From: AAAI Technical Report SS-94-07. Compilation copyright © 1994, AAAI (www.aaai.org). All rights reserved. Modeling Organizations: Methods Used by Real-world Designers Submission to the AAAI-94Spring Symposiumon ComputationOrganization Design Bryan Borys ComputationalOrganizational Design Lab Institute for Safety and SystemManagement University of SouthemCalifornia Los Angeles, CA90089-0021 213/740-4045 borys@usc.edu Formal,abstract computationaltechniques for modelingorganizations imply certain approachesto certain modelingproblems. Less formal techniques used by people involved in concrete design contexts in real organizations(real-world designers)face the sameproblems,but use different approaches.Foursuch problemsare considered: the problemsof simplification (whichfeatures to model, whichto ignore), granularity (what unit of analysis to adopt), simulation(howto verify accuracyand consistency of the model), and agency(howto be certain that the agents modeledwill respondto certain signals in certain ways). Approachesof computationalorganization design (COD)and real-world design (RWD)to each of these problems are comparedand somelessons both CODand RWDare drawn. Introduction hope that further refinementswill yield more faithful models. Analogousproblems face designers of organizations in "real-world" settings) Practitioners of real-world design (RWD)must also transform their chaotic, ambiguousand confusing observations of organizational 2behavior into a format amenableto design. Central problemsof computationorganization design (COD)result from the gap between, one hand, "natural" representations of organizations that are meaningfulto organizational members,and on the other hand the formal representations required by available computational techniques. CODproceeds by makingsimplifying assumptions about organizations which makecomputationeasier. COD modelerssacrifice faithfulness to reality in the nameof modelingexpediency, often with the 1 I use quotes aroundthe term "real-world" (in this instance only) in recognition of the fact that the distinction betweenreal-world design and COD is one of degree: "real-world" designers use formal techniquesjust as do those involved in creating or using COD systems. 2 For detailed discussions of this problem,see Bryan Borys, Where Do Rules ComeFrom?: ParticipantObservationof the Process of Administrative Innovation, unpublisheddoctoral dissertation, GraduateSchoolof 11 p. 2 Nonetheless, the activities of real-world by its explicit requirements -- they are thus more designers are more deeply embeddedin a vulnerable to bias than the CODpractitioner particular organizational context than that of COD pursuing abstract models. practitioners. The task of CODis to generate an For re’d-world designers, bridging the gap abstract representation of organizational behavior between organizations amenable to simulation, generation of design organizations involves applying two things: alternatives, First, their native understandings of the etc. The task of RWD is to generate and models of a concrete, workable design that will address organization they are attempting to (re)design. specific problemsin a specific organization. Second, the everyday methods they use to In pursing RWD,designers have access to, managethe rich complexity of everyday life. The and are influenced by, a rich set of organization- approaches they use for bridging the gap and specific knowledge(e.g., culture, standard managing the problems of organizational practice, knowledgeof individual modeling provide a fruitful idiosyncrasies). formal and abstract approaches developed in the Whatmight be considered extraneous issues from a CODperspective are counterpoint to more CODliterature. kept at the forefront of the RWD task. For Real-world instance, some real-world designers tend to Central ignore certain organizational features that might problems design: and approaches embarrass or offend) Thus real-world design is more deeply embedded in the rich, complex and Whatfollows is a set of observations of a often unprogrammableworld that sociologists group of real-world organization designers. with a penchant for neologism call "everyday These designers were members of a formal life." process improvementteam at a large electronics manufacturer I shall call "Electra. ’’4 The Electra This embeddedness has two implications that differentiate RWDfrom COD:First, design team was charged with using basic since RWD is embeddedin an organizational context, real- "business process reengineering" methods5 to world designers have access to knowledge that improve the speed of Electra’s engineering helps them get around tricky modeling change process. The most difficult problems-- they thus have a leg up on the more task was to map the existing engineering change abstract and formalistic process; this step took up about two-thirds of the CODdesigners. Second, part of this design team’s time. however, this same access sometimes leads realworld designers to generate models that are more I shall highlight some of the major problems influenced by the context of the design task than that faced these real-world designers and identify Business,StanfordUniversity,1992;and"’Seeingthe Elephant’: A GroundedTheoryof the Role of Knowledge Interests in BusinessProcessReengineering,"unpublished manuscript, submittedto the 1994Academy of Management meetings. 3 See Borys,1992,op cit. 4 Moredetaileddescriptionsandanalysesof the Electra case can be foundin Borys,1992,op cit. 5 See, for instance, Hammer, Michaeland Champy, James. 1993. Reengineeringthe Corporation:A Manifestofor Business Revolution, NewYork: HarperBusiness. 12 p. 3 howthe designers approachedthe problems. problemsor problemswhosesolutions lie 6beyondthe capacities of the designers. Thusthe implementationcontext (rather than merely the available evidenceand modelingtools) influences howthe modelis simplified. One result is that real-worlddesigns tend to more readily implementablethan CODdesigns, becauseimplementationissues are (often surreptitiously) importedinto the modeling process. Theother side of this coin, however,is that RWD maymerely replicate existing organizational problems, where moreformal and abstract CODtechniques force designers to address them. This suggests that design-forimplementability(DFI) is an important consideration: CODdesigns, generated by a COD expert far from actual organizations, mayignore crucial factors that impedethe utility of the resulting design. It also highlights, however,the danger that DFImayserve merely to legitimize the glossing of importantorganizational problems. Fromthis discussion I shall drawsomelessons, or at least somecautions, for those workingin COD. 1) Theproblemof simplification. All models simplify. CODmodels often simplify on the basis of computationalease, makingassumptionsthat are necessary to allow the use of available methodsand representations. Real-worlddesigners simplify, as well. The representationaltools available to real-world designerscan be quite rich: flowcharts, diagrams, organization charts and prose descriptions of activities can conveya richness of meaningand a level of complexitythat is infeasible for moreformalrepresentations. Nonetheless, these tools imposelimits on complexity, as well. Thusreal-world designers ’also face the problemof whenand howto simplify, to gloss, to ignore. Thequalitative difference betweenreal-world design and COD is that real-world designers often simplify on the basis of "practical" demands-- their simplifications facilitate successful action in a specific organizational context, not merely successful computationor accurate representation. Real-worlddesigns are often evaluatedin termsof their implicationsfor action -- evento the point where implementabilityis a greater concernthan accuracyin representation. For instance, designers tend to modelfeatures that create important and fixable problemsand ignore modelablefeatures that generate unimportant 2) The problemof granularity. In modelingbehavioral processes, an importantissue is to choosethe appropriateunit of analysis -- that is, should wemodel behaviors, tasks, inputs/outputs, "steps" in a process, etc.? CODmodelstypically adopt some stylized unit (e.g., "transaction")that is hardcodedinto the formal representational scheme. CODdesigns thus tend to consistently implement a particular choice of granularity throughoutthe model. 6 See Borys, 1992, op cit. 13 p. 4 an outsider could not comprehend it in greater detail. Such"black boxing"of certain areas protects them from outsiders’ meddling. It also, however,influences the granularity of real-world designers’ models. Thusthe choice of howfine-grained a modelshould be is influenced, for real-world designers, as muchby language, organizational structure, and organizationalpolitics as by the formal requirements of the modelingenvironment. Again we find a two-sided result: RWD sacrifices consistent implementationand is moreopen to bias than moreformal methods.But on the other hand it moreeasily captures modelsof task structures that are moreinherently meaningful than some CODontologies. For real-world designers, determiningthe appropriategranularity is a constant problem. Whatis appropriate is determinedin part by more or less "natural" chunksof activities. Thelevel of granularity maydiffer across parts of the organization that is being modeled.For instance, activities within a Financefunction are often governedby quite explicit rules governingquite "small" tasks. Theseprovide a languagefor renderingFinance-relatedactivities in a finegrained way;this languageis adoptableby realworld designers. The language of Design Engineering,by contrast, is full of "black boxes," in whichmanyactivities are implied, but not analyzed(e.g., "generate design alternatives"). Thusreadily available modelsof DesignEngineeringtend to be coarser-grained than those of Finance. A task in Finance may take one personseveral minutes; yet a task in Design Engineeringmaytake several people several months. Thelanguage of tasks in Finance 3) The problemof simulation. Amajorbenefit of COD is that it facilitates simulationof various organizational features. Such simulation allows the CODmodeler to generatea result, compareit to empiricalresults, and thus verify the accuracy and consistency of and in DesignEngineeringfacilitate modelingat quite different levels of detail. Organizationalpolitics seemto play a role in such distinctions. Givenreadily available models of DesignEngineering,it is nonetheless possible to delve into the design task in moredetail (e.g., by asking "Howdo you generate design alternatives?"). Designengineers are often resistant to such probing, however.Their defenseis that their workis not amenableto such fine-grainedanalysis. As I tried to uncoverthe fmedetail of a particular design engineering decision, for instance, twoengineersin separate conversations gave the sameresponse: They couldnot explain the steps involvedin that particular process; all they couldsay wasthat "it’s an engineeringdecision." Theyimpliedthat the model. For real-world designers, simulation and consistency-checkingis both morerich and less systematic and accurate. Since real-world designs are typically lists of features or steps in a paper flow chart, real-world designers must supply their ownmediumof simulation and test. The most natural and perhaps the most pervasive is the narrative: real-world designers "talk through" the implications of their models,checking whether the models "makesense." Theproblemfor real-world design is that forcing modelsto "makesense" in the context adopted by the real-world design group may introduce spurious elements. "Narrative 14 p. 5 simulation and checkingincreases the intemal simulation’’ forces the elementsof the modelinto a specific context in whichthey maynot make sense. Re’d-worlddesigners thus maysacrifice accuracy for meaningfulness. Oneby-productof narrative simulation is "filling in the gaps": a frequentsourceof spurious information. Narratives encouragethe narrator to introduce his or her ownknowledge in the place of informationthat missingfromthe story. For instance, Elecu’adesignersoften filled in the wholesin informants’accounts of the engineering changeprocess. Oncethey had mappedseveral pieces of the process, they talked through an exampleengineering change. At several points they found gaps in the model.The typically response wasto say "what they must meanis..." and to fill that gap with their own (often wrong) knowledge. Anotherby-productis that elementsthat appear logically inconsistent are often removedor glossed, rather than acceptedas evidenceof the designers’ lack of understanding.7 Narratives makesense only within a particular context. For instance, simulating the engineering change process modelby imaginingthe history of an engineering changerequiring a changeto a printed circuit board layout mightgenerate different issues than a narrative history of a consistency from a very formal standpoint. For real-world design, the process often cannot providethe samelevel of rigor. Instead, however,real-world design simulation increases the meaningfulnessand credibility of the model. 4) The problemof action. A popular COD conventionis to treat actors as "agents" that respondto certain (predefined) communicationswith certain (predefined) routines. Asignal to an agent triggers a specific response(or set of responses;or triggers it with certain probability). In contrast to this sanguineassumption, RWD takes the "agency problem"as a central issue. Thetacit assumptionunderlying the design of mostorganizationalstructures is that clear communications,coupled with clear and certain rewards and punishments,are sufficient to ensure appropriate behaviors. To assume otherwise is to becomequite pessimistic about the possibility of coordinatingthe behaviorof large numbersof people without tremendouscost (e.g., in face-to-face meetings). Nonetheless, everydayoccurrencescontinually challenge this assumption. People issue orders and requests that generate unexpectedresponses becausethe personreceivingthe signal interprets it differently than the sender intended. Such communications are moreor less deeply context-dependent;thus real-worlddesigners try to incorporatetheir native understandingsof signalling contexts into their designs. For real-world designers, the problemis: Howcan they be confident that a given signal will result in certain actions? Real-world designers havea rich set of schemasof agent fictitious changeinvolvinga newelectronic component. Theresult is that simulationin real-world design, insofar as it relies on placinga particular modelwithin a specific context, limits the robustnessof the simulation. Theflip side of this is that, again, the end result is a modelthat has greater a priori meaningfulness. For COD, 7 ibid. 15 p. 6 literature: for instance, that DesignEngineersare often not receptive to signals fromManufacturing about the "manufacturability"of their designs.) Thereceiver’s resources, skills, etc. A seconddimensionis the receiver’s capacity for respondingto the trigger. For those likely to act, but in unpredictable ways, signals must include directions for action. TheElectra design team distinguished between"process" problems, responsivenessand signalling contexts. These schemashelp them analyze typical contexts and decide on whichtypes of signals are better or worsetriggers of which types of behavior. These schemasaddress: first, the motivationsof particular receivers; second,the resources of the receiver; and third, the nature of the signal. The schemashelp real-world designers approachthe agencyproblemwith a rich repertoire of solutions. Thereceiver’s motivation, Oneaspect of the which stemmedfrom lack of triggers, and "training" problems,causedby a lack of understandingof howto respond to a trigger. A Draftsperson, for instance, maybe able to respondadequatelyto a request for a revised set of paperdrawings,but maynot be able to fulfill a request for the sameset of drawingsto be set signalling context is a presumptionof the receiver’s willingness to conformto the sender’s intent. Someactors are knownfor their "initiative" and thus can be expectedto take action given only information, rather than specific directions. At the other end of the initiative spectrumare agents whohavea over electronic mail. Thenature of the signal. In addition to the propensityand ability of the receiver to conform to the signal, anotheraspect of the signaling context is the nature of the signal. Oneway Electra designers conceivedof the force of the signal wasin terms of the "hardness"of the information conveyed. "Hard information" was unambiguousand final -- and which, for that reason, waspresumablyan effective stimulus of a particular behavior. Soft information,on the other hand, was knownby both parties to be ambiguousand subject to change and thus more likely to allow for multipleinterpretations and multiple responses. propensity to shirk from certain tasks. For such agents, simply conveyinginformation is not enough;for those whoare likely to not act, formal orders must be providedin addition to information. Anticipatingthe interpretation of a particular signal-to-act thus involvesinferring the likely state of the receiving agent. Thereare no general rules for addressingthis issue; real-world designers’ backgroundknowledgeabout the organization provides the basis for such inferences. For instance, according to a member of the Electra organizationdesign team, "Drafting won’t do anything without a WorkOrder." Such a statement, necessaryto specify the proper signal, is specific to Electra. Thereis no general statement about the relationship betweenDesign Engineering demands,Drafting responses and workorders outside Electra. (Althoughslightly moregeneral hypothesesmaybe available in the Theproblemis that hardness is contextual. At another organization, for instance, design engineers gave preliminarycost estimates to sales personnelso the latter couldplan their sales strategy. Oncethey had the information, however, somesales managersbeganusing it to quote prices for customers,thus locking the 16 companyinto those estimates. Whatseemedlike "soft," "guesswork"cost information to the engineer wasinterpreted as a formally binding cost estimate by a sales manager. The problemis that one cannot always foresee the context in whichinformationwill be interpreted. Thus real-world designers sometimes create an artificial contextby formallylabeling information. Electra organization designers used formal proceduresto bind a particular predefined interpretation to a particular predefinedcontext. For instance, cost information was formally and explicitly labeled "hard" and "soft" accordingto the particular task whichgeneratedit. Further, real-world designers mayattempt to manipulate the hardnessof the informationis specific contexts. "I’d like to firm up that info," said Jim, a memberof the Electra team, as he proposeda rule that requireda particular report to be interpreted as a final cost estimate, rather than "soft, waffieystuff." Suchrules not only tells the receiver howto interpret the cost data. Theyalso encouragesthe sender to be morecareful about howtheir signal mightbe received. Often, however,real-world designers cannot completelyand confidently specify the nature of a signal. Theythus resort to markersfor such ambiguity.For instance, "Designtells Drafting to makechangesto drawings," reads the Electra design; the designers could give no morespecific advice about howto tell Draftingto do so. Signals are invariablycontext-specific. In the CODparadigm, receivers’ motivations and capacities are modeledand certain specific signals are assumedto havecertain efficacy. Thusthe CODparadigmassumesa particular context. For real-world designers, such issues are subject to p. 7 continual revision, based on organization-specific knowledge.Real-world designers recognize that contexts shift. Theyrespondto such uncertainty by bindingspecifiable contexts to particular signals; but such a strategy can only be partially effective. Conclusions Real-worlddesigners do not necessarily use better strategies for solving problemsof organizationalrepresentation. Neither are all of their strategies available for researchers working in COD.But by describing, typifying and understandingreal-world design in natural settings, wemight find someuseful extensions to COD.The basic contrast between RWDand CODapproaches, as suggested by Table 1 below, is that RWD trades rigor for riclmess and meaningfulness. Both CODand RWDprovide screens for deciding what features to modeland whichto ignore. COD often proceeds by ignoring features that cannot be modeledwith available modeling techniques. RWD often proceeds by ignoring features that raise political problems,that violate organizationalnorms,or that raise issues that are not solvable by designers. Both CODand RWD must choose a level of analysis, a level of granularity. For COD,this choice is typically embodiedin an ontologythat is then consistently implementedthroughoutthe model. For RWD, granularity is a function of the natural-languageontologies of organizational members.It thus is moreinherently meaningful, but also shifts fromarea to area. Both CODand RWD have provisions for simulating and testing their models. COD 17 p. 8 simulation techniquesare quite rigorous comparedto RWD techniques, which often rely on the vagaries of narrative simulation. Nonetheless,nan’ative simulationis a better guarantor that the model"makessense" in a specific organizationalcontext. Both CODand RWDmust address the agencyproblem, or the question of howthe agents they modelwill respondto the signals they are provided. Thesignal-response set is dependentuponthe context in whichthe signal is sent and received. CODdesigners assumea particular type of signal context. RWD designers, by contrast, haverich schemasof signal contexts they employto analyzevarious contexts in detail; they attempt to use these schemasto bind certain contextualfeatures to the signal, with moreor less success. As are most formal techniques, CODis susceptibleto the criticism that gains to rigor comeat the cost of a lost of meaningfulness.The contrast with RWD approaches amplifies such a criticism. It also, however,highlights someof the costs to meaningfulness:inconsistency, bias, and reification of organizationalculture. Understandingthe sources of these costs may help us to makeCODmore meaningful without suffering the samepitfalls. 18 p. 9 Table 1: Comparisons between COD and organizational Problem Simplification RWD approaches to problems of modeling Question CODapproach Whatto include? four RWDapproach Excludes what cannot be Excludesissues that raise modeledby existing political problems. technique. Granularity Simulation Agency What’sthe level/unit of Consistent throughout the Uses natural taxonomies of analysis? model. activities. Howcan we check the Context-independent, Context-sensitive testing of accuracy and consistency of rigorous testing of many features amenableto our model? aspects simultaneous. narration. Howcan be certain that the Assumesa particular Uses rich knowledgeof agents in our modelbehavior signalling context. aspects of signalling context as we assumethey will? to bindparticular features to particular signals. 19