaR VaRlossβ)(xr)(x)(ExRisk Workshop SAMSI17•Control allocationfactor•Effectivenessof control–the probability of a control catching an error:•Deterministic control•Stochastic control•Costof control (per period)Risk Management: Control ModelNixxNiii,,1,1,101.0,1,))(()(11iibiiNjjiiidbxdtpxi.10,10 ,)(iiaiiigaxgxi.10,10)Bernoulli(~)(iiaiiigaxgxiRisk Workshop SAMSI18Model Formulation I, II, and III–Design problem: Given a budget B, –Model I: “Expected-Loss-Optimal” Control Structure–Model II: “β-VaR-Optimal” Control Structure–Model III: “β-CVaR-Optimal” Control Structure.,...,1,10,1,)(.. )(min111NixxBxtsxiNiiNiiiNiii)(xL,),(:min )(llxlxr)()(xrlxRisk Workshop SAMSI19Outline•Motivation and problem definition•Methodology•Experimental study•Real world application•Future researchRisk Workshop SAMSI20Experimental Study •Experimental design–Topological variation•Sequential, parallel, arbitrary–Process size•Small (4 tasks), medium (10 tasks), large (25 tasks)–Cost of control vs. Loss per error ( )•Expensive: (500, 1000, 2000, 4000, 10000), inexpensive: (25, 50, 100)–Tolerance level of risks (β)•β= 0.90, 0.95, 0.99–Error correlation•Independent, dependentiicd/Risk Workshop SAMSI21Experimental Results•As the process sizeincreases, (Table 113, 115, 117, 119, 121, and 123)–The optimal amount of control allocation in total increases–The optimal amount of control allocation at each task decreases–For sequential structure, the objective function value increases exponentially; for parallel structure, the magnitude remains the same. •As the tolerancelevelof risks (β) changes (Table 2 and 10; Table 22 and 30), –The optimal amount of control allocation in total increases–The impact of βat the task leveldepends on characteristics of the loss distributions•In the range of βvalue and loss distributions tested, the impact is insignificant.•As the ratio cost of control /loss per error( ) increases (Table 153-159)–The optimal amount of control allocation at each task decreases–The optimal amount of control allocation in total decreases i i c d /Risk Workshop SAMSI22Experimental Results (continue)•Optimal control allocations depend on risk objectives.–The relative importance of each task location changes accordingly–Tradeoffs when consider multi-risk objectives•For processes with sequential structure, holding other factors constant, –The highest control allocations occur at tasks towards the center of the process.•For processes with parallel structure, holding other factors constant, –The highest control allocations occur at the merging tasks of the process.Risk Workshop SAMSI23Outline•Motivation and problem definition•Methodology•Experimental study•Real world application•Future researchRisk Workshop SAMSI24•An Order Fulfillment ProcessThe Data:15tasks, 13internal tasks, 46errors that occur in different tasks, costs per error per type, frequencies of error occurrences, cost factors of controls, based 1200orders per month.Case StudyRisk Workshop SAMSI25The tasks:0) Clients place order, 1) Enter order information, 2)Check payer and insurance info. 3) Create/update contracts, 4) Prove prescription, 5)Prepare prescribed items, 6) Dispense from alternative source, 7) Submit drug orders to wholesaler, 8) Deliver medication, 9)Prepare and send claims to an insurance companyor 10) to the responsible party, 11) Collect payments, 12) Post payments and prepare vouchers, 13) Update ledgers, 14)insurer/clients pay bills.Results: Optimal Allocation of Control ResourceRisk Workshop SAMSI26The tasks:0) Clients place order, 1) Enter order information, 2) Check payer and insurance info. 3) Create/update contracts, 4) Prove prescription, 5) Prepare prescribed items, 6) Dispense from alternative source, 7) Submit drug orders to wholesaler, 8) Deliver medication, 9) Prepare and send claims to an insurance company or 10) to the responsible party, 11) Collect payments, 12) Post payments and prepare vouchers, 13) Update ledgers, 14) insurer/clients pay bills.Results: Optimal Objective Function ValuesRisk Workshop SAMSI27Outline•Motivation and problem definition•Methodology•Experimental study•Real world application•Future researchRisk Workshop SAMSI28•Summary–Risk management models for error associated risks in business process information flow•Future research–Sensitivity analysis of the effect of other factors on optimal control allocations and risk objectives•Loss per error, Control effectiveness, Cost structure of controls, Topological redesign, Analytic solution for CVaR –Managerial problems•Multi-Objective Optimization•Find the maximum confidence level βfor a given value-at-risk •Given the output errors, identify the most probable error sources •Many others.Future research