a framework for streamlining globally integrated services
TRANSCRIPT
A Framework for Streamlining Globally Integrated Services
Biplav SrivastavaIBM Research
Figure 1. Stages of a change during a service request and the time line.
Technology (IT) Change Management(CM) [6] whichseeks to control and reduce the risk of any alterationmade to an IT infrastructure in its hardware, softwareor attached network. A rapidly increasing segment ofthe IT industry today deals with offering services tobusinesses to manage their IT infrastructure. It is common to see a large organization having its managed ITsystems, e.g., servers or data centers, co-located withits operations but being managed from a remote location. Customers often demand that change management be followed for all remotely delivered services.The services span a variety of processes like performing patches, user account management, storage andbackup, etc. Again, the service request may originatein one division, be processed in the outsourced vendoror second division and audited in a third division.
The service requests are meant to be monitored byend-to-end Service Level Agreements (SLAs) as wellas SLAs for individual participants. As the numberof participants increase in delivering a service, it isa challenge to track the end-to-end performance on aservice request and deliver the service with the estimated savings . In practice, individual participants areactively tracked while end-to-end SLA tracking is passively aimed for. Unsurprisingly, service requests areoften inefficiently delivered .
Moreover, service requests rarely occur only once.As multiple service requests arrive over time, there aremany saving opportunities due to duplicated, redundant or unnecessary activities possible from the context of the requests. Unfortunately, such opportunitiesget undetected as delivery organizations just struggle
ChangeClose Time
t tChange ServiceImplementation OrderCompleted Completed
t
ChangeCreation ~
Tjme ~ . ~
t : t : :f f f
Service Approvals Time of ApprovalOrde~ Requested All Approvals ChangeRece ived Gained Scheduled
Start
Enterprises are transforming to Globally IntegratedEnterprises wherein, among other things, the deliveryof services is globally distributed spanning divisions,suppliers and geographies. A service itself can be atthe level of business processes like processing expensesor personnel re-location (assignment) , or IT changeslike applying a patch. Given this trend, streamliningto reduce the time and cost effort needed to perform aservice is more important than ever before. The reduction can come due to better coordination among activities, elimination of redundant activities, avoidance ofsituations that lead to additional costs and finally, automation of repetitive activities wherever possible. Wepropose a unified framework for streamlining servicesand apply it to cases from business and IT services.
Abstract
Companies are increasingly transforming to Globally Integrated Enterprises[7] (GIEs) . GIEs are organizations that not only operate on a multinational scalebut also are able to incorporate into their day-to-dayoperations the advantages of their global presence likelow-cost skills and unique local capabilities. So, the organizations are able to service their customers flexiblyfrom any place in the global organization depending onthe request characteristics and thereby make savings oneach request .
A service request in a GIE can span multiple subdivisions, geographies and if parts of the businessprocess are outsourced, to external vendors. For example, consider processing expenses. The expense requestmay originate in one division, processed in the second,paid in the third division and the request archived at anexternal vendor. Or an assignment request may originate in one division, processed in the second , monitored in the third division , and use services of externalvendors likes shipping companies and tax services.
Now consider the service domain of Information
1 Introduction
978-1-4244-3541-8/09/$25.00 ©2009 IEEE III
No. Steps Expected Time Scheme 1 Scheme 2 Schemel; Scheme2 Scheme2; Schemel1 Initiate request 1 1 1 1 12 Get management request 1 1 1 1 13 Get tax briefing confirmation 1 1 1 0 04 Get support groups confirmation 1 1 1 0 15 Make pre-assignment preparation 1 1 0 0 06 Transition for assignment 1 1 1 0 07 Transfer employees physical assets 1 1 0 0 08 Do compliance check 1 1 0 0 1
Necessary time (initiation) I 8 I 8 I 5 I 2 (5 unopt) I 4 (8 unopt) I
9 Initiate end-of-assignment request I 1
I1
I1
I1
I1
I10 Complete post-assignment preparation 1 1 0 0 111 Transition for end-assignment 1 1 1 1 112 Transfer employees physical assets 1 1 0 1 1
End-to-end necessary time I 12 I 12 I 7 I 5 I 8 I
Table 1. Two schemes for assignments and analysis of their optimization potential.
to monitor single requests end-to-end. Streamliningservice requests is important, and also the ability toidentify and optimize across service requests. Eventually, streamlining delivery should lead to avoidance ofsituations that result in additional costs (e.g., missedaudits, legal fines) and a blueprint for how automationcan be introduced to expedite repetitive activities.
In this paper, we present a unified framework forstreamlining services that can be used for both business and IT services. It focuses on representing theservice ecosystem in detail, analyzing the behavior ofsingle and multiple service requests, and looking forautomation opportunities. A benefit of the frameworkis that since it treats activities at different levels ofgranularity in a set of systematic, uniform, analyticalphases, it can guide exploration of any new analysistechniques for business or IT processes and its possibleapplication to the other.
In the remainder, we first define the stages of a service request and present two examples from businessand IT services. Then we state the problem of streamlining services, describe the solution as a unified framework and illustrate it in the two cases. Finally, weconclude by putting the work in the context of relatedwork[9].
2 Background
Consider a general characterization of the stages inprocessing a service". In a typical service, see Figure 1,a service request is received by a Service Manager (SM)and assigned to a Service Implementer (SI). Approvalis a key step (or sub-process) of the service by whichall stakeholders to a service give concurrence to theprocessing of the service. A service should not be executed without all approvals in place. The SI will initiate requests for approval specifying the reason for theservice, the business details (cost, benefit), the technical details (if any) and when it is planned, and wait tillall the approvals are obtained.
IThe terms are influenced by IT Change Management butcovers business services as well.
In practice, the approvers are divided into a smallprimary group whose approval must be taken at allcosts. Then there is a larger secondary group whoseapproval can be delayed during emergency and the service's execution started, but their approval must stillbe taken. Regardless of how the approvals are obtained, all approvals have to be documented for auditreasons.
After getting approvals, the service is processed atthe scheduled time and completed. Different participants in the service ecosystem are involved in the approval and execution of the service.
3 Motivation
We now take two examples of services: assignmentfor personnel re-location which is a business processlevel service and IT change management service forapplying patches. Despite their apparent difference ingranularity, we will later look at streamlining them using a common framework.
3.1 Business Service of Personnel Relocation (Assignment)
Let us take the example of assignments. A globalcompany will have the need to move people around onassignments based on business needs. As assignmentscan involve different countries, the activities includenot only relocation but also taxation, legal and immigration concerns. Let us consider two assignmentschemes that the company provides as shown in Table 1. The first scheme provides for long-term transferand is more generous in terms of provisions than thesecond scheme which is meant for a short-term transfer. The expected time needed to complete the process,in weeks, is also showrr'. Schemel can get an employeeon-board in 8 weeks (Steps 1-8) while Scheme2 gets theemployee on-board in 5 weeks (3 steps not needed inSteps 1-8).
2The numbers are kept uniform to de-emphasize the orthogonal, but important, issue of knowing each activity's performanceestimate accurately.
112
Step No. and Name People Action Data Needed Date Generated Tool(s)j Database(s) Used Specific To?Role Type
1. PAR Notification ST C N/A Email notification Lotus Notes Operating system2. Patch Relevance CI I Server Information, Applicable Operating system
PAR No. information3. Create CMR CM S Server Information, CMR ManageNow, CPMA, Customer specific
CI PAR No. Remedy4. Check for Patch Dependence CM S or C PAR No. Pre-requisites, Patch site Operating system& Sched ule Approval CI co-requisites, (Microsoft, IBM ftp site)
post-requisites5. Download Patch and Test CI S or C PAR No. Patch download Patch site Operating system
(Microsoft, IBM ftp site)6. Upload patch package CI S PAR No. Patch package Operating system7. Build Change CI C Technical details ManageNow/ CPMA Change and Customer
of patch8. Get Customer Approval CM S or C ManageNow/ CPMA Customer9. Implement change CI S Patch package Terminal Services Customer10. Propose to Close CI S Implementation Server Hotfix ManageNow/ CPMA Customer
successful? update11. Close Change CM I Verification with ManageNow/ CPMA
customer12. Close PAR ST I CMR Patch Notification System
Table 2. A patch process at an example service delivery center.
Depending of the scheme's details, a subset of Steps2-5 can be considered as approval steps. Individualsteps can now be distributed internally across the globeor out-sourced, like Step 3. Service-level-agreements(SLAs) can be used to monitor the performance of eachgroup doing the task.
3.2 IT Change Management Service forApply Patches
Table 2 shows the near-reality patch process managed by a remote service delivery team. Such a processdescription, or process model, consists of the activitiesinvolved at each step, the roles involved with the activities, the data items needed to process a step, thedata generated and the tools used. The Patch Analysis Report or PAR is a document to issue securityadvisories that may affect systems that are owned ormanaged by the IT vendor. It is issued by the Security Team (ST) in Step 1 and sent to platform teams(Le., Windows, AIX, etc.) of different accounts to beprocessed by their System Specialists (SS). The specialists (also called Change Builder, Change Implementerin ITIL[6]), in Step 2, check the relevance of the PARto the servers that are present in their account. Theinformation needed to make this decision is the PARnotice and the server description. In the 3rd Step, aChange Management Request (CMR) is created in atool like Remedy, ManageNow or CPMA by the ServiceManagement (SM) representative (also called ChangeManager in ITIL). The change is now sent to differentgroups for internal checking and scheduling approval(Step 4). The approval consists of determining a mutually acceptable change schedule and these teams determining if they have pre-requisite, co-requisite or postrequisite patches that need to be applied along withthe proposed patch changes. In Step 5, the systemspecialist knows the final list of patches that will beapplied. She downloads the patch from the patch siteand tests it. In Step 6, the patch is uploaded to the
server(s) where the patch will be applied. In Step 7,the change is built by the platform team and in Step8, customer's approval is solicited by the SM representative. In Step 9, the change is built and in Step 10, itis proposed to be closed. Now, the SM team takes overin Step 11 and closes the change after verifying withthe customer. Finally, in Step 12, the PAR is closed.
4 Problem and Solution
We are looking at the problem of how to reduce thetime and cost effort needed to perform a service whenthe delivery organization is globally distributed. Theend-to-end (customer-facing effectiveness and deliveryfacing efficiency) SLAs for the service are given. Theaim is to organize or revise a service delivery processso that it can meet or exceed the SLAs.
4.1 Solution Approach
Figure 2 presents a unified framework for streamlining services.
In Phase1, we describe the service (business processor IT) in sufficient detail to cover activities, their interrelationships, information needed and produced, andthe actors involved. The actual representation can bein any notation or modeling tool like WebSphere Business Integrator (WBI) Modeler, drawing tools like Visio, or a simple table representation that we use.
Next, we analyze the service for characteristics thatmay hamper the execution of a service request and propose changes. Some examples are:
• The service should have a single initiation and asingle termination activity. This way, a servicerequest can be easily tracked to closure.
• Any constraint on a service request's durationshould be applied on activities from all parallelpaths in the service. As example, in the expensing
113
Phase 1: Describe process in sufficientdetail to cover activities , their interrelationships, information needed andproduced , and actors involved
Phase2 : Analyze process for characteristicshampering execution of single servicerequest , and propose changes
Phase 3: Analyze process for characteristicsoptimizing execution of multiple servicerequests, and propose changes
Phase 4: Evaluate process for automationopportunities using a methodology likeAutoSeek and make recommendations
Phase 5: FolIO'MJpon proposals andrecommendations
Figure 2. Proposed framework for streamliningservices
process, the processing of the expense and the payment for the expense to an external vendor (e.g.,business credit cards) are parallel paths. If theyare not synchronized, increased cost due to delayrelated penalties to the vendor is possible.
• What is the information needed for making thedecision at a every process phase and can it bebroken down to few, clear choices? This conveysthe complexity of each activity and the technicalskill needed for it .
In Phase3, the service should be analyzed for characteristics optimizing execution of multiple service requests, and propose changes . Some examples are:
• Identify duplicate or redundant activities, and theconditions under which they occur, if a service request is followed by a subsequent one.
• What are the undesirable situations that may happen in-between these requests?
In Phase4, we evaluate the service for automationopportunities. As noted earlier, activities in businessservices are mostly manual and need domain knowledge. In IT services, many activities can be performedin scripting software. A methodology like AutoSeek[8]can help make recommendations.
Finally, one needs to follow-up on the streamlining recommendations and measure the savings realized.Now, we apply the framework to the two examples.
4.2 Streamlining in Assignment BusinessProcess
In the assignment case, let us consider two assignment schemes that a company provides as shown in Ta-
ble 1. Following Phase1 from the framework, we haverepresented activities and their inter-relationships butwe did not model information needs and actors. ForPhase2, suppose there was no streamlining opportunity found.
Moving to Phase3, it is possible that two assignment(service) requests can come one-after-another due tobusiness needs. For example, an employee was on assignment on Schemel and upon its end, needs to bemoved to Scheme2 as the project needs a few moremonths. Or the employee was on Scheme2 when theproject scope was small but it not needs to be expandedand Schemel is applicable. The last two columns indicate the incremental activities that need to be takenafter assignment Schemel (Scheme2) is followed byScheme2 (Schemel) . We will refer to them as Scheme12and Scheme21, respectively. Note that Scheme12, tillinitiation, needs to take only 2 weeks if the redundancyof the steps is identified but otherwise will take 5 weeks.Similarly, Scheme21 needs to take 4 weeks if the redundancy of the steps is identified but otherwise will take8 weeks. For the full business process duration, thesavings are more and shown in the table.
Note that if each step is handled by individualgroups in the context of only the current service request, they will fail to recognize the redundancies andconsequently the savings possible. They are measuredon SLAs of their steps based on Service1 or Service2,and even if they meet the SLAs in this case, the company will still lose the opportunities of savings since thesteps, (like Step 3-8 for Scheme 12) were redundant.The savings here was possible to the tune of removing3 out of the 5 steps. There are other ramifications aswell:1. The company loses the ability to quickly take decisions on situations (like Scheme12 or Scheme21) asbusiness demands than competitors who are structuredin a different way.2. For unplanned events, new and difficult situationsarise for the company that could need additional expenses to resolve. Assignments are affected by taxation, legal and immigration concerns as well. In thesituation of Scheme12, the assignee would already be inthe work country. If the second assignment's start willbe delayed and the event was not planned well in advance, there will be a gap in the two assignments. Thecompany needs to be prepared to address new taxation,legal, immigration and employee's salary concerns during the gap.3. A role should have been present to monitor the performance of the assignment business process across theorganization. But the role now needs to look for specific situations like gap in assignments and have con-
114
CMR Record
r----;;;;~=_____l ~pproveffi list:Gro ups whoseapprovals have to besoli cited and thei rchange wi ndows ;Groups whoseapprovals cou ld be
~.,..---,,..., auto matica llydetermi ned
Policy-based Appro valEnabler (PA E)
In concrete terms, using AA (see Figure 3), theChange Implementer (CI) could create a CR that triesto minimize the reasons for approval delay. This is doneby pre-filling CR fields with all relevant and availableprocess details at the time approvers need them viadata integration. Moreover, if approvers can give preference, whether desirable or undesirable, on changesthat were created in the past, we could use them increating a new change . These patterns, or policies, canaddress resource conflicts, technical or business concerns about a change . Furthermore, even if the approvers are unaware of the existence of such policies,we could use their historical approval patterns to helpthem discover their policies.
Figure 3. AA architecture; data flow is marked.
AA was piloted in an IBM managed account for amonth covering the processes of software installation,patching, account creation and hardware installation,covering historically 95% of the changes in that account. We used a profiling tool to measure the timetaken for approval before the insertion of AA and whileusing it . Patching was the most frequent change during the pilot period. We found that for patching, AAcould bring 44% improvement in effectiveness (reduction of end-to-end duration) and--Sl % improvement inefficiency (reduction in effort expended) using minuteas the basis. Across all processes, >-25 % improvementin effectiveness and >-22 % improvement in efficiencywas achieved. Considering that approvers are paid atdifferent rates based on their level of expertise, the actual improvements in efficiency are even larger. Wealso found that approvers were willing to specify andsubscribe to policies. While policy elicitation in general is hard, helping approvers discover their changepreferences/ policies using learning techniques can beeffective.
duce approval delays for IT Changes. AA solutionconsists of: (a) automat ically populating the ChangeRecords (CRs) (b) partially automating the processingof low-complexity approvals tasks needing easily available data and (c) standardizing approval escalations.AA uses data integration and policy-based technologies to focus the approvers' attention on high-risk andskill-based decisions.
4.3 Streamlining in IT Changes
In the case of IT Changes, let us consider patch management as shown in Table 2. Following Phase1 fromthe framework , we have represented the activities, theirinter-relationships, the information needs and actors,but additionally also the tools used and what a patchrequest can be specific to. The more information wecan gather, the more analysis opportunities it opens.
For Phase2 of the framework, we evaluate the complexity of each activity in the service and characterizeit into three categories: as data checking step (I-type),a simple step (S-type) where output can be easily enumerated, and a complex step (C-type) where processing requires in-depth technical understanding. We alsoverify if each step is indeed separately needed due to aseparate actor, input, or processing requirement.
For Phase3 of the framework , we recognize that step4 in the process , which deals with getting approvaland scheduling a change, can be combined for multiplepatch requests. In fact, we advise the patch servicerequest to be initiated for a set of patches at regularlyscheduled time unless there is a critical patch needingimmediate implementation.
For Phase4 of the framework, we use theAutoSeek[S] method for making automation recommendations. Autoseek can help select steps which aregood automation candidates based on their importancein the process, the types of their information processing, and the effort needed to elicit, write and validatethe rules.
Building on the insights from process analysis , asPhase5 of the framework, we developed a cross-processsolution called Approval Accelerator (AA)[9] to re-
tingency plans. Most companies do not have this roleand their ability to detect events like gaps is lacking.4. The employee needs to perform redundant stepswhich can delay their business responsibilities. Moreover, in the situation of gaps in assignments, they needto plan for personal contingencies that again takes timeaway from their business responsibilities.5. Since all groups performing individual steps canclaim that they are meeting their SLAs, the role managing the end-to-end process , if it exists, and the employee, needs to repeatedly convey the sense of businessurgency. The directly affected parties are the employeeand the business project.
Moving to Phase4 of the framework, since the natureof act ivit ies in the service are mostly manual, we donot find any simple automation choices. So, we endthe analysis with the recommendation to re-organizethe process to account for Service12 and Service21.
115
5 Discussion and Related Work
Streamling globally-integrated services builds onprevious work in service optimization, automation andprocess modeling. We now discuss connections to someof the closely related ones.
LEAN[II] is emerging as a key trend in process simplification. It introduces the concept of value deliveredper step as the basis of evaluating the efficiency of eachstep in a process. The set of steps necessary to delivera service is called a value stream. LEAN focuses onreducing waste by identifying value added at each stepin the entire value stream of the process and eliminating those steps that do not add value, but resultin waste. Waste could be in any form, viz. systemdowntime, waiting time for approvals, time wasted incopy-pasting data between tools to make them worktogether, etc. The proposed framework also evaluatesthe benefits of each process activity but can work withreduced process information and extends the analysisto multiple service requests and automation.
Automation is an important aspect of process optimization and has been considered in detail by variousauthors[2, 1, 8]. While selecting the tasks/activitiesto automate, many authors tend to take the costreduction as the deciding factor. For example, in thesix-step methodology illustrated in Brown[2] , the identification of the tasks that should be considered forautomation is based on the resultant cost-reductions(Step 3). Advantages of automation, in addition to significant cost-reduction, is that it results in other benefits like increased speed of execution (decisions aremade quickly based on the business rules, rather thanwaiting for the manual operator's decision), better integration, implicit standardization (results in uniformityacross the multiple instantiations), better interactionswith other delivery processes, and better reporting andservice level management.
Rules has been a prominent AI tool in automatingthe processing of real-world applications[5, 3]. Thereis an active research community in AI investigating issues related to rules management [10] and in application areas like IT system management [4] on how to usethem for specific scenarios. The proposed frameworkattempts to find situations so that rule-based automation is cost-effectively feasible.
6 Conclusion
Managing the delivery of services in an ecosystemof globally distributed entities spanning divisions, suppliers and geographies is challenging. As enterprisesembrace the Globally Integrated Enterprises model,streamlining the service to reduce the time and costeffort needed to perform a service is more important
than ever before. We proposed a unified framework forstreamlining services and apply it to cases from business and IT services. The framework tries to identifythe savings that can come due to better coordinationamong activities, elimination of redundant activities,avoidance of situations that lead to additional costsand finally, automation of repetitive activities wherever cost-effectively possible. The characteristics of theframework is that it treats the services at different levels of granularity in a set of systematic, uniform stepslooking for improvement opportunities in single servicerequests, multiple service requests and automation. Ithas been validated on cases from business and IT-levelservices, and in the field for IT Change Management.References
[1] A. Brown and J. Hellerstein. Reducing the cost ofit operations - is automation always the answer? InProceedings of the 10th Workshop on Hot Topics inOperating System., 2005.
[2] A. Brown and A. Keller. A best practice approach forautomating it management processes. In Proceedingsof the IEEE/IFIP Network Operations and Management Symposium (NOMS 2006)., 2006.
[3] B. R. Group. Business rules manifesto:The principles of rule independence. Inhttp://www. omg.org/technology/ documents/bms.spec.caialoq. htm, 2006.
[4] J. O. Kephart. Research challenges of autonomic computing. In ICSE '05: Proc. 27th Inti. Conf. SoftwareEngg., pages 15-22, New York, NY, USA, 2005. ACMPress.
[5] T. E. Nordlander. AI surveying: Artificial intelligencein business. MS Thesis, De Montfort University. Athttp://www.4c.ucc.ie/web/ upload/publications/mastersThesis/ArtijiciaLlntelligence_in_Business.pdj, 2001.
[6] OGC. IT infrastructure library (ITIL). Inhttp://www.itil.co.uk/.• 2006.
[7] S. Palmisano. The globally integratedenterprise. In Available online at:http://www.ibm.com/ibm/governmentalprograms/samforeignaffairs.pdf., 2006.
[8] B. Srivastava. Autoseek: A method to identify candidate automation steps in IT change management.In Proc. Intelligent Management (IM-2007), Munich,Germany., 2007.
[9] B. Srivastava and S. YellamRaju. Enabling selectiveautomation of human decision-making using rules aspreferences in a service-industry application. AAAI2008 Workshop on Preferences., 2008.
[10] A. Uszok, J. M. Bradshaw, M. Johnson, R. Jeffers,A. Tate, J. Dalton, and S. Aitken. Kaos policy management for semantic web services. IEEE IntelligentSystems, 19(4):32-41, 2004.
[11] J. P. Womack. Lean thinking. In Daniel T. Jones FreePress; New Ed edition; ISBN-l0: 0743231643,2003.
116