L8 - framework for organlzlng, deflnlng, and sLandardlzlng buslness processes Lo effecLlvely plan and conLrol and org. so LhaL Lhe org. can use lnLernal knowledge for exLernal advanLage L8 developmenL (beyond M8 = manufacLurlng resource plannlng) drlven by need for sLronger lnLegraLlon beLween funcLlonal enLerprlse sllos (1970s-1980s) 1960s: early compuLers, reorder polnL (8C) sysLems and early maLerlal requlremenLs plannlng (M8) CompeLlLlve focus = cosLs roducL-focused manufacLurlng Plgh-volume producLlon MagneLlc Lape = only sLorage medlum o Cne-dlmenslonal, buL manufacLurlng = Lwo-dlmenslonal AvallablllLy of random access memory made M8 posslble 1970s: M8 and compuLer hardware and sofLware developmenLs CompeLlLlve focus = (LargeL-) markeLlng, producLlon lnLegraLlon and plannlng Lver fasLer and hlgher capaclLy dlsk (random access) sLorage lnLegraLlon beLween forecasLlng, masLer schedullng, procuremenL and shop floor conLrol M8 ll = manufacLurlng resource plannlng Mld 70s: blrLh of ma[or sofLware vendors (laLer key L8 vendors) o SA o Cracle o l8M 1973: l8M's manufacLurlng managemenL and accounL sysLems (MMAS) = precursor Lo L8 SynchronlzaLlon of new sofLware appllcaLlons wlLh release of new hardware sysLems 1980s: M8 ll CompeLlLlve focus = quallLy ! need for process conLrol ldea of lnLegraLed sofLware packages Lo replace several sLand-alone sysLems unlx operaLlng sysLem offered posslblllLy of real-Llme capLure of LransacLlons ! beLLer declslon supporL Cracle: SCL daLabase sysLem ! sofLware could be run on compuLers from dlfferenL vendors (flexlblllLy) lnLegraLlon of lnformaLlon ln a conslsLenL, effecLlve manner across Lhe enLerprlse ! L8 foundaLlon 1990s: M8 ll and early L8 sysLems # uramaLlc growLh of L8 sofLware and sysLems due Lo anLlclpaLed year 2000 (?2k) problem ! L8 as one way Lo address needed flxes Lo legacy sysLem sofLware LhaL was noL ?2k compllanL 1rend: lndusLry consolldaLlon 2000s: sofLware vendor consolldaLlon 2000-2002: pressure Lo downslze on sofLware companles Cne buslness growLh opLlon = acqulslLlons and mergers o L.g. eopleSofL/!.u. Ldwards merger " ComplemenLary sofLware producLs " LlLLle overlap ln sofLware offerlngs " More compleLe sofLware porLfollo # LefL lndusLry wlLh Lwo ma[ors players (Cracle and SA), buL wlLh sofLware capablllLles of orlglnal 3 players 1he fuLure of L8 L8 sysLems = maLure locus: easy conflguraLlon o ShorL lmplemenLaLlon cycles o Medlum and large pro[ecLs: sLlll pro[ecL managemenL lssues o 1allored L8 sofLware packages " nlche markeLs, nlche producLs & vendors o ush" lnformaLlon based on user-deflned lnLeresL areas o lnLelllgenL sysLems " SlmulaLlons
Lnterpr|se |ntegrat|on - ls process of achlevlng unlLy of efforL among Lhe varlous subsysLems ln Lhe accompllshmenL of Lhe organlzaLlon's Lasks. Lk has been developed as Lhe Lool Lo enable enLerprlse lnLegraLlon. ln facL, L8 sysLems have been adverLlsed as Lhe comprehenslve packaged sofLware soluLlons LhaL seek Lo lnLegraLe Lhe compleLe range of buslness' processes and funcLlons ln order Lo presenL a hollsLlc vlew of Lhe buslness from a slngle lnformaLlon and l1 archlLecLure. n|gh d|fferent|at|on and h|gh |ntegrat|on: - As an organlzaLlon expands, groups of employees are dlfferenLlaLed along varlous dlmenslons LhaL lnclude funcLlonal, producL and geographlcal speclallzaLlon. - CrganlzaLlonal boundarles can creaLe dlsparaLe workflows and subculLures LhaL can hlnder Lhe exchange of knowledge across dlfferenL deparLmenLs. - Such organlzaLlonal fragmenLaLlon can lower Lhe vlslblllLy requlred Lo provlde one-sLep soluLlons $ Lo cusLomers. - ln addlLlon, envlronmenLal facLors such as globallzaLlon, lnLense compeLlLlon and heavy regulaLlon are also demandlng such large organlzaLlon Lo work closely LogeLher ln order Lo enacL changes qulckly LhroughouL Lhe whole organlzaLlon. - lL ls Lherefore opLlmal for an organlzaLlon Lo have boLh a hlgh level of dlfferenLlaLlon as well as a hlgh level of lnLegraLlon. 1he greaLer level of dlfferenLlaLlon, Lhe more dlfflculL lL ls Lo achleve a hlgh level of lnLegraLlon. - lnLegraLlon should be dynamlc, allowlng people Lo handle changes as well as Lo vary Lhelr sLrengLh of lnLerdependencles beLween Lhe componenLs.
Understand|ng Lnterpr|se Integrat|on: - ueflnlLlons of lnLegraLlon are ofLen brushed across lnLulLlvely as coordlnaLlon and cooperaLlon - Coord|nat|on - ls deflned as managlng dependencles beLween acLlvlLles - Cooperat|on - refers Lo mulLlple parLles worklng LogeLher - Integrat|on - Lhe managemenL of dependencles beLween Lhe componenLs for comblned beneflLs. ln Lhe auLhors vlew lL ls much more Lhan LhaL. lnLegraLlon should brlng LogeLher componenLs LhaL are lndependenL Lo each oLher. lL lnLroduces opporLunlLles for Lhe dlvlslons Lo coordlnaLe and collaboraLe ln Lhe fuLure. - Lnterpr|se |ntegrat|on - one sLream of deflnlLlons emphaslzed Lhe deflnlLlon on buslness componenLs of Lhe enLerprlse, coverlng aspecLs such as buslness funcLlons, knowledge and process lnLegraLlon. L.g. lnLegraLlon makes mulLlple unlLs, funcLlons and slLes of large organlzaLlons work LogeLher Lo lncrease capaclLy, lmprove performance, lower cosL sLrucLure and dlscover opporLunlLles for lmprovemenL LhaL do noL appear unLll you look across funcLlons" - Lnterpr|se |ntegrat|on - anoLher sLream deflnes Ll Lhrough Lhe lnLegraLlon of Lechnologlcal sysLems. lL reflecLs Lhe capablllLy Lo lnLegraLe a varleLy of dlfferenL sysLem funcLlonallLles". - Lnterpr|se |ntegrat|on - auLhors: whaL appears Lo be mlsslng ls Lhe dlscusslon of enLerprlse lnLegraLlon aL a more sLraLeglc level. lnLegraLlon should resulL ln synerglsLlc ouLcomes such as aLLalnlng beLLer revenues or Lhe ease Lo pursue new opporLunlLles. - 8us|ness |ntegrat|on - refers Lo Lhe creaLlon of LlghLer coordlnaLlon among Lhe dlscreLe buslness acLlvlLles conducLed by dlfferenL lndlvlduals, so LhaL a unlfled buslness process ls formed. % - System |ntegrat|on - refers Lo Lhe creaLlon of LlghLer llnkages beLween dlfferenL compuLer-based lnformaLlon sysLems and daLabases.
1he def|n|t|on of |ntegrat|on: btloqloq compooeots toqetbet ooto o commoo plotfotm, wblcb cteotes tbe optloos fot compooeots to estobllsb, cbooqe ot voty tbelt stteoqtb of lotetJepeoJeoce omooq ooe oootbet, fot o syoetqlstlc ootcome.
Lk sysLems are lnLended Lo enable seamless flow of lnformaLlon across Lhe whole organlzaLlon. ln plannlng lLs L8-enabled enLerprlse lnLegraLlon sLraLegles, an organlzaLlon should Lake lnLo accounL Lhe sLraLeglc, buslness and sysLem facLors LhaL may affecL Lhe effecLlveness of Lhe enLerprlse lnLegraLlon. ! Cne buslness facLor LhaL can affecL Lhe effecLlveness of L8-enabled enLerprlse lnLegraLlon sLraLegles ls Lhe level of slmllarlLy beLween buslness Lasks, buslness unlLs and buslness processes. ! ComplexlLy and Lask lndependence affecL Lhe sulLablllLy of mechanlsms LhaL enable lnLegraLlon ! SLandardlzaLlon ls llkely Lo be more approprlaLe Lhan merely asklng Lhe affecLed unlLs Lo manually ad[usL Lo each oLher ! 8ecause of daLa and process sLandardlzaLlon ln L8 sysLem, lnLegraLlon becomes more dlfflculL and achlevemenL of beneflLs are less llkely lf Lhe unlLs belng lnLegraLed are hlghly dlfferenLlaLed ! Soclal sLrucLures play an lmporLanL role: e.g. globally dlspersed organlzaLlons LhaL seek Lo achleve enLerprlse lnLegraLlon wlll have Lo manage culLural facLors such as Lhe power dlsLance among peers, subordlnaLed and superlors as well as Lhe preference for collecLlve or lndlvldual acLlvlLles
llndlngs of Lhe paper: $ 1he enLerprlse lnLegraLlon sLraLegles were unclear and people aL dlfferenL hlerarchlcal levels had dlfferenL lnLerpreLaLlon of Lhe enLerprlse lnLegraLlon ob[ecLlves $ 1he noLlon of one enLerprlse" was noL spread clearly Lo Lhe sLaff ln Lhe lower hlerarchlcal levels Clven Lhe varlous lnLegraLlon posslblllLles, senlor managemenL musL declde on whaL one enLerprlse" means Lo Lhem and communlcaLe Lhls dlrecLlon of lnLegraLlon wlLh clarlLy and commlLmenL down Lo Lhe resL sLaff ln Lhe organlzaLlon. $ Lack of clarlLy abouL Ll scope and prlorlLlzaLlon of ob[ecLlves encouraged many users Lo Lhrow ln" as many requlremenLs as posslble durlng deslgn phase $ 1he vlslon of enLerprlse lnLegraLlon was noL kepL consLanLly ln slghL $ Cross-8u lnformaLlon vlslblllLy was noL glven much aLLenLlon $ Large no. Cf sLakeholders made lL unclear where Lhe auLhorlLy Lo make declslons lay & $ LxLernal companles were relucLanL Lo allocaLe Llme Lo LesL Lhe lnLerface $ lnformaLlon sharlng ls noL a common pracLlce - soclal barrler lragmenLaLlon ln buslness ls reflecLed when some users are unable Lo undersLand how Lhelr acLlon would affecL Lhe oLher deparLmenLs. ConLlnuous efforLs should be made Lo coordlnaLe across unlLs Lo connecL separaLe sllos LogeLher.
uwlzeyemungu, L. 8aymond, L|nk|ng the Lffects of Lk to Crgan|zat|ona| erformance: Deve|opment and In|t|a| Va||dat|on of an Lva|uat|on Method, lnformaLlon SysLems ManagemenL, 27:23-41, 2010
aper presenLs a meLhod Lo evaluaLe Lhe conLrlbuLlon of an L8 sysLem Lo organlzaLlonal performance. lLs orlglnallLy lles ln Lhe facL LhaL Lhe whole range of performance lndlcaLors acLually used by managers are lnLegraLed ln Lhe evaluaLlon of Lhe sysLem.
% L8 lmplemenLaLlon remalns however one of Lhe mosL slgnlflcanL challenges for lS pracLlLloners ln Lhe pasL decade". % L8 lmplemenLaLlon pro[ecLs requlre large lnvesLmenLs and enLall conslderable rlsks LhaL musL be mlLlgaLed ln an approprlaLe manner. 1hls paper alms Lo propose a new quallLaLlve meLhod for Lhe ex-posL evaluaLlon of L8 sysLems. 1hls meLhod llnks Lhe L8 sysLem on one hand wlLh Lhe lndlcaLors of organlzaLlonal performance, Lhrough Lhe auLomaLlon, lnformaLlonal and LransformaLlonal effecLs LhaL resulL from Lhe lmplemenLaLlon and use of Lhe sysLem.
roblems ln evaluaLlng Lhe effecLs of l1 on performance: $ lnLanglblllLy of numerous cosLs and beneflLs assoclaLed Lo l1 $ 1he manlfesLaLlon of l1 effecLs over Llme and aL varlous level $ 1he dlfflculLy ln dlssoclaLlng l1 from Lhe work sysLems Lhey supporL.
LvaluaLlon Models of l1 LffecLs: & Casual models - aLLempL Lo demonsLraLe a cause-effecL relaLlonshlp beLween l1 lnvesLmenLs and organlzaLlonal performance & rocess models - preoccupled wlLh Lhe chaln of evenLs by whlch l1 lnvesLmenLs provlde value added Lo Lhe organlzaLlon. & ConLlngency models - underllned by Lhe ldea LhaL Lhe effecLs of l1 on organlzaLlonal performance are noL dlrecL, buL Lhey dlffer wlLh Lhe former ln LhaL Lhey propose LhaL Lhls lmpacL depends upon Lhe allgnmenL or flL" of l1 wlLh oLher fundamenLal dlmenslons of Lhe flrm's managemenL such as lLs sLraLegy, sLrucLure and envlronmenL. & 8alanced scorecard (8SC) - used Lo evaluaLe lS/l1 performance ln general and L8 sysLems ln parLlcular. ln case of L8 evaluaLlon, lL comblnes and lnLer-relaLes mulLlple perspecLlves. 1hanks Lo 8SC lL was recognlzed LhaL Lhe way ln whlch Lhe sysLem ls deslgned and Lhe manner ln whlch lL ls used and managed are lmporLanL Lo achleve l1 lnvesLmenL ob[ecLlves.
I|nd|ngs - Lk effects: '
L8 evaluaLlon meLhods ln seven sLeps:
(
Tutorial 2
LlghL, 1he ma|ntenance |mp||cat|ons of the custom|zat|on of Lk software, !ournal of SofLware MalnLenance and LvoluLlon: research and pracLlce, 2001
1be totlooole fot kl softwote oJoptloo ooJ lts costomlzotloo lmproved sofLware sLrucLure & ouLsourclng ablllLy of malnLenance 1ap lnLo avallable skllls base for sofLware 8educed levels of enLropy & lmproved documenLaLlon 8esoluLlon of year 2000 problem (L8 sofLware = year 2000 compllanL) Common l1 and org. blueprlnL (useful ln Lhe conLexL of globallzaLlon, M&A, eLc.) Chance Lo adopL besL pracLlce buslness models embedded wlLhln Lhe sofLware costomlzotloo - cbooqes / oJJltloos to tbe ovolloble fooctlooollty lo tbe stooJotJ kl softwote Cenerlc cusLomlzaLlon: wrlLLen once and used ln Lhe whole org. Local cusLomlzaLlon: wrlLLen once, buL only used locally undersLandlng Lhe malnLenance lmpllcaLlons of L8 sofLware cusLomlzaLlon )
1. Change funcLlonallLy Changlng Lhe sofLware CusLomlzaLlon needs Lo be re-wrlLLen, re-LesLed and re-lmplemenLed upgrades of Lhe L8 sysLem may requlre malnLenance of Lhe cusLomlzaLlons 2. Addlng funcLlonallLy lncorporaLlng addlLlonal funcLlonallLy absenL from Lhe sLandard L8 sofLware MalnLenance = upgrades & malnLenance of Lhe llnk beLween cusLom developmenL and Lhe L8 sofLware 3. rocess auLomaLlon Could requlre conslderable reworklng when upgrades arrlve 4. Amendlng reporLs / dlsplays Lach change ln lLself = small, buL malnLalnlng all changes & assoclaLed documenLaLlon = slgnlflcanL underLaklng o Can be underLaken aL nexL upgrade of L8 sysLem ! uslng L8 sofLware Lo reduce malnLenance efforL (= orlglnal drlver for Lhe declslon Lo lmplemenL) 3. new reporLs no llkellhood of upgrades, unless new daLa needs Lo be enLered ln Lhe reporLs MalnLenance efforL = creaLlon of reporLs 8lsk: sLaff creaLlng own reporLs o 8educed Lhrough lmplemenLlng a reporL generaLor package cooclosloo. Level of accepLance of L8 sofLware: complex L8 sofLware cannoL always accommodaLe organlzaLlonal demands o CusLomlzaLlon vs. malnLenance " upgrades (mlghL Lake a loL of efforL) " Cngolng malnLenance ouLslde of any upgrade lmplemenLaLlon " 8lsk managemenL Culckness of malnLenance lssues CosLs of malnLenance CuLsourclng malnLenance Lo L8 vendor, where sLandard sofLware ls noL cusLomlzed " L8 sofLware llfecycle cosLs musL be consldered
Luo & SLrong, A Iramework for Lva|uat|ng Lk Imp|ementat|on Cho|ces, 2004
& Successful L8 lmplemenLaLlon requlres flL beLween L8 sysLem & supporLed org. processes * & AssumpLlons of L8 vendors don'L maLch processes ! sysLem or processes adapLed & L8 lmplemenLaLlon = serles of lndependenL cusLomlzaLlon and lmplemenLaLlon pro[ecLs & L8 characLerlsLlcs: packaged sofLware, complexlLy, sysLem adapLaLlon, CrganlzaLlon adapLaLlon & 8esource-based perspecLlve: L8 lmplemenLaLlon affecLed by organlzaLlon's l1-based resources (+ how Lhey are assembled, coordlnaLed & deployed) 1.) 1anglble l1 resources (e.g. l1 lnfrasLrucLure) 2.) lnLanglble l1 resources (e.g. knowledge bases) 3.) Puman l1 resources (e.g. Lechnlcal and managerlal skllls) CusLomlzaLlon 1.) 1echnlcal cusLomlzaLlon 1.1) Module selecLlon (whlch modules Lo lmplemenL) 1.2) 1able conflguraLlon (selecL conflguraLlon opLlons ln Lhe Lables, very complex & Llme- consumlng when lnLerdependencles among opLlons across varlous Lables & modules, beneflLs: Lallor wlLhouL codlng, full vendor supporL, ease of fuLure upgradlng) 1.3) Code modlflcaLlon (some sysLems Lo supporL cusLomlzaLlon, hlghesL cosL & rlsk, posslble lncompaLlblllLy wlLh newer verslons of sysLem, some lnLegraLlon beneflLs may be losL) 2.) rocess cusLomlzaLlon 2.1) no change (only changes ln Lasks & resources) 2.2) lncremenLal change (also changes ln relaLlonshlps among Lasks & relaLlonshlps among Lasks & resources) 2.3) 8adlcal change (fundamenLal reLhlnklng & radlcal redeslgn of buslness processes, lncl. performance measures)
CapablllLy requlremenLs for L8 cusLomlzaLlon: 1.) 1echnlcal change capablllLles 1.1) undersLandlng defaulL L8 sysLem processes, conflguraLlon & bullL-ln opLlons 1.2) AblllLy Lo develop & modlfy large-scale sofLware ln a neLworked daLabase envlronmenL 1.3) AblllLy Lo manage large-scale sysLems developmenL pro[ecLs "+ 2.) rocess change capablllLles 2.1) undersLand exlsLlng buslness processes & buslness envlronmenL 2.2) AblllLy Lo deslgn new or changed buslness processes, as well as lmplemenL Lhese deslgns 2.3) Capable of managlng & coordlnaLlng large-scale buslness process changes
' CrganlzaLlon consldered Lo have hlgh capablllLy of elLher Lype, lf lL has broad scope and greaL depLh ln all Lhree relevanL ablllLles
' 8eslsL LempLaLlon Lo over cusLomlze (e.g. over cusLomlze sysLem as Lechnlclan)
' ulfferenL capablllLles needed for Lypes of L8 cusLomlzaLlon:
' 8oLh Lechnlcal change & process change capablllLles wlll change over Llme as companles go Lhrough L8 lmplemenLaLlon process ! perlodlcally evaluaLe capablllLles ! plan paLh of lmplemenLaLlon pro[ecLs (more dlfflculL pro[ecLs become feaslble Lhrough learnlng)
Puq, Puq, CuLrlghL, 8k through Lk: Avo|d|ng change management p|tfa||s, !ournal of Change ManagemenL, 2007.
! ConLrlbuLlon of L8 sysLems Lowards 88 lmplemenLaLlon can be slgnlflcanL ln Lerms of scope, conflgurablllLy and lnLegraLlveness ! SLudy uses open-ended quesLlonnalre (3 companles: PosplLal, global consumer company, global consulLlng company) uslng slx mgmL. dlmenslons Change managemenL dlmenslons: 1.) Leadershlp - descrlbes Lop managemenL's ablllLy Lo dlrecL roles and responslblllLles based on company mlsslon sLaLemenL "" 2.) 8arrlers Lo change - descrlbes how barrlers Lo reachlng consensus, barrlers Lo learnlng, barrlers Lo moLlvaLlon and barrlers Lo operaLlonal success were ldenLlfled. 3.) CommunlcaLlons - descrlbes how employees, parLners and suppller are educaLed abouL Lhe change, whaL was done Lo obLaln supporL from Lhem. 4.) lmplemenLaLlon of change and conLrol - descrlbes Lhe hlerarchy of declslons, dlssemlnaLlon of Lhe change concepLs, Llme scales and ma[or checkpolnLs of Lhe pro[ecL, progress revlew and needed lnLervenLlons Lo overcome obsLacles. 3.) eople culLure facLor - descrlbes how employees were empowered, wheLher leadershlp was shared, whaL klnd of efforL were puL lnLo removlng culLural barrlers and whaL klnd of problem-solvlng Lools were used Lo reach consensus. 6.) Change revlew - descrlbes how company measures Lhe aLLalnmenL of Lhe sLraLeglc, LacLlcal and conLrol goals. WhaL ls 88? 88 ls noL downslzlng or resLrucLurlng, lL's noL lndenLed Lo lmprove shorL-Lerm flnanclal resulLs by geLLlng rld of people and [obs, 88 ls a culmlnaLlon of varlous managemenL pracLlces & schools of LhoughL As opposed Lo 1oLal CuallLy ManagemenL, 88 does noL focus on lncremenLal change or gradual lmprovemenL of processes, raLher lL seeks radlcal redeslgn and drasLlc lmprovemenL of Lhe processes Lk |mp|ementat|on |ssues 1.) MaLchlng funcLlonallLy 2.) CrganlzaLlonal reslsLance 3.) 1echnlcal supporL & lnLegraLlon Avo|d|ng the p|tfa||s (a|ong 6 management d|mens|ons): 1.) uevelop mlsslon sLaLemenL & shared values, deflne roles & responslblllLles, expecLed ouLcomes ( Why change ls needed, why 88 efforL useful, how mgmL. enable employees Lo lmplemenL change ( 1op managemenL commlLmenL (!) ( Appralse employees of poLenLlal changes Lo deslgn & scope, waLch for unreallsLlc worker expecLaLlons (employees hold mgmL. Lo lLs promlses) 2.) ollLlcs: lf change aLLempLs LhreaLen power & preferences of groups ! confllcL & sLruggles ( CeL supporL of well-known lndlvlduals & oplnlon leaders (1sL sLep!) ( Pelp employees say goodbye" Lo old Lechnology, help Lo spend some Llme ln Lhe neuLral zone (l.e. 88 ldeas)(l.e. offer enough lead Llme Lo geL used Lo new processes), help Lhem come Lo grlps wlLh new ldeas 3.) LffecLlve communlcaLlon can dlspel confuslon & employee reslsLance ( LducaLe employees ( Cffer slngle polnL of conLacL (preferably lnLraneL / corporaLe porLal) ( Clve opLlon Lo offer oplnlons, commenLs & suggesLlons ( CommunlcaLlon of change ldes Lhrough mulLlple channels 4.) arLnershlp lS & process Leams Lo lead re-englneerlng efforLs "# ( reparaLlon, Analysls (Lop mapplng of exlsLlng sysLems), deslgn (of new processes), fleld LesLlng, acLual lmplemenLaLlon ( AL preparaLlon, mgmL. needs Lo play leadershlp & communlcaLlon role ln assesslng buslness sLraLegy, deflnlng Lhe pro[ecL's scope & lnlLlaLlng Leam bulldlng ( lmplemenLaLlon & conLrol ! Leam bulldlng (!) 3.) CulLure seLs boLh llmlLs & dlrecLlon of movemenL of behavlor ln organlzaLlons ( CulLure can only be changed when employees learn new seLs of behavlor ( L8-moLlvaLed 88 requlres culLural reform (new culLure LhaL supporLs new work procedures) ( Lmployees should know whaL ls expecLed from Lhem & whaL Lo expecL from Lhe company ( Clve employees Llme Lo accepL org. change efforL 6.) AssessmenL (! guaranLees process lnLegraLlon and rellablllLy!) ( ro[ecL progress ln comparlson Lo Lhe seL ouLllne ( AssessmenL of how Lhe L8 sysLem works across funcLlonal boundarles ( ConLlngency plans for any unforeseen obsLacles ( CuaranLee LhaL new re-englneered processes are foolproof (helps avoldlng surprlses & evenLual fallure of pro[ecL)
Tutorial 3
llqbt, nollooJ & wllls, Lk and best of breed: a comparat|ve ana|ys|s, 8uslness rocess ManagemenL !ournal, vol. 7 no. 3, 2001.
L8 crlLlclzed for lnflexlblllLy and noL meeLlng speclflc organlzaLlon and lndusLry requlremenLs. L8 sofLware funcLlonallLy ls lacklng, Lhe lmpllclL buslness model does noL represenL Lhelr own and Lherefore reenglneerlng buslness processes ln llne wlLh Lhls presenLs ma[or dlfflculLles. L8 sulLes Lend Lo only have one besL class appllcaLlon (eoplesofL for P8, Cracle for flnanclals eLc.)
8o8 (besL of breed) ls a sLraLegy LhaL lnLegraLes componenLs of sLandard package and/or cusLom sofLware (comblnlng ln-house and Lhlrd parLy sofLware). uue Lo Lhe absence of besL ln class sLandard sofLware some companles have developed cusLom componenLs Lhemselves. o 1he ob[ecLlve of 8o8 ls Lo develop enLerprlse sysLems LhaL are more closely allgned wlLh Lhe buslness process of an organlzaLlon. 1he alm for Lhe sLraLegy ls enLerprlse lnLegraLlon and process orlenLaLlon. o 1hls approach also provldes an lnfrasLrucLure LhaL accommodaLes Lhe lmplemenLaLlon of new or lmproved appllcaLlon and buslness process, Lhereby provldlng companles wlLh a consLanL sLaLe-of-arL capablllLy. o Lach 8o8 componenL can be lmplemenLed as a sLand-alone appllcaLlon.
WlLh Lhe L8 sysLem one sulLs lLs buslness process Lo Lhe sysLem, whlle ln Lhe 8o8 you sulL Lhe sysLem Lo flL your processes.
80/20 rule approach - Lo shorLen Lhe Llme consumlng dlscusslon leadlng Lo consensus, you only alm Lo meeL 80 of deslred funcLlonallLy by your sofLware and Lhe remalnlng 20 would be lefL Lo addlLlonal MlS developmenLs ouLslde of Lhe package. "$
C|ean s|ate approach - lmplemenL Lhe besL pracLlces Lo your buslness processes from Lhe vendor. C|ean s|ate reeng|neer|ng - you redeslgn your processes yourself.
& Increased comp|ex|ty - dlglLal Lechnology noL only deallng wlLh complexlLy buL also creaLlng lL o lalse assumpLlon LhaL more compuLers ls beLLer o lor every 23 lncrease ln complexlLy (as Lo Lolerance of amblgulLy, llloglcal quesLlons or lnconslsLencles posed by users) Lhere ls an lncrease ln Lhe sofLware complexlLy by 100 o As Lhe complexlLy ls lncreased, people geL losL ln Lhe lnLerconnecLedness - no one knows whaL wlll happen lf you change one parL of Lhe sysLem
& St||| the same |egacy systems - dlfferenL dlvlslons or faclllLles ofLen made lndependenL purchases, many companles ended up havlng several lnsLances of Lhe same L8 sysLems o ln Lhe end L8 sysLems became [usL anoLher subseL of Lhe legacy sysLems Lhey were supposed Lo replace & Cost of |mp|ementat|on - 73 of L8 lmplemenLaLlons were consldered fallures "% o 1here ls llLLle sLaLlsLlcal evldence LhaL Lhe beneflLs of L8 lmplemenLaLlon ouL-welgh Lhe cosLs and rlsks o LnLerprlse sysLems noL only falled Lo dellver on Lhelr grand promlse, buL ofLen slmply aggravaLed Lhe problems Lhey were supposed Lo solve" & Custom|zat|on vs. p|a|n van|||a o lmplemenLlng plaln-vanllla Lype of Lhe sysLem doesn'L glve you any comparaLlve advanLage o CusLomlzlng Lhe already complex L8 sofLware creaLes even more complexlLy and even larger rlsks " WlLhouL lnLlmaLe knowledge of how Lhe lnLegraLed pleces of modules acLually work leads Lo ln-house bugs LhaL are hard Lo foresee and expenslve Lo flx " SofLware conLalns unknown defecLs, lL's noL posslble Lo guaranLee LhaL Lhe program as a whole has no flows & Data rocess|ng o 8ecause of Lhe dlfflculLy and complexlLy of Lhe new sofLware lLself, companles Lend Lo forgeL abouL Lhe daLa, necesslLaLlng enormous reworklng Lo properly clean up Lo lnLegraLe Lhe daLa. o 8lg amounL of daLa conLalnlng myrlad hldden errors only adds Lhe complexlLy and uncerLalnLy of Lechnologlcal soluLlons
& sycho|og|ca| barr|er o 8uslness execuLlves wanL Lo conLlnue Lo belleve LhaL Lechnology wlll lower cosLs, lmprove processes and reduce Lhe slze of workforce o LxecuLlves don'L llke Lo hear abouL downsldes of Lechnology
$ SCA (Serv|ce Cr|ented Arch|tecture) - lL proposes Lo overcome Lhe problem lnvolved wlLh updaLlng and changlng legacy sysLems by bulldlng modular cross-sysLem buslness processes. 1hese processes would connecL Lhe relevanL pleces of funcLlonallLy from varlous l1 sysLems, Lhereby maklng lL easler Lo change processes Lo adapL Lo ne buslness goals. o nowever many dlfflculL Lechnlcal problems musL be solved before SCA can become Lhe backbone for a new sLraLeglc archlLecLure, lncludlng robusL proLocols for accesslng Lhe appllcaLlons, hlgh-quallLy lnLegraLed daLa sLores and a sound meLhodology for managlng Lhe over all processes. "& o 1he ha||mark of SCA ls Lhe fundamenLal modularlLy of lLs sofLware buslness processes. Companles can bulld (or purchase) buslness modules for Lhelr core processes. o SLlll Lhe comp|ex|ty ls carrled over wlLh Lhls Lechnology. SCAs become addlLlonal layers of code superlmposed on Lhe exlsLlng layers. 1haL means lL ls posslble LhaL a process wlll fall aL some polnL due Lo some faulL ln Lhe layers below and ln order Lo undersLand and flx LhaL problem, sofLware englneers wlll need Lo deal wlLh Lhe layers of enLerprlse appllcaLlons below Lhe modular buslness process.
lece of advlce? - Closer and closer communlcaLlon and collaboraLlon beLween l1 and buslness sldes of Lhe organlzaLlon.
lostet ooJ 1oecke, Descr|b|ng the L|ephant: 1he D|fferent Iaces of I1 as Serv|ce, Cueue, volume 3 lssue 6, !uly/AugusL 2003
uescrlpLlon of ma[or Lechnology Lrends LhaL are drlvlng many relaLed efforLs, namely Lhe LransformaLlon from verLlcally lnLegraLed buslness unlLs Lo horlzonLally lnLegraLed, servlce-orlenLed sysLems.
1he organlzarlon of l1 resources as a seL of more-or-less lndependenL sllos, each responslble for a dlsLlncL enLerprlse funcLlon or appllcaLlon ls a commonplace. lL's noL uncommon for lndlvldual sllos Lo be ldle 90 of Lhe Llme because of Lhe need Lo provlde excess capaclLy for occaslonal peak loads.
uecooplloq vettlcolly ooJ loteqtotloq botlzootolly of Lhe l1 sLrucLures allows for reuse of funcLlon wlLhln appllcaLlons and economles of scale ln Lerms of resource usage. verLlcal decoupllng - sLandardlzlng lnLerfaces among appllcaLlon componenLs PorlzonLal lnLegraLlon - adopLlng unlform managemenL lnLerfaces so LhaL large number of resources, dlsLrlbuLed over whaL used Lo be dlsLlncL sllos, can be allocaLed, used, monlLored and managed ln a common and auLomaLed manner, lmprovlng uLlllzaLlon and reduclng operaLlons cosLs
CtlJ - a sysLem LhaL uses open, general purpose proLocols Lo federaLe dlsLrlbuLed resources and Lo dellver beLLer-Lhan-besL-efforL quallLles of servlces. Crld Lechnologles allow: 1) Cn-demand access Lo compuLlng capablllLles, 2) 1he federaLlon of dlsLrlbuLed resources Lo meeL end-user requlremenLs. CtlJ loftosttoctote - horlzonLal lnfrasLrucLure lnLegraLlon layer.
utlllty compotloq. a) A separaLlon beLween servlce provlder and consumer, b) 1he ablllLy Lo negoLlaLe a deslred quallLy of servlce from Lhe provlder.
Oo-JemooJ - broad Lerm used Lo denoLe Lechnologles and sysLems "' LhaL allow users or appllcaLlons Lo acqulre addlLlonal resources Lo meeL changlng requlremenLs.
uoto ceotet ootomotloo - refer Lo producLs LhaL enable Lhe coordlnaLed managemenL of resources wlLhln an enLerprlse, e.g. Lo keep large number of machlnes up Lo daLe wlLh Lhe laLesL paLches.
clostet - denoLes a non-shared-memory. 1he relaLlvely low cosL of clusLers makes Lhem excellenL power planLs for grld/uLlllLy/on-demand compuLlng sysLems.
Ootsootcloq - a Lhlrd parLy offers Lo run all or parL of an enLerprlse's l1 operaLlons.
5oftwote os o setvlce - Lhe Web ls used Lo provlde many cusLomers wlLh access Lo funcLlons LhaL have been speclflcally deslgned for Lhls mode of use. SaaS ls an approach Lo wrlLlng appllcaLlons and exposlng lnLerfaces Lo users.
5etvlce-OtleoteJ Atcbltectote - an approach Lo deslgnlng sysLems LhaL faclllLaLes Lhe reallzaLlon of Lhe l1-as-servlce and horlzonLal lnLegraLlon goals menLloned earller. An SCA ls a seL of servlces LhaL seeks Lo achleve Lhe clean separaLlon of lnLerface and lmplemenLaLlon needed Lo reallze oLher deslrable properLles such as locaLlon Lransparency and loos coupllng beLween servlce and cllenL. web setvlces - seL of Lechnologles for reallzlng SCA. 5etvlce-OtleoteJ loftosttoctote - Lhe use of SCA approaches Lo Lhe problem of resource managemenL.
PorlzonLal lnLegraLlon: Appllcotloos use wotklooJ moooqets Lo coordlnaLe Lhelr access Lo physlcal resources. AppllcaLlon and lLs workload manager are noL verLlcally lnLegraLed. lnsLead Lhey blnd dynamlcally Lo resources vla common qtlJ loftosttoctote. 1he resources Lhemselves may lmplemenL varlous vlttoollzotloo approaches Lo enhance Lhe flexlblllLy wlLh whlch Lhey serve Lhelr users.
LnLerprlses LhaL wanL Lo supporL a range of appllcaLlons on shared resources face Lhe problem LhaL dlfferenL appllcaLlons and Lhelr assoclaLed workload managers do noL lnLegraLe aL Lhe lnfrasLrucLure level. SoluLlon: lnLroducLlon of common horlzonLal layer LhaL deflned and lmplemenLs a conslsLenL seL of absLracLlons and lnLerfaces for access Lo and managemenL of shared resources. 1hls ls called qtlJ loftosttoctote. A grld lnfrasLrucLure musL provlde a seL of Lechnlcal capablllLles as follows: % 8esource modellng - descrlbes avallable resources, Lhelr capablllLles and Lhe relaLlonshlps "( beLween Lhem Lo faclllLaLe dlscovery, provlslonlng and quallLy of servlce managemenL. % MonlLorlng and noLlflcaLlon - provldes vlslblllLy lnLo Lhe sLaLe of resources % AllocaLlon - assures quallLy of servlce across an enLlre seL of resources for Lhe llfeLlme of Lhelr use by an appllcaLlon % rovlslonlng, llfe-cycle managemenL and decommlsslonlng - enables allocaLed resource Lo be conflgured auLomaLlcally for appllcaLlon use % AccounLlng and audlLlng - Lracks Lhe usage of shared resources and provldes mechanlsms for Lransferrlng cosL among users
vlttoollzotloo tecbooloqles - Lechnologles LhaL lmplemenL a layer on resources LhaL boLh provldes flexlble conLrol of Lhe physlcal resource concepL (e.g. wlLh respecL Lo performance) and aL Lhe same Llme supporLs mulLlple vlrLual lnsLances on Lhe same physlcal resource wlLh good lsolaLlon.
Tutori al 4
Lee, Slau and Pong, Lnterpr|se |ntegrat|on w|th Lk and LAI, CommunlcaLlons of Lhe ACM, vol. 46, no. 2, 2003
Lnterpr|se App||cat|on Integrat|on (LAI) - auLomaLes Lhe lnLegraLlon process wlLh less efforL Lhan LhaL requlred wlLh L8. 1o some degree, LAl can even lncorporaLe L8, Lhus servlng Lo connecL L8 sysLems.
1wo dlfferenL approaches: lotetoollzotloo represenLed by L8 and xtetoollzotloo Lhrough Lhe use of LAl
( !"#$%"&'()&#(*" represented by Lk " LnLerprlse lnLegraLlon - capablllLy Lo lnLegraLe a varleLy of dlfferenL sysLem funcLlonallLles " ln early 1990s, Lwo dlsLlncL sysLem lnLegraLlon approaches: L8 addresslng operaLlonal lnLegraLlon and daLa warehouslng focuslng on lnformaLlonal lnLegraLlon Lo supporL declslon- maklng. " ln mld 1990 LlA emerged Lo make sysLem lnLegraLlon posslble wlLh lower cosLs and less programmlng. " 1echnlcal and 8ehavloral lnLegraLlon - Lo achleve aglllLy and flexlblllLy, lL ls necessary Lo have boLh Lechnlcal and behavloral lnLegraLlon. 1he blggesL challenge may be Lhe behavloral lnLegraLlon (e.g. redlsLrlbuLlon of roles and responslblllLles).
LnLerprlse 8esource lannlng - ploneer ln enLerprlse lnLegraLlon by offerlng a sysLem LhaL accompllshed Lhe lnLegraLlon of dlfferenL operaLlonal and LransacLlon daLa. ' ln order Lo lmplemenL L8, mosL companles musL flrsL reenglneer Lhelr buslness processes Lo adopL L8 sLandard buslness processes. o lor some companles lL's advanLageous as Lhey can geL rld of old legacy ") sysLems eLc., for oLhers lL's lmposslble ' Approaches Lo lmplemenLlng L8: comptebeoslve (lnvolvlng LoLal efforL Lo lmplemenL all modules of Lhe L8 package wlLh buslness process reenglneerlng), voolllo (less buslness process reenglneerlng and requlrlng L8 funcLlonallLles ln only one slLe) ooJ mlJJle-tooJ (an approach LhaL falls beLween Lhe oLher Lwo exLremes). ' Maln dlsadvanLage of L8: lnLernal focus, now seen as an dlsadvanLage due Lo lncreaslng focus on changes ln Lhe exLernal envlronmenL
( ,-#$%"&'()&#(*" through the use of LAI ' 8aslc concepL of LAl ls Lhe exLernallLy of enLerprlse lnLegraLlon wlLh lower cosLs and less programmlng uslng exlsLlng appllcaLlons. ' LlA - buslness-compuLlng Lerm for ploos, metboJs ooJ tools almed aL moJetolzloq, coosollJotloq ooJ cootJlootloq Lhe overall compuLer funcLlonallLy ln an enLerprlse. ' LAl may lnvolve developlng LoLally new ouLlook of an enLerprlse's buslness and lLs appllcaLlons, deLermlnlng how exlsLlng appllcaLlons flLs lnLo Lhe new vlew and Lhen devlslng ways Lo efflclenLly reuse whaL already exlsLs whlle addlng new appllcaLlons and daLa. ' LAl uses speclal mlddleware LhaL serves as a brldge beLween dlfferenL appllcaLlons for sysLem lnLegraLlon. All appllcaLlons are freely communlcaLlng wlLh each oLher Lhrough common lnLerface layer raLher Lhan Lhrough polnL-Lo-polnL lnLegraLlon. (LxLenslve programmlng ellmlnaLed) ' LAl approach has been embraced by companles due Lo lLs ablllLy Lo reuse exlsLlng compuLer appllcaLlons Lo sLreamllne processes and Lhe relaLlve convenlence of conflguraLlon. ( LAl lmplemenLaLlon lnvolved an exLenslve long-Lerm lnvesLmenL ln deslgn. (1lme consumlng) ( LAl lmplemenLaLlon requlres LhaL Lhere ls a sLrong communlcaLlon, coordlnaLlon and cooperaLlon beLween lnformaLlon Lechnology and buslness personnel. ( LAl requlres buslness-mapplng processes, as a crlLlcal aspecL here ls Lo comblne separaLe sysLems' buslness processes.
neverLheless Lhe ulLlmaLe goal of LAl ls Lhe flexlblllLy or aglllLy LhaL carefully archlLecLed lnLegraLlon brlngs Lo Lhe enLerprlse, permlLLlng rapld response Lo new buslness opporLunlLles.
Lk or LAI? "* Lk ls: ) ush-orlenLed as L8 forces an organlzaLlon Lo accepL sLandard lnLegraLed buslness processes - Lhls can generaLe a slgnlflcanL reslsLance from members of an organlzaLlon. ) 8oLLom-up approach, as lndlvlduals wlLhln Lhe organlzaLlon cannoL selecL Lhelr own buslness processes for use ln Lhe new sysLem. ) L8 supporLs cenLrallzed buslness sLraLegy LAI ls: ) ull-orlenLed as Lhe exlsLlng appllcaLlons and buslness processes are used Lo map and lnLegraLe separaLe funcLlonallLles of an enLerprlse lnLo a form LhaL ls more accepLable by Lhe employees. ) LAl supporLs decenLrallzed buslness processes
Cerlcke eL al., Success lacLors of AppllcaLlon lnLegraLlon: An LxploraLory Analysls, CommunlcaLlons of Lhe AssoclaLlon for lnformaLlon SysLems, 2010.
Mckeen, uevelopmenLs ln lS pracLlce: Lnterpr|se App||cat|on Integrat|on, Commun|cat|ons of the AIS, vol. 8, 2002
Lnterpr|se App||cat|on Integrat|on (LAI) refers Lo Lhe plans, meLhods and Lools almed aL modernlzlng, consolldaLlng, lnLegraLlng and coordlnaLlng Lhe compuLer appllcaLlons wlLhln an enLerprlse.
What LAI g|ves us? ! lL serves as Lhe connecLlon/broker/LranslaLor llnklng auLonomously deslgned appllcaLlons lnLo a coheslve whole. ! lL embraces Lhe dlverslLy (heLerogenelLy) LhaL wlll always be parL of Lhe buslness ! lL conslders l1 an lnevlLable parL of large-scale sysLems deslgn and developmenL.
#+
lour LargeLs of LAl: 1. Data-|eve| |ntegrat|on - daLa-level lnLegraLlon LhaL ls noL only llmlLed Lo daLa and daLabases buL whlch also lncludes dlsLrlbuLed ob[ecLs (l.e. Lhe comblnaLlon of daLa, loglc and communlcaLlons wlLhln a slngle enLlLy). 8y means of a dlsLrlbuLed archlLecLure, ob[ecLs can be comblned Lo execuLe Lhe whole buslness funcLlons produclng an eleganL lnLegraLlon approach.
2. App||cat|on-|eve| |ntegrat|on - llnklng dlverse appllcaLlons ln order Lo accompllsh speclflc buslness processes. MosL popular ls Lhe messaglng-orlenLed mlddleware (MCM), where message brokers LransporL lnformaLlon beLween appllcaLlons by ldenLlfylng, Lransformlng and rouLlng messages Lo Lhe approprlaLe appllcaLlons on an evenL-drlven basls.
3. rocess-|eve| |ntegrat|on - lnLegraLlng Lhe flow of loglc among lnLegraLed appllcaLlons, 1ransacLlon rocesslng (1) as a Lool Lo effecL LransacLlon lnLegraLlon.
4. Inter-organ|zat|ona|-|eve| |ntegrat|on - process-level lnLegraLlon Lyplcally focuses on brlnglng LogeLher processes wlLhln an organlzaLlon Lo obLaln maxlmum value whlle supporLlng Lhe flow of lnformaLlon and loglc among Lhese processes. nexL sLep of Lhe lnLegraLlon ls Lo llnk processes beyond Lhe organlzaLlon Lo lnclude Lradlng parLners - boLh suppllers and cusLomers. 1he ulLlmaLe goal ls Lo blnd all Lradlng communlLy LogeLher.
LAI too|k|t: o Asynchronous event]message transport - MCM (message-orlenLed mlddleware) - enables asynchronous rouLlng of buslness evenLs beLween appllcaLlons. 1hey can also defer dellvery of lnformaLlon unLll appllcaLlon ls avallable. o 1ransformat|on eng|nes - Lools used Lo converLlng daLa and buslness evenLs from one formaL Lo anoLher. (L.g. from cusLomer's Lo ours formaL). #" o Integrat|on 8rokers - also known as message brokers, Lhey allow Lo rouLe and manlpulaLe buslness evenLs lnLelllgenLly beLween mulLlple appllcaLlons and daLa sLores. (L.g. recelve an order requesL from Web appllcaLlon and rouLe Lhese requesLs Lo one or more appllcaLlons, bearlng Lhe formaL expecLed by Lhe desLlnaLlon appllcaLlons) o 8us|ness process management frameworks - Lools LhaL enable buslness loglc Lo be separaLed from process flow loglc, Lwo dlsLlncL caLegorles: ptocess ootomotloo (framework LhaL allows mulLlple dlsparaLe sofLware componenLs Lo parLlclpaLe ln an lnLegraLed buslness process flow) and wotkflow (focuses on process sLeps performed by human lnLeracLlon wlLh Lhe sysLem).
Slx key sLraLegles for managlng LAl:
1. Create a Corporate Integrat|on Strategy 1) 1otqet sttoteqlc oppllcotloos - ldenLlfy Lhe appllcaLlons LhaL requlre lnLegraLlon and declde whlch of Lhose are sLraLeglc 2) 8ecome oo Al xpett - ldenLlfy Lhe mosL approprlaLe levels Lo focus Lhe lnLegraLlon efforL and LAl LoolklL needed 3) lJeotlfy tbe 5totos Ooo costs - conducL scenarlo plannlng Lo assess Lhe cosLs of Lhe sLaLus quo", whlch should represenL Lhe rough esLlmaLes of addlLlonal developmenL, malnLenance and Llme-Lo-markeL cosLs 4) 8ollJ tbe 8osloess cose fot Al - map Lhe advanLages of he planned LAl LoolklL 3) stlmote kesootces - LAl lnlLlaLlve needs people, Lools, new skllls, educaLlon and procedural and sLrucLural changes 6) cteote o lloo - ldenLlfy necessary Lasks Lo lnLegraLe sLraLeglc appllcaLlons and comblne Lhose lnLo a workable plan speclfylng Lhe loglcal order of Lhe work 7) 5ell tbe Al 5ttoteqy to Moooqemeot - LAl sLraLegy requlres senlor l1 managemenL's full endorsemenL
2. Assemb|e 1he LAI 1oo|k|t - ldenLlfy Lhe key polnLs of your LAl LoolklL based on your sLraLeglc appllcaLlons LhaL are Lo be lnLegraLed. useful checkllsL: & lotetool ooJ/ot extetool loteqtotloo - are you lnLegraLlng lnLernally or exLernally wlLh cusLomer or boLh? & 8osloess ptocess moooqemeot - LAl musL supporL Lhls feaLure Lo manage changes Lo buslness processes when Lhey happen & 5ecotlty - daLa encrypLlon eLc., especlally lmporLanL when deallng wlLh enLlLles ouLslde your own organlzaLlon & ose of use - especlally lmporLanL lf your buslness users wlll be malnLalnlng buslness ## process or workflow lnformaLlon & 1ecbooloqy Moooqemeot looctlooollty - Lhe LAl LoolklL musL Lell ln deLall whaL lL ls dolng wlLh Lhe lnfrasLrucLure and noLlfy when Lhere are problems
3. Dep|oy nub-And-Spoke Des|gn - Lhree organlzlng prlnclples wlLh Lhls deslgn: 1. uon'L connecL anyLhlng dlrecLly Lo anyLhlng, 2. ueslgn appllcaLlon Lo be auLonomous and don'L allow Lhem Lo share daLabases dlrecLly, 3. All knowledge of lnLerconnecLlons ls removed from Lhe source and placed wlLhln Lhe hub.
4. Create an Integrat|on Core Competency 1eam Lach hub should have lLs own Leam Lo Lake responslblllLy for: $ MalnLalnlng all lnLegraLlon documenLaLlon, $ erformlng (and assumlng ownershlp for) Lhe deLalled deslgn of each lnLerface $ ConsLrucLlng Lhe mlddle pleces of Lhe lnLerface $ LsLabllshlng besL pracLlces $ erformlng broker markeLplace evaluaLlons, and $ AdmlnlsLerlng all mlddleware sofLware producLs.
S. ke|ntegrate Legacy App||cat|ons
SLraLegles for relnLegraLlon: ! kefac|ng - replaclng sLandard emulaLlon screens wlLh graphlcal user lnLerface (Cul), lnLegraLlon ls faclllLaLed by Lhe use of common fronL-ends ! kepurpos|ng - llke refaclng, usage of Cul for Lhe presenLaLlon layer. 1he dlfference ls LhaL repurposlng allows for bulldlng ln new buslness loglc (as updaLlng) even Lhough Lhe legacy apps remaln unchanged ! kestructur|ng - separaLlng presenLaLlon and buslness loglc componenLs wlLhln an appllcaLlon. Cnce separaLed, Lhese componenLs are wrapped ln new lnLerfaces and lnLegraLed lnLo a number of new refaclng or repurposlng appllcaLlons. ! ke-eng|neer|ng - rebulldlng Lhe enLlre appllcaLlon ! 8eplaclng - replace appllcaLlon wlLh off-Lhe-shelf soluLlon #$
6. "1he Co||aborat|ve Lnterpr|se" - lnLroduclng Lhe collaboraLlon beLween Lhe Lradlng parLners and Lherefore creaLlng glanL Lhe collaboraLlve enLerprlses".
Tutori al 5
1urban, Sharda, De|en & k|ng, Chapter 2
Data warehouse - pool of daLa - Lype of daLabase, prlmarlly focused on declslon supporL and provldes lmproved analyLlcal capablllLles. Data warehous|ng - a dlsclpllne LhaL resulLs ln appllcaLlons LhaL provlde declslon supporL capablllLy, allows ready access Lo buslness lnformaLlon and creaLes buslness lnslghL.
1he daLa warehouse lnLegraLes varlous daLabases LhroughouL Lhe organlzaLlon lnLo a slngle, ln-house enLerprlse unlL Lo generaLe a slngle verslon of Lhe LruLh for Lhe company.
uaLa warehouse also allows maklng daLa avallable ln real Llme Lo Lhe declslon makers who need lL (keol 1lme uoto woteboosloq jkuw]).
uaLa are usually sLrucLured Lo be avallable ln a form ready for analyLlcal processlng acLlvlLles, such as: onllne analyLlcal processlng (CLA), daLa mlnlng, querylng, reporLlng).
CharacLerlsLlcs of daLa warehouslng (malnly abouL daLa access): a) Sub[ecL orlenLed - daLa organlzed by deLalled sub[ecL, conLalnlng only lnformaLlon relevanL for declslon supporL, b) lnLegraLed - placlng daLa ln conslsLenL formaL, c) 1lme varlanL - deLecLlng Lrends, devlaLlons and long-Lerm relaLlonshlps for forecasLlng and comparlsons, leadlng Lo declslon maklng, Llme as Lhe only lmporLanL dlmenslon LhaL daLa warehouse most supporL, d) nonvolaLlle - afLer enLered lnLo a daLa warehouse, users cannoL change Lhe daLa.
AddlLlonal characLerlsLlcs: l. Web based - Lyplcally deslgned Lo provlde an efflclenL compuLlng envlronmenL for Web-based apps ll. 8elaLlonal/mulLldlmenslonal - uses elLher relaLlonal sLrucLure or mulLldlmenslonal sLrucLure lll. CllenL/server - uses archlLecLure LhaL provldes easy access for end users lv. 8eal Llme - access Lo currenL daLa and analysls capablllLles v. MeLadaLa - conLalnlng daLa abouL daLa, abouL how daLa ls organlzed and how Lo efflclenLly use lL. 1hree Lypes of meLadaLa: 1. 5yotoctlc metoJoto (descrlblng composlLlon of daLa), 2. 5ttoctotol metoJoto (descrlblng sLrucLure), 3. 5emootlc metoJoto (descrlblng meanlng of daLa ln speclflc domaln). 1he prlmary purpose of meLadaLa should be Lo provlde conLexL Lo Lhe reporLed daLa, LhaL ls, lL should provlde enrlchlng lnformaLlon LhaL leads Lo Lhe creaLlon of knowledge. #%
Data mart - ls a subseL of daLa warehouse, usually focuslng on a parLlcular sub[ecL or deparLmenL. - uepeoJeot Joto mott - ls a subseL LhaL ls creaLed dlrecLly from Lhe daLa warehouse, has an advanLage of provldlng conslsLenL and quallLy daLa. - loJepeoJeot Joto mott - ls a small warehouse deslgned for a sLraLeglc buslness unlL or a deparLmenL, buL lLs source ls noL LuW (LnLerprlse uaLa Warehouse).
Cperat|ona| Data Stores (CDS) - Lype of daLabase used as a Lemporary sLaglng area for a daLa warehouse, sLores only very teceot lofotmotloo. lL ls used for shorL-Lerm declslons lnvolvlng mlsslon- crlLlcal apps, raLher Lhan long-Lerm declslon, whlch are more assoclaLed wlLh LuW (petmooeot lofotmotloo). *Cper marLs - creaLed from CuS, when operaLlonal daLa needs Lo be analyzed mulLldlmenslonally.
Ma[or componenLs of Lhe daLa warehouslng process: 1. uaLa sources - daLa are sources from mulLlple lndependenL operaLlonal sysLems and posslbly from exLernal provlders. uaLa may also come from onllne LransacLlon processlng (CL1), L8 sysLems or Web daLa. 2. uaLa exLracLlon and LransformaLlon - daLa are exLracLed and properly Lransformed uslng L1L (exLracL, Lransform, load). 3. uaLa loadlng - daLa are loaded lnLo sLaglng area, where Lhey are Lransformed and cleansed. 4. Comprehenslve daLabase - LuW supporLs all declslon analysls by provldlng relevanL summarlzed and deLalled lnformaLlon orlglnaLlng from many dlfferenL sources. 3. MeLadaLa malnLenance 6. Mlddleware Lools - enable access Lo Lhe daLa warehouse, e.g. by creaLlng SCL querles. 1here are many fronL-end appllcaLlons LhaL users can use Lo lnLeracL wlLh daLa, such as daLa mlnlng, CLA, reporLlng Lools and daLa vlsuallzaLlon Lools.
aper aLLempLs Lo presenL a framework for bulldlng a 8l sysLem.
8uslness lnLelllgence sysLems comblne operaLlonal and hlsLorlcal daLa wlLh analyLlcal Lools Lo presenL valuable and compeLlLlve lnformaLlon Lo buslness planners and declslon makers. Cb[ecLlve of 8l: ( 1o lmprove Lhe Llmellness and quallLy of lnformaLlon, ( 1o enable managers Lo be able Lo beLLer compare Lhelr poslLlon Lo Lhe compeLlLlon, ( 1o analyze changlng Lrends ln markeL share, ( 1o analyze changes ln cusLomer behavlor and spendlng paLLerns, cusLomer preferences, ( 1o analyze company capablllLles and markeL condlLlons.
#& 8l ls an area of ueclslon SupporL SysLem (uSS), whlch ls an lnformaLlon sysLem LhaL can be used Lo supporL complex declslon-maklng, and solvlng complex, seml-sLrucLured or lll- sLrucLured problems. 8l can be presenLed as an archlLecLure, Lool, Lechnology or sysLem LhaL gaLhers and sLores daLa, analyzes lL uslng analyLlcal Lools, faclllLaLes reporLlng, querylng and dellvers lnformaLlon or knowledge LhaL ulLlmaLely allows organlzaLlons Lo lmprove declslon maklng. 8l ls Lhe process LhaL Lransforms daLa lnLo lnformaLlon and Lhen lnLo knowledge.
8l Lools have evolved from belng LxecuLlve lnformaLlon sysLem (LlS) and ueclslon SupporL SysLem (uSS) Lo provlde much more ablllLy ln lnformaLlon dellvery and ablllLy Lo supporL Lechnlques such as query, reporLlng and ad hoc analysls and mulLldlmenslonal analysls, whlch are known as CLA.
Coals ln order Lo achleve daLa conslsLency: " 1|me||ness - daLa wlLhln Lhe sysLem should be synchronlzed wlLh all oLher appllcaLlons " Accuracy - Lhe daLa should encompass every daLa from any oLher appllcaLlon " Acceptance - Lhe users convlnced of Llmellness and accuracy of daLa should be able Lo acLlvely use Lhe sysLem as a supporL for declslon-maklng 1he concepL of 8l can be decomposed lnLo Lhree parLs: l) uaLa CapLure/AcqulslLlon wlLh Lhe L1L process ll) uaLa SLorage lll) uaLa Access and Analysls
Data Capture]Acqu|s|t|on w|th the L1L process 8efore daLa ls loaded from operaLlonal daLabase and exLernal sources lnLo Lhe daLa warehouse, lL needs Lo processed: - LxLracLlon and Cleanse - selecLed daLa ls consolldaLed and fllLered ouL from varlous forms of polluLlon. uaLa Cleanse valldaLes and cleans up Lhe exLracLed daLa Lo correcL #' lnconslsLenL, mlsslng or lnvalld values. - 1ransform - daLa ls lnLegraLed lnLo sLandard formaLs and applles buslness rules LhaL map daLa Lo Lhe warehouse schema. - Load - cleansed daLa ls loaded lnLo Lhe daLa warehouse.
Data Storage - uoto woteboose - daLa warehouse ls a copy of LransacLlon daLa speclflcally sLrucLured for query and analysls and ls lnformaLlonal, analysls and declslon supporL orlenLed, noL operaLlonal or LransacLlon processlng orlenLed. lLs speclflc purpose ls Lo supporL buslness declslons, noL buslness operaLlons. - uoto Motts - small slzed daLa warehouses, Lyplcally creaLed by lndlvldual deparLmenLs or dlvlslons Lo faclllLaLe Lhelr own declslon supporL acLlvlLles. - MetoJoto - daLa abouL daLa LhaL allows users Lo saLlsfy Lhelr lnformaLlon need abouL Lhe daLa warehouslng sysLem and lLs conLenL.
Data Access and Ana|ys|s - Access Lools (referred Lo as fronL-end), conslsLs of Lools and Lechnlques LhaL provlde a buslness user wlLh dlrecL, lnLeracLlve access Lo daLa, whlle hldlng Lhe Lechnlcal complexlLy of daLa reLrleval. - Some of Lhe Lools: o CLA - provlde users wlLh Lhe means Lo explore and analyze large amounLs of daLa, lnvolvlng complex compuLaLlons, Lhelr relaLlonshlps and vlsually presenL resulLs ln dlfferenL prospecLlve. 1yplcal appllcaLlons performed on mulLldlmenslonal daLa: l) koll-op - daLa ls summarlzed wlLh lncreaslng generallzaLlon ll) utlll-Jowo - lncreaslng levels of deLall are revealed urlll up & down:
#( lll) 5llce ooJ Jlce - performlng pro[ecLlon operaLlons on Lhe dlmenslons Sllce:
ulce:
lv) llvotloq - cross LabulaLlon lvoLlng:
o Data m|n|ng - uslng a varleLy of Lechnlques Lo ldenLlfy nuggeLs of lnformaLlon or declslon maklng knowledge ln bodles of daLa and exLracLlng Lhese ln such a way LhaL Lhey can be puL Lo use ln Lhe areas such as declslon supporL, predlcLlon, forecasLlng and esLlmaLlon, |t |s the search for re|at|onsh|ps and global paLLerns LhaL exlsL ln large daLabases buL are hldden among Lhe vasL amounL of daLa. #) o Mach|ne |earn|ng (ML) - parL of emerglng ArLlflclal lnLelllgence. ML ls a famlly of meLhods LhaL aLLempL Lo allow machlnes Lo acqulre knowledge for problem solvlng by showlng Lhem hlsLorlcal cases.
SharlaL & nwakanna, Lnterpr|se kesource |ann|ng And Its Iuture ke|at|onsh|p 1o Dec|s|on Support System, !ournal of 8uslness & Lconomlcs 8esearch, volume 4, number 12, 2006
aper looks aL Lhe developmenL of L8 and uSS, wlLh a focus on Lhe dlfferences beLween Lhe Lwo sysLems ln Lerms of Lhelr evoluLlon and appllcaLlon and Lhe poLenLlal for convergence ln Lhe fuLure. ) L8 and uSS have evolved ln parallel and as a resulL Lhose organlzaLlons, whlch have already lmplemenLed L8 are now havlng problems lnLegraLlng uSS and daLa warehouslng lnLo Lhelr sysLems. ) L8 vendors say lL's comparaLlvely slmple Lo as on uSS appllcaLlon, buL Lhls ls noL Lrue. MosL L8 speclallsLs are noL experlenced wlLh uSS.
Dec|s|on Support System (DSS) ls a compuLer based lnformaLlon sysLem deslgned Lo supporL all phases of seml-sLrucLured and unsLrucLured declslon maklng. lL ls composed of corporaLe daLa (daLa warehouse, a seL of analyLlcal models and Lools, a knowledge base and an easy-Lo-use user lnLerface.
Lk ls an lndusLry Lerm for lnLegraLed, mulLl-module appllcaLlon sofLware packages LhaL are deslgned Lo serve and supporL mulLlple buslness funcLlons".
AlLhough L8 and uSS have evolved separaLely, Lhere ls an lncreaslng Lrend Lowards lnLegraLlng Lhe Lwo. 1hls lncreaslng lnLegraLlon of Lwo sysLems ls poLenLlally valuable ln a number of ways. CrganlzaLlons LhaL succeed Lo do so are able Lo lncrease Lhelr lnLelllgence denslLy and lmprove boLh quallLy and Lhe avallablllLy of declslon supporL lnformaLlon.
L8 has lLs orlglns ln operaLlonal 1ransacLlon rocesslng SysLems, whereas uSS ls concerned wlLh declslon supporL and sLraLeglc plannlng.
1hree soluLlons Lo our problem: 1. Lk vendors fac|||tate |ncorporat|ng |n the|r Lk data warehouse the data from app||cat|ons that are outs|de of the Lk. 8uL, Lhe L8 conLrol over Lhe daLa warehouse ls conLrary Lo Lhe baslc ob[ecLlve of uSS. lf vendors creaLe and conLrol Lhe daLa warehouse for an lndusLry, Lhere wlll be no compeLlLlve advanLage, because every organlzaLlon ln Lhe same lndusLry wlll have Lhe same daLa warehouse and Lhe same uSS. 2. 1wo |ndependent data warehouses wou|d be bu||t: one for non-Lk app||cat|ons and one for Lk app||cat|on. 1hls approach ls parLlcularly advanLageous for organlzaLlons LhaL already have an efflclenL and esLabllshed uSS. AdvanLage: allows free access Lo Lhe daLa, buL does noL allow Lo vlew an organlzaLlonal daLa as an lnLegraLed whole. ulsadvanLage: ln pracLlce lL #* ls noL posslble Lo comblne Lhe daLa from one daLa warehouse Lo Lhe daLa ln anoLher daLa warehouse unless a consolldaLed vlew of Lhe daLa ls creaLed aL a hlgher level as enLerprlse daLa warehouse. 3. A standard organ|zat|ona| data warehouse |s bu||t to reference the data |n both Lk and non-Lk app||cat|on and there |s no Lk data warehouse. AdvanLages: allows Lo lnLegraLe all areas, glves greaLer freedom of access and cholces wlLh regards Lo Lhe daLa warehouslng Lools and Lechnology and does noL Lle organlzaLlon Lo any parLlcular L8 appllcaLlon.
Any problems? ) 1here ls sLrong dlsagreemenL beLween Lhose wlLh Lechnlcal experLlse and Lhose who are solely lnLeresLed ln Lhe buslness aspecL of L8 and uSS sysLems. ) 1he ma[or dlsLlncLlon beLween uSS and L8 ls Lhe facL LhaL Lhey are deslgned Lo perform Lwo dlfferenL Lasks aLe Lwo dlfferenL level of Lhe organlzaLlon (sLraLeglc vs. operaLlon).
Tutori al 6
!uklc, Mode|||ng strateg|es and a|ternat|ves for data warehous|ng pro[ects, CCMMunlCA1lCnS Cl 1PL ACM, vol. 49, no. 4, 2006
1here ls no consensus among researchers and pracLlLloners abouL Lhe mosL approprlaLe daLa modellng sLraLegles for daLa warehouslng pro[ecLs.
1wo maln reasons for creaLlng a daLa warehouse as a separaLe analyLlcal daLa sLore: 1. 1he performance of operaLlonal querles can be severely dlmlnlshed lf Lhey musL compeLe for compuLlng resources wlLh analyLlcal querles. 2. lL ls ofLen lmposslble Lo sLrucLure a daLabase LhaL can be used ln a sLralghLforward manner for boLh operaLlonal and analyLlcal purpose.
Data mart - ls a daLa sLore based on Lhe same prlnclples as a daLa warehouse, buL wlLh more llmlLed scope. uaLa marL ls usually smaller and focuses on a parLlcular deparLmenL or sub[ecL. ulmenslonal modellng as a prlnclple daLa marL modellng Lechnlque. lL uses Lwo Lypes of Lables: facLs & dlmenslons. ulmenslonally modeled daLa marL: $+
Data warehouse mode||ng opt|ons: - Inmon's opt|on (Lk-mode|ed data warehouse) - envlslons a daLa warehouse as an lnLegraLed daLabase modeled by uslng Lhe LradlLlonal daLabase modellng Lechnlque (L8 modellng). AfLer a daLa warehouse ls creaLed, lL serves as a source of daLa for dlmenslonally modeled daLa marLs and for any oLher analyLlcally useful daLa seLs. 1he ldea ls Lo have physlcally sLored cenLral daLa warehouse modeled as an LnLlLy-8elaLlonshlp model. All lnLegraLlon of Lhe daLa underlylng operaLlonal daLa sources occurs wlLhln a cenLral daLa warehouse L8 model.
- k|mba||'s opt|on (D|mens|ona||y mode|ed data warehouse) - envlslons Lhe daLa warehouse as a collecLlon of dlmenslonally modeled daLa marLs. 1he approach ls analogous Lo Lhe prevlous approach when lL comes Lo Lhe uLlllzaLlon of operaLlonal daLa sources and Lhe L1L process. 1he dlfference ls Lhe modellng Lechnlque used for modellng Lhe daLa warehouse. ln Lhls approach Lhe seL of commonly used dlmenslons ls deslgned flrsL. lacL Lables correspondlng Lo Lhe sub[ecLs of analysls are Lhen added. A seL of dlmenslonal models ls creaLed where each facL Lable ls connecLed Lo mulLlple dlmenslons and some of Lhe dlmenslons are shared by more Lhan one facL Lable. ln addlLlon Lo Lhe orlglnally creaLed seL of conformed dlmenslons, addlLlonal dlmenslons are lncluded as needed. 1he resulL ls a daLa warehouse LhaL ls a collecLlon of lnLerLwlned dlmenslonally modeled daLa marLs.
Inmon's Lk mode|ed data warehouse k|mba||'s d|mens|ona||y mode|ed data warehouse $" 8equlremenL Lo creaLe L8 model as Lhe flrsL sLep ulmenslonally modeled sLrucLures are creaLed wlLhouL creaLlng an underlylng L8 model for Lhem lf ln Lhe fuLure oLher Lypes of analyLlcal daLa sLores wlll be needed ln addlLlon Lo dlmenslonal sLrucLures Lhen here we have more power lf dlmenslonal sLrucLures are everyLhlng a company wlll ever requlre Lhen lL's perfecL - slmpler, qulcker, less cosLly uaLa modellng phase requlres hlgh level of experLlse and conslderable upfronL commlLmenL lL lacks enLerplsewlde focus and concenLraLes prlmarlly on Lhe lndlvldual buslness unlLs or groups of users CreaLed model ls usable ln a number of dlfferenL ways
- Creat|on of |ndependent Data Marts - Lhere ls consensus abouL lnapproprlaLeness of uslng Lhls soluLlon as a sLraLegy for deslgnlng a daLa warehouse. 1here ls an unnecessary repeLlLlon of Lhe L1L efforL and Lhe lnablllLy for cross-deparLmenL analysls and communlcaLlon. SLlll, Lhls approach ls presenL ln companles very ofLen. Why? ( Lack of lnlLlal enLerprlsewlde focus when daLa analysls ls concerned ( no cross-deparLmenL collaboraLlon ( 8udgeLlng sLrucLure as an obsLacle Lo common soluLlon ( SomeLlmes Lhere ls a cholce of creaLlng daLa marLs or dolng noLhlng.
Chaudhury, uayal, narasayya, An Cverv|ew of 8us|ness Inte|||gence 1echno|ogy, CCMMunlCA1lCnS Cl 1PL ACM, vol. 34, no. 8, 2011
1he cosL of daLa acqulslLlon and daLa sLorage has decllned slgnlflcanLly. 1hls has lncreased Lhe appeLlLe of buslness Lo acqulre very large volumes ln order Lo exLracL as much compeLlLlve advanLage from lL as posslble.
1yplcal archlLecLure for supporLlng 8l wlLhln an enLerprlse:
Data sources: Lhe daLa comlng Lo 8l ofLen comes from dlfferenL sources (mulLlple operaLlonal daLabases from mulLlple operaLlonal daLabases from dlfferenL deparLmenLs buL also from exLernal sources)
$# Data movement: back-end Lechnologles for preparlng Lhe daLa for 8l are referred Lo as xttoct- 1toosfotm-looJ (1l) tools. ln order Lo supporL 8l Lasks ln Lhe oeot teol tlme Complex LvenL rocesslng Lnglne ls used.
Data warehouse: daLa for 8l ls Lyplcally loaded lnLo Lhe Joto woteboose. 8elaLlonal uaLabase ManagemenL SysLem (8u8MS) ls a popular cholce here. 1o Lackle Lhe 8lg uaLa problem, whlch cannoL be handled by 8u8MS, englnes based on MopkeJoce paradlgm are belng LargeLed. MopkeJoce - ls a programmlng model and an assoclaLed lmplemenLaLlon for processlng and generaLlng large daLa seLs wlLh a parallel, dlsLrlbuLed algorlLhm on a clusLer. A Map8educe program ls composed of a Map() procedure LhaL performs fllLerlng and sorLlng (such as sorLlng sLudenLs by flrsL name lnLo queues, one queue for each name) and a keduce() procedure LhaL performs a summary operaLlon (such as counLlng Lhe number of sLudenLs ln each queue, yleldlng name frequencles).
M|d-1|er servers: complemenLary severs Lo daLa warehouse servers LhaL provlde speclallzed funcLlonallLy for dlfferenL 8l scenarlos. CLA servers (onllne analyLlc processlng) efflclenLly expose Lhe moltlJlmeosloool vlew of daLa Lo appllcaLlons/users and enable Lhem for common 8l operaLlons, such as: llltetloq, Aqqteqotloo, utlll-Jowo, llvotloq. 8eporLlng servers enable deflnlLlon, efflclenL execuLlon and renderlng of reporLs LnLerprlse search englnes supporL Lhe keyword search paradlgm over LexL and sLrucLured daLa ln Lhe warehouse. uaLa mlnlng englnes enable ln-depLh analysls of daLa LhaL goes well beyond whaL ls offered by CLA and provldes Lhe ablllLy Lo bulld predlcLlve models Lo help answer Lhe quesLlons abouL fuLure predlcLlons 1exL analyLlc englnes can analyze large amounLs of LexL daLa and exLracL valuable lnformaLlon LhaL would oLherwlse requlre slgnlflcanL manual efforL
Iront-end app||cat|ons: appllcaLlons Lhrough whlch users perform 8l Lasks, such as spreadsheeLs, enLerprlse porLals, dashboards eLc.
web-ooolytlcs enables undersLandlng how vlslLors Lo a company's Web slLe lnLeracL wlLh Lhe pages.
Data storage: $$ Access structures - declslon supporL querles requlre operaLlons such as fllLerlng, [oln and aggregaLlon. Speclal daLa sLrucLures Lo supporL Lhese operaLlons: o loJex sttoctotes - enables assoclaLlve access based on Lhe values of a parLlcular column. lndex scans and lndex lnLersecLlons (mulLlple query fllLerlng condlLlon handllng) can slgnlflcanLly reduce and ln some cases ellmlnaLe Lhe need Lo access Lhe base Lables. o MotetlollzeJ vlews - maLerlallzlng summary daLa, much less general Lhan lndex sLrucLure o lottltlooloq - dlvldlng Lables and lndexes lnLo smaller, more manageable unlLs o colomo-otleoteJ stotoqe - all values of a parLlcular column are sLored alongslde. 1wo maln advanLages: greaLer daLa compresslon and only columns accessed ln Lhe query need Lo be scanned. (8ead-only envlronmenL)
Data Compress|on - compresslon can reduce Lhe amounL of daLa LhaL needs Lo be scanned and lL can lower sLorage and backup cosLs.
uery process|ng: 8l ls capable of handllng Lhe querles on Lhe mulLldlmenslonal level, l.e. lnvolvlng a seL of numerlc measures LhaL are Lhe ob[ecLs of analysls. 1he aLLrlbuLes of a dlmenslon may be relaLed vla a bletotcby of relaLlonshlp.
CLA Servers - Cnllne AnalyLlc rocesslng, supporLs operaLlons such as fllLerlng, aggregaLlon, plvoLlng, rollup and drlll-down on a mulLldlmenslonal vlew of Lhe daLa.
CLA servers are lmplemenLed uslng elLher ( MulLldlmenslonal sLorage englne (MCLA) - server dlrecLly supporL Lhe mulLldlmenslonal vlew of daLa Lhrough a sLorage englne, fasL query response Llme, excellenL lndexlng properLles buL relaLlvely poor sLorage uLlllzaLlon, ( A relaLlonal u8MS englne (8CLA) - Lhe mulLldlmenslonal model and lLs operaLlons have Lo be mapped lnLo relaLlons and SCL querles. MosL 8CLA use o Star schema Lo represenL Lhe mulLldlmenslonal daLa model. 1he daLabase conslsLs of a slngle facL Lable and a slngle Lable for each dlmenslon. Lach row of Lhe facL Lable conslsLs of a polnLer Lo each of Lhe dlmenslons LhaL provlde lLs mulLldlmenslonal coordlnaLes and sLores Lhe numerlc measures for Lhose coordlnaLes. Lach dlmenslon $% Lable conslsLs of columns LhaL correspond Lo aLLrlbuLes of Lhe dlmenslon. SLar schemas do noL expllclLly provlde supporL for aLLrlbuLe hlerarchles. o Snowf|ake schema - reflnemenL of Lhe sLar schema where Lhe dlmenslonal hlerarchy ls expllclLly represenLed by normallzlng Lhe dlmenslon Lables.
( Cr a hybrld comblnaLlon (PCLA) - comblnlng 8CLA and MCLA by spllLLlng sLorage of Lhe daLa ln MCLA and relaLlonal sLore. (
Arlyachandra 1 and WaLson P , Wh|ch Data Warehouse Arch|tecture Is Most Successfu|?, 8uSlnLSS ln1LLLlCLnCL !Cu8nAL , vCL. 11, nC. 1, 2006
Cngolng dlscusslon abouL Lhe besL daLa warehouse archlLecLure. 1wo glanLs of Lhe daLa-warehouslng fleld: " Inmon - who advocaLes Lhe hub-and- spoke archlLecLure " k|mba|| - who advocaLes Lhe daLa marL bus archlLecLure wlLh conformed dlmenslons
lour measures used Lo assess Lhe success of Lhe archlLecLures: 1) lnformaLlon quallLy, 2) sysLem quallLy, 3) lndlvldual lmpacLs and 4) organlzaLlonal lmpacL.
lndependenL daLa marLs archlLecLure scored Lhe lowesL on all measures, noL surprlslngly as lL conflrms Lhe convenLlonal wlsdom LhaL lndependenL daLa marLs are a poor archlLecLural soluLlon. llndlngs suggesL also LhaL Lhe federaLed archlLecLure ls noL an opLlmal long-Lerm soluLlon.
1he dlfferences ln resulLs of hub-and-spoke, bus and cenLrallzed archlLecLures are sufflclenLly small LhaL no clalms can be made for a parLlcular archlLecLure's superlorlLy over Lhe oLhers. no slngle archlLecLure ls domlnanL, each of Lhe archlLecLure ls equally successful for Lhelr lnLended purposes. no surprlse abouL Lhe ouLcome, also because over Llme archlLecLures became slmllar Lo each oLher.
AuLhors found LhaL Lhe hub-and-spoke archlLecLure ls Lyplcally used wlLh more enLerprlsewlde lmplemenLaLlons and larger warehouses. SLlll, lL's Lhe mosL expenslve and Llme-consumlng soluLlon.
Tutori al 7
WlnLer, 8lschoff & WorLmann, kevo|ut|on or Lvo|ut|on? kef|ect|ons on In-Memory App||ances from an Lnterpr|se Informat|on Log|st|cs erspect|ve, ln: Lehner & lller (2011), pp. 23-34.
1he paper quesLlons wheLher lMu8MS ls a revoluLlon or evoluLlon, Lhe flndlngs lndlcaLe LhaL Lhere are slLuaLlons where ln-memory appllances are a useful exLenslon Lo exlsLlng l1 supporL concepLs, whlle oLher slLuaLlons do noL requlre such supporL. AuLhors regard ln-memory appllances as an evoluLlon, bot oot o tevolotloo of l1 supporL from an enLerprlse lnformaLlon loglsLlcs perspecLlve.
otetptlse lofotmotloo loqlstlcs (ll) - slmllar as 8uslness lnLelllgence, buL 8l focuses on Lhe 'local' use of provlsloned daLa, whlle LlL has an enLerprlse-wlde daLa provlslonlng perspecLlve. LlL lS ls based on Lhe daLa-warehouslng lnfrasLrucLure LhaL collecLs, sLores and lnLegraLed relevanL daLa LhaL ls furLher modlfled and enhances wlLhln speclflc reporLlng processes. LlL archlLecLure usually ls comprlsed of flve layers: $ 1. Source sysLems, 2.daLa LransformaLlon (L1L), 3. uaLa warehouse, 4. uaLa marLs, 3. 8eporLlng sysLems.
ConcepLual LlL archlLecLure poses Lhree key challenges: $ Speed - rlslng daLa volumes, lncreaslng needs for daLa lnLegraLlon and ever more dlverse daLa analysls posslblllLles $ lnLegraLlon - speclallzed lnfrasLrucLures wlLh hlgh degree of complexlLy rlses lssue of hlgh operaLlng and developmenL cosLs $ llexlblllLy - need Lo analyze any deslred buslness ob[ecL from any perspecLlve
ln-memory appllances - fundamenLals and poLenLlals: % lnLroduclng new concepLs llke masslve parallel processlng LhaL lmproves processlng speed even furLher $' % 1he exponenLlal prlce drop of processlng power and maln memory fosLers break-Lhrough lnnovaLlons % Columnar daLabases whlch are seL orlenLed raLher Lhan record orlenLed remlses Lo address Lhree LlL challenges: $ Speed - daLa updaLed and propagaLed lncremenLally and ln real-Llme lnLo Lhe analyLlcal envlronmenL. $ Integrat|on - LransacLlonal and declslon-relaLed daLa ls managed ln an lnLegraLlve manner. $ I|ex|b|||ty - analysls paLhs are noL llmlLed by pre-fabrlcaLed aggregaLes, so LhaL daLa can be lnLegraLed and analyzed from any perspecLlve.
ln conLrasL Lo Lhe old approach (lnmon hub- and-spoke approach) deplcLed, ln Lhe new landscape Lhere ls no dupllcaLlon [ln- memory daLabase lnsLead of daLa marLs ooJ L1L layer]. 1hls also lncreases Lhe speed as you don'L sLore your daLa ln marLs and you have blgger posslblllLles for daLa.
lnLroducLlon of ln-memory appllances ls reasonable under cerLaln characLerlsLlcs of your buslness: Low/moderaLe daLa volume/no need for lnLegraLlon - Lhls paLLern ls abouL operaLlonal analyLlcs. 1he LransacLlonal sysLem aL hand has no performance lssues. 1here ls no slgnlflcanL need for an ln-memory appllance. Plgh daLa volume/no need for lnLegraLlon - also abouL operaLlonal analyLlcs, hlgh amounLs of daLa can be handled by lnLroduclng an ln-memory appllance. Low/moderaLe daLa volume/need for lnLegraLlon - Lhe need for lnLegraLlon drlves Lhe appllcaLlon of 8l/uWP. 1here ls no slgnlflcanL value added by an ln-memory appllance. Plgh daLa volume/no need for lnLegraLlon - also abouL 8l/uWP. Plgh amounL of daLa can be handled by lnLroduclng an ln-memory appllance.
ln-memory should be regarded noL Lo be dlsrupLlve or as Lhe end" of exlsLlng LlL archlLecLures and approaches, buL raLher as an evoluLlon LhaL creaLed slgnlflcanL progress under cerLaln clrcumsLances.
$(
lller & Pagedorn, 8us|ness 8enef|ts and App||cat|on Capab|||t|es Lnab|ed by In-Memory Data Management, ln: Lehner & lller (2011), pp. 43-36.
aper presenLs posslble appllcaLlon use of ln-memory daLa managemenL, skeLch of how one can ldenLlfy buslness domalns and processes where lMuM could add slgnlflcanL value. WhaL klnd of analyses and declslon processes flL besL Lo Lhe capablllLles of ln-memory Lechnology?
CapablllLles of ln-memory daLa managemenL: ! Short response t|me - hlgh daLa volumes for analyses, slmulaLlons or plannlng runs can be processed very qulckly due Lo shorL access Lo read Llmes and well as hlgh compuLaLlon raLes. ! Un|f|ed transact|ona| and ana|yt|c data process|ng - Lhe separaLlon of LransacLlonal and analyLlc daLa processlng ln LradlLlonal appllcaLlon archlLecLures can be overcome wlLh lMuM. Cruclal for Lhese lmprovemenLs ls Lhe huge, sLlll sLrongly lncreaslng capaclLy of maln memory (2 LerabyLe used ln pracLlce now). ! Ana|ys|s of ||ne |tem data - aggregaLlon for Lhe acceleraLlon of response Llmes ls no longer necessary. Analyses and plannlng runs can be based on Lhe orlglnal daLa records. CurrenL resLrlcLlons due Lo pre-deflned aggregaLes are obsoleLe.
(wbot ls o lloe ltem? A llne lLem ls a group of daLa and lLs sLrucLure LhaL ls or can be repeaLed mulLlple Llmes ln lLs documenL. lL holds Lhe daLa LhaL ls Lhe focus of Lhe documenL, and a parL of a documenL or LransacLlon LhaL conLalns dlsLlncL parL LhaL can sLand alone. lor lnsLance, an order LhaL has Lwo llne lLems could be spllL lnLo Lwo orders LhaL would be able Lo be fllled separaLely. lL does noL necessarlly conLaln lnformaLlon LhaL can be applled Lo a group of lLems, alLhough someLlmes Lhls lnformaLlon ls lncluded. 1hls feeds lnLo whaL Lype of daLa LhaL a llne lLem conLalns. wbot type of Joto Joes o lloe ltem cootolo? ln a purchase order, or oLher commerce relaLed documenL Lhese pleces of daLa wlll be found ln Lhe llne lLem: urpose of llne, roducL or Servlce lu, CuanLlLy, unlL of Measure, rlce or lee, 1axes, Speclal Pandllng)
1yplcol Appllcotloo Ateos fot lMuM. ctltetlo fot selectloo of bosloess ptocesses fot lo-memoty tecbooloqy ose. ( Data dynam|cs - Pow ofLen and predlcLably do daLa change? ( kange of var|at|on - 1o whaL exLend do key flgures Lyplcally change and how severe ls Lhe correspondlng lnfluence on corporaLe success? ( Number of ana|ys|s opt|ons - Pow many alLernaLlves or worklng hypoLheses shall ldeally be compared? ( Urgency of ana|ys|s resu|ts - 1o whaL exLend ls lL urgenL Lo obLaln analysls resulLs, e.g. Lo meeL glven deadllnes, carry ouL subsequenL and dependenL process sLeps or Lo explolL resulLs whlle sLlll meeLlng wlLh speclallsLs and managers? ( Comp|ex|ty of ana|ys|s - ls an analysls comparaLlvely complex due Lo underlylng algorlLhms or daLa sLrucLures? ( Data vo|ume - are huge daLa volumes processed ln Lhe consldered buslness acLlvlLles? $)
ApparenLly, Lhe use of lMuM ls promlslng ln areas wlLh a h|gh |eve| of data dynam|cs and var|at|on, many ana|ys|s opt|ons, urgent|y needed resu|ts, h|gh comp|ex|ty and huge data vo|umes.
lo ptloclple ooe coo ocbleve tbe followloq effects by osloq lMuM. ' Mass|ve |ncrease |n ana|ys|s frequency Lhrough a drasLlc reducLlon of response Llmes ' n|gher ana|ys|s f|ex|b|||ty as predeflned reporLlng hlerarchles become obsoleLe and a mulLlLude of opLlons can be explored ' Lower data |atency, up Lo and lncludlng evaluaLlons ln real-Llme ' Lnhancement of data bandw|dth, e.g. hlsLorlc and currenL daLa can be analyzed LogeLher ' n|gher |eve| of deta|| Lhrough access Lo slngle documenLs and llne lLem daLa
8eal-llfe examples of lMu8 appllcaLlons: 1. MonlLorlng of sales and lnvenLory daLa ln reLall lndusLry Lo avold ouL-of-sLock slLuaLlons 2. CuallLy monlLorlng - durlng producLlon, large amounL of maLerlal daLa and equlpmenL parameLers are capLured conLlnuously 3. ManagemenL of complalns - complalns abouL deflclencles are puL lnLo a sLaLlsLlc and hlsLorlc conLexL. (cusLomer segmenLaLlon, based on Lhe complaln once recelved, you can accommodaLe Lhls cusLomer)
LxploraLory analysls posslble: a) rof|tab|||ty ana|ys|s - lMuM allows for lnvesLlgaLlon of conLrlbuLlon marglns wlLhouL any predeflned level of aggregaLlon (noL resLrlcLlng yourself Lo predeflned producL or markeL) b) Customer segmentat|on - huge amounLs of cusLomer and markeL daLa have Lo be explored wlLh respecL Lo flexlblllLy chosen comblnaLlons of characLerlsLlc values.
Complex analysls: a) Ava||ab|e-to-rom|se - provldes avallable quanLlLles of requesLed produces and correspondlng dellvery daLes b) "I|ex|b|e pr|c|ng |n sa|es" - ln negoLlaLlons wlLh cusLomer dlfferenL opLlons for prlces and dlscounLs can be lnvesLlgaLed dlrecLly
lasL consolldaLlon: a) Up-to-date sa|es f|gures - provldlng consolldaLed resulLs wlLh mosL currenL daLa as fasL as posslble
AdapLlve plannlng: a) roduct|on p|ann|ng - plannlng runs can be compleLed wlLh lMuM ln shorL Llme lnLervals wlLh always up-Lo-daLe sLarLlng parameLers
Analysls of uaLa from Consumer uevlces: a) ln-memory Lechnology enables processlng of exLreme volumes of hlghly dynamlc consumer daLa $* from embedded sysLems, as Lhey appear e.g. ln domesLlc homes. (e.g. elecLrlclLy consumpLlon)
vom 8rocke, !an, ueborLoll, SLefan, Muller, Cllver, and 8euLer, nadlne (2014) "now In- memory 1echno|ogy Can Create 8us|ness Va|ue: Ins|ghts from the n||t| Case," CommunlcaLlons of Lhe AssoclaLlon for lnformaLlon SysLems: vol. 34, ArLlcle 7.
WlLh |n-memory techno|ogy all daLa and appllcaLlons are kepL ln Lhe compuLer's maln memory Lo avold expenslve mechanlcal hard-drlve, reduce laLency Llmes and lncrease Lhe ablllLy Lo process large volumes of daLa or complex daLa. 1hls approach resulLs ln a slgnlflcanL lncrease of lnformaLlon processlng capaclLy.
( 1here's a shorLage of Lhe klnd of experLlse LhaL ls requlred Lo reallze buslness value from ln- memory Lechnology (purpose of Lhls arLlcle).
1echno|ogy push vs. techno|ogy pu|| $ 1echnology push - Lechnologlcal developmenLs are broughL abouL by sclence and new Lechnologlcal advances are key drlvers of lnnovaLlon. $ 1echnology pull - cusLomer need ls Lhe orlgln of Lechnologlcal developmenLs.
ln-memory Lechnology ls drlven by boLh Lechnology push and need pull. 1he need pull ls explalned by Lhe emergence of 8lg uaLa" and Lhe deslre Lo collecL lnLelllgence from lL ln order Lo galn compeLlLlve advanLage. New tecbooloqles ote oeeJeJ to collect vost omooot of complex Joto.
.&/**0 - one of Lhe mosL promlnenL plaLforms for 8lg uaLa processlng, whlch lmplemenLs Lhe fundamenLal programmlng model MopkeJoce. Map8educe ls a baLch query processor LhaL ls speclflcally deslgned for analyzlng whole daLa seLs ln ad hoc fashlon, buL noL for LransacLlon sysLems and real-Llme processlng. ln conLrasL, ln-memory Lechnology addresses boLh of Lhese aspecLs of daLa analysls ln addlLlon Lo provldlng hlgh-performance analyLlcal capablllLles.
In-memory techno|ogy and |ts f|ve pr|mary character|st|cs: 1. Data |s stored ent|re|y |n the computer's ma|n memory |nstead of on d|sk-based hard dr|ves. 1haL enables dramaLlc decrease ln Lhe Llme requlred Lo access lL. 2. Mu|t|p|e CUs can process para||e| requests, Lhereby uslng Lhe avallable compuLlng resources fully. Many operaLlons can be execuLed concurrenLly. 3. M|xed row- and co|umn-or|ented storage |s app||ed. lnsLead of row-based-only approach LhaL ls lmplemenLed ln LradlLlonal relaLlonal daLabase managemenL sysLem, whlch allows %+ slngle Luples Lo be read qulckly buL lL's noL well sulLed Lo readlng a seL of resulLs from a slngle column. Column-orlenLed daLabases on Lhe oLher hand are well sulLed Lo calculaLlons LhaL are execuLed on a slngle or only few columns. 1hls lmproves daLa compresslon and allows for masslve parallel processlng and efflclenL memory access. 4. IMD8MS offer compress|on techn|ques ||ke d|ct|onary encod|ng and run-|ength-encod|ng that s|gn|f|cant|y reduce data s|ze. S. IMD8MS |mp|ement an |nsert-on|y approach, so a database does not a||ow app||cat|ons to perform updates or de|et|ons on phys|ca||y stored tup|es of data. uaLa changes are recorded wlLhln Lhe same daLabase Lable, allowlng more efflclenL querles of hlsLorlcal daLa.
1he ob[ecLlve of hlgh-performance compuLlng ls Lo mlnlmlze Lhe average response Llme of a Lask ln a real Llme", so Jotloq glven evenL.
*AppllcaLlons of ln-memory capablllLles: 1) ConLlnuous capLurlng of Lhe cusLomer daLa ln order Lo geL Lhe beLLer plcLure, gaLherlng Lhe daLa ln order Lo ensure fasL reacLlon Lo lncldenLs (e.g. elevaLors' producer). 2) Crawl Lhe lnLerneL daLa Lo look for all lnformaLlon assoclaLed wlLh your company. 3) LnLerprlse search" - Lhe lostoot koowleJqe lloJet LhaL can analyze who has creaLed, updaLed, downloaded and forwarded a documenL and dlsplay Lhese connecLlons ln order Lo connecL lndlvlduals wlLh slmllar experLlse or lnLeresLs. 4) 1alk Lo me" - dlsplaylng daLa by verbal commands
llndlngs of Lhe paper: l. 1be tecbolcol cbotoctetlstlcs of lo-memoty tecbooloqy ptovlJe fltst-otJet ooJ secooJ- otJet effects. 1 sL order - we observe an lncrease ln Lhe lnformaLlon-processlng capablllLy provlded by ln- memory Lechnology, such as lncreased speed of slngle calculaLlons or LransacLlons.
ll. 1be secooJ-otJet effects of lo-memoty tecbooloqy moolfest lo oJvooceJ bosloess ooolytlcs ooJ tbe coovetqeoce of oolloe ttoosoctloo ptocessloq (Ol1l) ooJ oolloe ooolytlcol ptocessloq (OlAl). 2 nd order - Lhe posslblllLles LhaL arlse from Lhe lncrease ln lnformaLlon processlng capaclLy. ln- memory Lechnology allows CLA and CL1 sysLems Lo be converged by ellmlnaLlng Lhe lnformaLlon laLency of convenLlonal L1L processes and enables analyses of hlsLorlcal" daLa aL Lhe Llme a LransacLlon ls performed.
lll. 1be secooJ-otJet effects of lo-memoty tecbooloqy ote Jtlveo by teJoceJ loteocy tlmes ooJ tbe oblllty to ptocess lotqe volomes of complex Joto.
1hree flrsL order effecLs as drlvers for second-order effecLs (reducLlon of laLency Llmes, Lhe ablllLy Lo process large daLa volumes and Lhe ablllLy Lo process complex daLa).
lv. 1be voloe cteotloo tbtooqb lo-memoty tecbooloqy ls testtlcteJ by tbe copobllltles of tbe %" ovetoll soclo-tecbolcol sttoctotes ooJ ptocesses. 1he reallzaLlon of beneflLs wlll be challenglng because roles and responslblllLles are noL clearly deflned, processes are characLerlzed by lmprovlsaLlon and ad-hoc managemenL or human [udgmenL and declslon-maklng are requlred. new governance sLrucLures musL be adapLed, e.g. Lo allow for more auLonomous declslon maklng.
Tutori al 8
8aars and kermper, Management Support w|th Structured and Unstructured DataAn Integrated 8us|ness Inte|||gence Iramework, lnformaLlon SysLems ManagemenL, vol. 23: 132-148, 2008
1hls paper presenLs Lhe lnLegraLlon of componenLs for handllng unsLrucLured daLa - Lhree approaches glven.
1hree lnLeracLlng Lrends hlghllghL Lhe need for such soluLlons: 1. More LurbulenL, global buslness envlronmenL, 2. AddlLlonal pressures Lo unvell valld rlsk and performance lndlcaLors Lo sLakeholders 3. AggravaLed challenges of effecLlvely managlng more and more densely lnLerwoven processes
Structured data - Lhe daLa LhaL ls asslgned Lo dedlcaLed flelds and LhaL can Lhereby be dlrecLly processes wlLh compuLlng equlpmenL. MosL sallenL Lools for analysls of such daLa: reporLlng, daLa mlnlng and CLA.
1hree maln approaches Lo handle unsLrucLured daLa: 1. Integrated resentat|on - structured and unstructured content are s|mu|taneous|y accessed v|a |ntegrated user |nterface. ( AuLomaLlzed Lrlgger for a parallel search for flLLlng conLenL ln a documenL reposlLory, sulLlng unsLrucLured Lo sLrucLured daLa, CLA daLa and Lhe selecLed documenLs wlll be presenLed slde by slde ( Maln beneflLs: funcLlons Lo access sLrucLured and unsLrucLured daLa can be used LogeLher ln an efflclenL and sLralghLforward manner and users have Lo geL accusLomed Lo only one sysLem wlLh one user lnLerface only ( arallel screenlng can uncover and vlsuallze oLherwlse neglecLed lnLerrelaLlons
2. Ana|ys|s of Content Co||ect|on - ana|ys|s based on a structured descr|pt|on of content |tems w|th metadata ( 8elevanL meLadaLa can elLher be enLered manually by end users or could be consLrucLed ouL of usage logs and search querles %# ( ldenLlflers of Lhe conLenL lLems are LreaLed as facLs LhaL are sub[ecL Lo analysls, whereas meLadaLa flelds are used for classlflcaLlon purposes and serve as analysls dlmenslons, ( 1hree sLeps necessary for Lhls approach: exLracLlon of meLadaLa, lnLegraLlon lnLo sLrucLured daLa reposlLory and lnLegraLed analysls ( Maln beneflLs: Lhls Lype of approach enables compleLely new Lypes of analyses and Lhereby ln-depLhs lnslghL lnLo buslness lnLerrelaLlons
3. D|str|but|on of Ana|ys|s kesu|ts and Ana|ys|s 1emp|ates - d|str|but|on of 8I know|edge that can be eff|c|ent|y shared ( 1hls approach presupposes LhaL Lhere lndeed ls 8l knowledge LhaL can be efflclenLly shared and LhaL ls of some relevance for a sufflclenL number of users ( 1hls analysls mlghL noL be essenLlal for oLher segmenLs - buL Lhe knowledge on how Lo achleve slgnlflcanL resulLs and how Lo presenL Lhem effecLlvely ls ln order Lo check for slmllar developmenLs ( SLeps ln Lhls approach: daLa exLracLlon, daLa analysls and reflnemenL, Lransfer of analysls resulLs or analysls LemplaLes Lo kM Lools for dlsLrlbuLlon ( Maln beneflL: faclllLaLed and more effecLlve and efflclenL appllcaLlon of analysls sysLems and meLhods ( Lspeclally more complex analysls (e.g. lnvolvlng callbraLlon or parameLerlzaLlon) are made wldely accesslble
uavenporL , Compet|ng on ana|yt|cs, ueclslon Maklng, 2006
1hree key aLLrlbuLes among analyLlcs compeLlLors: 1. W|despread use of mode||ng and opt|m|zat|on ' AnalyLlcs compeLlLors look beyond baslc sLaLlsLlcs ' use of predlcLlve modellng Lo e.g. ldenLlfy mosL proflLable cusLomers %$ ' oollng daLa generaLed ln-house and daLa from exLernal sources for comprehenslve undersLandlng of cusLomers 2. An enterpr|se approach ' noL a slngle klller app" buL raLher mulLlple appllcaLlons supporLlng many parLs of Lhe buslness ' AnalyLlcs gaLhered by cenLrallzed groups Lo ensure LhaL crlLlcal daLa and oLher resources are well managed and LhaL dlfferenL parLs of Lhe organlzaLlon can share daLa easlly, wlLhouL problems of lnconslsLenL formaL, deflnlLlons and sLandards 3. Sen|or execut|ve advocates ' Companywlde embrace of analyLlcs requlres leadershlp from execuLlves aL Lhe very Lop ' Lower-level people lack Lhe perspecLlve and Lhe cross-funcLlonal scope Lo change Lhe culLure ln any meanlngful way ' Lxperlence of Lop managemenL: knowlng when Lo run wlLh numbers and when Lo run wlLh guLs
Sources of sLrengLh of analyLlcs compeLlLors: 1. 1he r|ght focus ) AnalyLlcs should encourage facL-based declslons, buL Lhere should be a dlrecLlon for resource-lnLenslve efforLs, llke e.g. lncreaslng cusLomer loyalLy, cusLomer servlce and relaLed areas such as prlclng and promoLlons ) lL's rlsky for your focus Lo become Loo dlffuse or Lo loose clear slghL of Lhe buslness purpose behlnd each 2. 1he r|ght cu|ture ) lnsLllllng a companywlde respecL for measurlng, LesLlng and evaluaLlng quanLlLaLlve evldence ) Lmployees should be urged Lo make declslons based on hard facLs 3. 1he r|ght peop|e ) AnalyLlcal flrms need analyLlcal people, good analysL besldes all analyLlcal skllls should also have Lhe ablllLy Lo express complex ldeas ln slmple Lerms and have Lhe relaLlonshlp skllls Lo lnLeracL well wlLh declslon makers (comblnaLlon of analyLlcal, buslness and relaLlonshlp skllls) 4. 1he r|ght techno|ogy ) uoto sttoteqy - sysLems LhaL snaLch Lhe daLa from every concelvable source, daLa handllng musL be able Lo presenL daLa ln sLandard formaLs, lnLegraLe lL, sLore lL ln a %% daLa warehouse and make lL easlly accesslble Lo anyone and everyone ) 8osloess lotelllqeoce softwote - Lools LhaL allow people Lo exLracL, Lransform and load (L1L) daLa for analysls for declslon-maklng purposes ) compotloq botJwote - Lhe volume of daLa processes ls a challenge for hardware, many analyLlcs companles are converLlng Lhelr hardware Lo 64-blL processors LhaL can cope wlLh large amounLs of daLa more qulckly
8oyd & Crawford, S|x rovocat|ons for 8|g Data, 2011
8|g data ls noLable noL because of lLs slze, buL because of lLs relaLlonallLy Lo oLher daLa. uue Lo efforLs Lo mlne and aggregaLe Lhe daLa, 8lg uaLa ls fundamenLally neLworked. lLs value comes from Lhe paLLerns LhaL can be derlved by maklng connecLlons beLween pleces of daLa, abouL an lndlvldual, abouL lndlvlduals ln relaLlon Lo oLhers, abouL groups of people or slmply abouL Lhe sLrucLure of lnformaLlon lLself. Slx provocaLlons of 8lg uaLa, why lsn'L 8lg uaLa always a CreaL uaLa: 1. Automat|ng kesearch Changes the Def|n|t|on of know|edge 8lg uaLa creaLes radlcal shlfL ln how we Lhlnk abouL research. WlLh enough daLa numbers speak for Lhemselves". uo Lhey? no. 1hls sLaLemenL reveals an arroganL undercurrenL ln many 8lg uaLa debaLes where all oLher forms of analysls can be sldellned by producLlon llne of numbers, prlvlleged as havlng a dlrecL llne Lo raw knowledge. 8lg uaLa ls abouL exacLly rlghL now, wlLh no hlsLorlcal conLexL LhaL ls predlcLlve.
2. C|a|ms to Cb[ect|v|ty and Accuracy are M|s|ead|ng Worklng wlLh 8lg uaLa ls sLlll sub[ecLlve and whaL lL quanLlfles does noL necessarlly have a closer clalm on ob[ecLlve LruLh - parLlcularly when conslderlng messages from soclal medla slLes. ! All researchers are lnLerpreLers of daLa. uaLa needs Lo be lmaglned as daLa ln Lhe flrsL lnsLance and Lhls process of Lhe lmaglnaLlon of daLa enLalls an lnLerpreLaLlve base. ! uaLa cleanlng" process - maklng declslons abouL whaL aLLrlbuLes and varlables wlll be counLed and whlch wlll be lgnored. 1hls ls an lnherenLly sub[ecLlve process. ! Large daLa seLs from lnLerneL sources are ofLen unrellable, prone Lo ouLages and losses and Lhese errors are even furLher magnlfled when mulLlple daLa seLs are used LogeLher. ! 1o make sLaLlsLlcal daLa clalms abouL a daLaseL, we need Lo know where daLa ls comlng from.
3. 8|gger Data are Not A|ways 8etter Data 1here ls a problemaLlc underlylng eLhos LhaL blgger ls beLLer, LhaL quanLlLy necessarlly means quallLy. %& 1wlLLer as a popular source for mlnlng 8lg uaLa, buL worklng wlLh 1wlLLer daLa has serlous meLhodologlcal challenges LhaL are rarely addressed by Lhose who embrace lL. (1wlLLer makes accesslble only 10 of daLa, whlch ls noL represenLaLlve for Lhe whole 1wlLLer communlLy, whlch ls cerLalnly noL represenLaLlve for Lhe socleLy) o 8esearchers comblne blg daLa seLs Lo obLaln even blgger daLa seLs. Whlle one of Lhe seLs conLalns any errors, by connecLlng several such seLs we are drownlng ln errors.
4. Not A|| Data Are Lqu|va|ent Some researchers presume LhaL daLa ls lnLerchangeable. " uaLa Laken ouL of Lhe conLexL looses lLs meanlng and value. ConLexL maLLers. " When Lwo daLaseLs can be modeled ln a slmllar way, Lhls does noL mean LhaL Lhey are equlvalenL or can be analyzed ln Lhe same way. " L.g neLworks produced Lhrough soclal medla and resulLlng from communlcaLlon Lraces are noL necessarlly lnLerchangeable wlLh oLher soclal neLwork daLa. " 1wo popular Lypes of soclal neLworks derlved from daLa Lraces: o ArLlculaLed neLworks - neLworks resulLlng from people speclfylng Lhelr conLacLs Lhrough a medlaLlng Lechnology (frlends, followers eLc.) o 8ehavloral neLworks - neLworks derlved from communlcaLlon paLLerns, cell coordlnaLes and soclal medla lnLeracLlons (people who are acLually communlcaLlng, are Lagged LogeLher ln phoLos eLc.) " 8oLh behavloral and arLlculaLed neLworks cannoL be equallzed Lo personal neLworks
3. Iust 8ecause |t |s Access|b|e Doesn't Make |t Lth|ca| unanswered quesLlons of whaL ls Lhe sLaLus of so-called publlc" daLa on soclal medla slLes? - Any daLa on human sub[ecLs lnevlLably ralse prlvacy lssues, and Lhe real rlsks of abuse of such daLa re dlfflculL Lo quanLlfy - lL's uneLhlcal Lo [usLlfy one's acLlons as eLhlcal slmply because Lhe daLa was accesslble. !usL because conLenL ls publlcly accesslble lL doesn'L mean LhaL lL was meanL Lo be consumed by [usL anyone - Many eLhlcs boards do noL undersLand Lhe processes of mlnlng and anonymlzlng daLa, leL alone errors can cause daLa Lo become personally ldenLlflable. - ulfference beLween belng publlc and belng ln publlc
6. L|m|ted Access to 8|g Data Creates New D|g|ta| D|v|des Cnly soclal medla companles have access Lo really large soclal daLa - especlally LransacLlonal %' daLa 1op unlverslLles are llkely Lo buy large daLa, and Lhe sLudenLs of Lhese unlverslLles are more llkely Lo geL a [ob aL Lhe daLa companles, lncreaslng Lhe gap beLween Lop unlverslLles graduaLes and perlpheral unlverslLles' sLudenLs. 8lg uaLa researches wlLh access Lo proprleLary daLa seLs are less llkely Lo pose quesLlons LhaL are conLenLlous Lo soclal medla company and LhaL could resulL ln Lhelr access Lo daLa belng cuL (chllllng effecL on Lhe klnd of research quesLlons posed) new dlglLal dlvlde: 8lg uaLa 8lch and 8lg uaLa oor 1hree classes of people ln Lhe realm of 8lg uaLa: Lhose who creaLe daLa, Lhose who collecL Lhe daLa and Lhose who analyze Lhe daLa (lasL group ls Lhe smallesL and more prlvlleged)