Smart Business Process Review Sudhendu Rai 29 th

Smart Business Process Review Sudhendu Rai, 29 th May 2015

Agenda • Introduction – What is Smart Business Process? – The need for Smart Business Process – Prior experience • Program Strategy – Program strategic vision – Why PARC? – SBP research directions and examples • Pipeline Overview – – Transaction Processing Operations Optimization Business Process Representation and Optimization Web-based Optimization tool Futures • Why now? • IP strategy • Xerox, Commercial and Government opportunities PARC | 2

What is Smart Business Process? A business process that is (continually) optimized and operated using data-driven simulation optimization and real-time productivity analytics encapsulated in standalone and web-based software solutions Data collection using IT systems & customized technologies (RFID, handheld devices) Workflow models using simulations Smart Business Process Data analysis & simulation optimization framework Real-time operational techniques PARC | 3

The need for Smart Business Process Cheap data collection sensing technology offers opportunity for process data collection on multiple dimensions Service operations are constantly challenged with designing efficient processes and to operate them profitably Service processes are getting data-rich through evolution of underlying process management platforms Transaction print and mail operations was paying $3 M in financial penalties due to changes in regulatory compliance laws HSBC standardization using HSBC One – IBM Filenet, SPC controller Wireless scanner for manual process event data logging Regulatory compliance requires finegrained data collection RFID technology Struggling with low operating margins Low margins & unhappy clients in CMS/XS business TP capability standardization, margin improvement, service integration OOM – Lynnfield Prisma - Erlanger MCP 2 MCP 3 Video cameras Mobile phone apps Process modeling & optimization can enable breakthrough process productivity gains Analytical models using simplified abstractions useful to gain insights but do not scale to real world problems Data driven simulation modeling and optimization of processes can be the answer. Additional innovation needed • Process characterization and abstraction • Data collection, cleansing and fusing • Service structure optimization • Real-time management techniques PARC | 4

Lean Document Production: An instance of SBP for Print Production Services Domain Data collection tools • • • Wireless handheld event data collection tool Job/shop templates Standardized questionnaires Data collection using IT systems & customized technologies (RFID, handheld devices) Workflow models using simulations Smart Business Process SO techniques • • • Ranking & selection Greedy algorithms Simulated annealing Data analysis & simulation optimization framework Real-time operational techniques PARC | 5

Program Strategic Vision Advance the science of data-driven domain specific service process (simulation) optimization, instantiate through software tools and platforms and demonstrate significant business results through their deployment Xerox – Develop scalable process optimization tools/platforms for CPS business – Expand optimization methodology and techniques to other Xerox Services verticals and horizontal capabilities Commercial – Generate commercial partnerships for business process optimization research (e. g. AIG, HSBC, FX, …) Government – Seek funding in relevant areas of interest to government agencies. These include simulation optimization and healthcare process improvement PARC | 6

Smart Business Process: Why PARC? • The program has many challenging research problems that need to be solved • It will add a new dimension to PARC’s existing strength in analytics, model-based reasoning, big data and digital manufacturing through advancements in the science of process optimization • It has broad applicability across a wide range of business process domains inside and outside of Xerox • Three examples of the Science of Process Optimization • Data-driven simulation optimization – Optimization of non-linear stochastic processes where the objective function is evaluated using simulation – • • models driven by actual process data A very rich area of research with a regular full-track at WSC – a premier conference in the field of stochastic discrete-event simulation Process flexibility – Cellular vs departmental workflows – Flexibility allocation in a service enterprise Workflow design and scheduling in the presence of heavy-tail job size distribution

Recognizing the new trend in Operations Research: From Problem-driven OR to Data-Driven OR Problem driven OR – The starting point is a problem identified by an academic or an industry professional, and the challenge is to find answers by developing new theory or new insights about the problem at hand Data driven OR – The starting point is not a specific problem, but rather a large data set that allowed us to identify new opportunities References Simchi-Levi, D. “OM Research: From Problem-Driven to Data-Driven Research”, Manufacturing & Service Operations Management Vol. 16, No. 1, Winter 2014, pp 2 -10

Simulation Optimization – Discrete/Continuous Optimization via simulation where g(x) is the single objective represented as the expected value of a random variable where represents the randomness. denotes the d-dimensional vectors with integer components is the continuous parameter space The distribution of is unknown function of the decision variable x but can be determined via simulation models Why is this hard? • Simulation is very expensive • The randomness makes it hard to compare and evaluate outcomes during the optimization iterations and provide guarantees on optimality.

Government funded simulation optimization research • Discrete optimization via simulation • Ranking and selection • Efficient simulation budget allocation • Continuous optimization via simulation • • • Page 10 Random search methods Response surface methodology Stochastic gradient estimation Stochastic approximation Sample average approximation Stochastic constraints Variance reduction techniques Model-based stochastic search methods Markov decision processes US Government funding agencies • National Science Foundation • Air Force Office of Scientific Research • Department of Energy • NIH • Office of Naval Research Many of these have been awarded within the last 5 years

Government funded Simulation Optimization (SO) Grants Active (SO)NSF Grants ( CMMI - Div Of Civil, Mechanical, & Manufacturing Innovation ) Title Award Number Amount Award Date Optimization and Sampling in Stochastic Simulation 1453934 $500, 000 2/1/2015 New Approaches for Simulation-Based Optimal Decision Making A New Paradigm for Simulation Optimization: Marriage between Expectation-Maximization and Model-Based Optimization Methodology for Optimization via Simulation: Bayesian Methods, Frequentist Guarantees, and Applications to Cardiovascular Medicine 1434419 $220, 000 1/1/2015 1413790 $122, 444 8/14/2013 1254298 $400, 000 7/1/2013 Stochastic Simulation Optimization: An Optimized Approach 1233376 $260, 000 9/15/2012 Design Principles for Parallel Simulation Optimization 1200315 $250, 515 7/1/2012 Design Principles for Parallel Simulation Optimization A New Paradigm for Simulation Optimization: Marriage between Expectation-Maximization and Model-Based Optimization 1200162 $149, 714 7/1/2012 1130761 $197, 269 9/1/2011 Combining Gradient and Adaptive Search in Simulation Optimization 0856256 $350, 004 7/1/2009 Mathematical Optimization and Operations Research The Air Force Office of Scientific Research The Office of Naval Research (ONR) Mathematical Optimization and Operations Research program supports basic and applied research in mathematical optimization. The primary focus of the program's basic-research portion is linear, nonlinear, integer and combinatorial optimization. Theoretical development, algorithm design and analysis, computational techniques, and software prototypes for solving large-scale problems are of interest. Techniques that produce provably optimal or near-optimal solutions, as well as techniques applicable to large problem domains are favored. This includes, but is not limited to, cutting-plane and polyhedral techniques for mixed-integer programming and interior-point and first-order algorithms for conic/convex optimization. Innovative techniques for dealing with uncertainty from stochastic optimization, robust optimization and simulationbased optimization are of growing interest. The primary focus of the applied-research portion of the program is on the development of innovative models and solution techniques for Navy-relevant problems in resource allocation, such as those arising in mission planning, command control, and logistics. Current application areas of interest are maritime mission planning and unmanned sensor management and allocation. 2010 - Decision-theoretic methods in simulation optimization 2015 - Optimization of Complex Systems using Imperfect Data from Large-scale Simulations Page 11 U 01 GM 070694/GM/NIGMS NIH HHS/United States A Simulation Optimization Approach to Epidemic Forecasting

Exploring Process Flexibility for Service Process Design Seminal work on process flexibility Jordan, W. C, Graves, S. C. “Principles on the Benefits of Manufacturing Process Flexibility”. Management Science Vol. 41, No. 4, April 1995. Multi-stage supply chain Humana Lexington Mailroom Tasks Flexibility is expensive How much flexibility is enough and how to allocate it optimally is still an active area of advanced research Operators • • Graves SC, Tomlin BT (2003) Process flexibility in supply chains. Management Sci. 49(7): 907– 919. Labor $230 K/mo Queuing networks Iravani SM, Van Oyen MP, Sims KT (2005) Structural flexibility: A new perspective on the design of manufacturing and service operations. Management Sci. 51(2): 151 – 166. Call centers Wallace RB, Whitt W (2005) A staffing algorithm for call centers with skill-based routing. Manufacturing Service Oper. Management 7(4): 276– 294. Humana Offshore Network Proof of optimality of closed chains Chou MC, Chua GA, Teo C-P, Zheng H (2010 b) Design for process flexibility: Efficiency of the long chain and sparse structure. Oper. Res. 58(1): 43– 58. Simchi-Levi D. , Wei, Y. “Understanding the performance of the long and sparse designs in process flexibility”. Operations research. Vol. 60 No. 5, Sep-Oct 2012 pp 1125 -1141 Labor $70 K/mo Open Issues In Flexibility Research • For example: With asymmetric demand, closed chain may not be optimal • Legros B. , Jouini O. , Dallery Y. , “A flexible architecture for call centers with skillbased routing” International Journal of Production Economics 159 (2015), 192 -207 • “The most well-known architectures with limited flexibility such as chaining fail against such symmetry. We propose a new architecture referred to as single pooling with only two skills per agent and demonstrate its efficiency” Inflexible network Page 12 Limited Flexibility Closed Chain Network

Developing solutions to deal with high levels of task size variability in service processes • Let X be a random variable with cdf F(x) = P[X≤ x] and complementary cdf (ccdf) Fc(x) = P[X>x]. We say here that a distribution F(x) is fat-tailed if Fc (x) ~ cx-a 0<a<2 In the limit of x->∞ • • Size-based binning policies are more effective when task size distribution is heavy-tailed in distributed server processing Methods have to be adapted to take into account production characteristics such as setups, job arrival patterns, multiple job types etc. Page 13 F(x) = Pr{ X ≤ x } 1. 0 k: smallest job size p: largets job size a: Exponent in the Bounded Pareto distribution M: Mean h: Number of cutoff points pdf Bounded Pareto Distribution f(x) = (akax-a-1)/(1 -(k/p)a) x 0=k x 1 x 2 xi job size 1 On Choosing a Task Assignment policy for a distributed server system: M. Harchol-Balter, M. E. Crovella, C. D Murta, Lecture Notes in Computer Science, Springer Berlin/Heidelberg, ISBN 0302 -9743 (2004) xh=p

Pipeline Overview • • Transaction processing characterization & optimization Healthcare Call centers Transportation Supply chain IPI 0 expansion budding strategic • Enterprise analytics and Business Intelligence Data-driven OR • Data collection methods • Simulation optimization • Process management infrastructure • Automation of optimized process Business process representation Process flexibility optimization Market Development = prototype = product IPI 2 • • • Big data process simulation Cloud-based architecture Compensation design Robust optimization Robotics process automation Revenue management Supply chain optimization Novel data collection technologies IPI 1 random • • = proof of principle Gov • = concept Prototype Development Multi-site routing Scheduling tool Gov = idea Technology Development Web-based optimization tool • Automated cell design • Equipment optimization • Operator optimization • Automated simulation animation Com Concept Development Com Idea Generation Wireless data collection tools • • • LDP-Lite Advanced LDP Feedback: PARC strategy / program objectives / sponsor activity / market activity PARC | 14

Pipeline Overview • • Transaction processing characterization & optimization Healthcare Call centers Transportation Supply chain IPI 0 expansion budding strategic Data-driven OR • Data collection methods • Simulation optimization • Process management infrastructure • Automation of optimized process Business process representation Process flexibility optimization Market Development = prototype = product IPI 2 • • • Big data process simulation Cloud-based architecture Compensation design Robust optimization Robotics process automation Revenue management Supply chain optimization Novel data collection technologies IPI 1 random • • = proof of principle Multi-site routing Scheduling tool Gov • = concept Prototype Development Gov = idea Technology Development Web-based optimization tool • Automated cell design • Equipment optimization • Operator optimization • Automated simulation animation Com Concept Development Com Idea Generation Wireless data collection tools • • • LDP-Lite Advanced LDP Feedback: PARC strategy / program objectives / sponsor activity / market activity PARC | 15

TP is $60 B market opportunity

Page 18

Healthcare Transaction Processing Operations Characterization Tray: A tray of envelopes received from post office Stack: A group of sorted documents (claims, correspondences) Batch: A group of up to 25 claims. Batches are formed after documents are scanned DCN (Document Control Number): Each claim gets a unique DCN at the time of scanning Data for analysis & assessment • Reports from web-portals can provide data on claims volume, tray tracking, transmission volumes, billing, throughput by operator and by document type • Custom reports provide richer more specific data • FEPS/ABC and P&L provide financial data

Preliminary insights & research opportunities • Operations shows high variability in demand, operator processing rates for onshore and offshore tasks making it challenging for capacity planning • Task mix not known prior to document scanning making it difficult for simulation modeling of onshore process • Process flexibility in onshore and offshore applications not optimized • Additional data on forms (diagnosis codes, treatment codes, cost etc. ) could be used for value-added data analytics Page 20

Representation and Optimization of Business Processes • Current process for onboarding a new doc type in a TP environment • Typical timeframe is between 1 -3 months depending on workflow complexity • Business rules encoded in C# within MCP Project Goal: • How to create a domain specific instantiation of a TP workflow within the BA workbench to reduce time for onboarding new clients Technical Challenges: • Abstraction and representation of complex real-world TP business process and associated constraints • Automated process analytics and optimization to enable business agility Page 21

Pipeline Overview • • Transaction processing characterization & optimization Healthcare Call centers Transportation Supply chain IPI 0 expansion Big data process simulation Cloud-based architecture Compensation design Robust optimization Robotics process automation Revenue management Supply chain optimization Novel data collection technologies Business process representation strategic Data-driven OR • Data collection methods • Simulation optimization • Process management infrastructure • Automation of optimized process Multi-site routing Scheduling tool Gov budding Process flexibility optimization Market Development = prototype = product IPI 2 random • • = proof of principle IPI 1 • = concept Prototype Development Gov = idea Technology Development Web-based optimization tool • Automated cell design • Equipment optimization • Operator optimization • Automated simulation animation Com Concept Development Com Idea Generation Wireless data collection tools • • • LDP-Lite Advanced LDP Feedback: PARC strategy / program objectives / sponsor activity / market activity PARC | 22

Centralized Print Services (CPS): Web-based optimization tools Business Challenges Functional Overview • CPS is a highly fragmented $930 M/y business with ~1800 geographically non-collocated print shops embedded in many industry verticals with little standardization in process and infrastructure • Enable global deployment of equipment, labor, workflow and configuration optimization services within CPS leveraging a web-based infrastructure with a goal to support new account signings, same account revenue growth and renewal initiatives Web-based Optimization Tool • • A scalable architecture with standardized data gathering templates and automated reporting Automated simulation optimization capabilities Equipment selection Operator optimization Cell design optimization Labor flexibility optimization Automate process flow visualization Automated workflow animation Page 23 Technical Overview

SO in Print Process Optimization Equipment optimization Operator optimization Page 24 Rai, Gross, Ettam, “Simulation-based optimization using greedy techniques and simulated annealing for optimal equipment selection within print production environments”

Why now? In the context of existing projects Project Category Areas of innovation Needed Resources Value Chain Needed Value Chain In-Place Current source of funding Future source of funding WBOT P Cell design optimization, labor flexibility optimization, Automate process flow visualization 2 HCL for 2 H (subject to BG input in early June). XRCI resources Technology team to productize & rollout Delivery consultants to utilize the tools Xerox strategy contracting, Commercial Multi-site I TAT prediction models, Routing methods XRCI resources Technology team to productize & rollout Multi-site routing platform under development within CPS Xerox strategy contracting, Commercial TP Service Operations Optimization I Domain knowledge, data collection, modeling & optimization methods PARC Too early to engage technology team Support from TP business to help with innovation activities Xerox strategy contracting, Commercial Business Process Representation & Optimization E TP Process abstraction and integration into BI workbench PARC Xerox strategy contracting

IP Strategy Areas of intellectual property – Data collection tools and methods – Business process modeling and representation – Operations research methods • • • Simulation optimization techniques Data analytics Real-time process management Process flexibility Robust optimization, revenue management and pricing – Multi-domain process integration (e. g. call centers/print shops/mailrooms) – Hardware automation PARC | 26

Smart Business Process: Xerox Opportunities • Centralized Print Services • Scalable platform for enterprise optimization and big data analytics and visualization • Multi-site routing and production • Transaction processing capability • Single site optimization • Multi-site consolidation • Process automation • Other areas: – – Call center Transportation Supply chain …

Smart Business Process: Commercial Opportunities • LDP consulting projects generated $3. 9 M in professional services revenue for Xerox ($2. 5 M for XIG) • Prior client engagements show commercial need of this initiative in other process domains – Credit card production (FDR) - > Similarities with manufacturing process optimization – Legal paper processing (HSBC) -> Banking transaction processing – KYC (Know your customer) (HSBC) -> Client onboarding anti-money laundering – Data center (HSBC) -> Was relevant to Xerox ITO (now ATOS) – Schneider Electric -> Discussions with ITO and Schneider. ITO/XS wrote a white paper with us on “How Smart Manufacturing Delivers Performance Improvement” – Insurance processes (AIG) -> Sudhendu to give a presentation in early June at AIG

Commercial Opportunities Model: Innovate/deliver significant Print Production Services Optimization Solutions serving a $100 B/y US Printing market value for Xerox and leverage it with outside clients WBST (Enterprise) TBD WBOT (Enterprise) 20+ CPS accounts Revenue from commercial projects Kaiser $3. 9 M Permanente Advanced LDP (Large Ops) FDR HSBC HK Fidelity BNYM (XS) HSBC US LDP-Lite (Small Ops) Full-LDP (Mid-sized Ops) Staples 200+ Small Xerox Managed Service Sites 100+ Mid-sized Xerox Managed Service Sites ($200 profit impact across Xerox value chain) 2003 2004 2005 2006 2007 2008 Continued usage in CPS 2009 2010 2011 2012 2013 2014 2015 PARC | 29

Government Opportunities: Simulation Optimization (SO) Active (SO)NSF Grants ( CMMI - Div Of Civil, Mechanical, & Manufacturing Innovation ) Title Award Number Amount Award Date Optimization and Sampling in Stochastic Simulation 1453934 $500, 000 2/1/2015 New Approaches for Simulation-Based Optimal Decision Making A New Paradigm for Simulation Optimization: Marriage between Expectation-Maximization and Model-Based Optimization Methodology for Optimization via Simulation: Bayesian Methods, Frequentist Guarantees, and Applications to Cardiovascular Medicine 1434419 $220, 000 1/1/2015 1413790 $122, 444 8/14/2013 1254298 $400, 000 7/1/2013 Stochastic Simulation Optimization: An Optimized Approach 1233376 $260, 000 9/15/2012 Design Principles for Parallel Simulation Optimization 1200315 $250, 515 7/1/2012 Design Principles for Parallel Simulation Optimization A New Paradigm for Simulation Optimization: Marriage between Expectation-Maximization and Model-Based Optimization 1200162 $149, 714 7/1/2012 1130761 $197, 269 9/1/2011 Combining Gradient and Adaptive Search in Simulation Optimization 0856256 $350, 004 7/1/2009 Mathematical Optimization and Operations Research The Air Force Office of Scientific Research The Office of Naval Research (ONR) Mathematical Optimization and Operations Research program supports basic and applied research in mathematical optimization. The primary focus of the program's basic-research portion is linear, nonlinear, integer and combinatorial optimization. Theoretical development, algorithm design and analysis, computational techniques, and software prototypes for solving large-scale problems are of interest. Techniques that produce provably optimal or near-optimal solutions, as well as techniques applicable to large problem domains are favored. This includes, but is not limited to, cutting-plane and polyhedral techniques for mixed-integer programming and interior-point and first-order algorithms for conic/convex optimization. Innovative techniques for dealing with uncertainty from stochastic optimization, robust optimization and simulationbased optimization are of growing interest. The primary focus of the applied-research portion of the program is on the development of innovative models and solution techniques for Navy-relevant problems in resource allocation, such as those arising in mission planning, command control, and logistics. Current application areas of interest are maritime mission planning and unmanned sensor management and allocation. 2010 - Decision-theoretic methods in simulation optimization 2015 - Optimization of Complex Systems using Imperfect Data from Large-scale Simulations Page 30 U 01 GM 070694/GM/NIGMS NIH HHS/United States A Simulation Optimization Approach to Epidemic Forecasting