leading a data team that supports the business into a ...analytic presentation & enterprise...
TRANSCRIPT
Leading a Data Team that Supports the Business into a Cohesive Data-driven UnitCanon Data Science & Analytics – An Overview
Apr – 2018
UsmanShahbazSeniorManagerDataScience&Analytics,Canon
Usmanhasmorethan13yearsofexperienceindatascience,datastrategyandadvanceddataanalytics.WorkingasSeniorManagerDataScience&AnalyticsatCanon,heisprogressivelyworkingtowards
developingfull-scaleanalyticalcapabilityacrossmultiplecustomerchannelsthuscreatingstrategicleverageforthebusiness.PriortojoiningCanon,Usmanhadledteamswithintelcoandretailindustriestoeffectivelydeliverondatastrategy,consumerinsights,datavisualisation andadvancedanalyticalmodelling.Usmanis
currentlyenrolledforaPhDinMachineLearning.HealsoholdsanMBAandaBachelor’sdegreeinElectricalEngineering.
Without Data You are just another person with an OpinionW. Edwards Deming “ ”
@rev16
About Canon – Our Journey in Oceania
Howitallbegan
Intheearly1930’s,inasmallroominTokyo,someyoungentrepreneurshadadream.TocreateJapan’sfirst-ever35mmcamerawithfocalplaneshutter.
Thesuccessofthatfirstprototypewasthebeginningofsomethingmore,therestishistory!
Canon Oceania Journey
Tobeadata-drivenorganizationwhere weanalyse andpredictconsumerbehaviour toarticulateorganisational strategyandoptimise ourconsumerinteractions.
LEVEL-1
Identifying Data Sources & Data
Islands
LEVEL-2
Bridging Data sources through
Rapid Prototyping
LEVEL-4
Diagnostic/Predictive Insights fuel
business hypothesis
“Data as a way of work”
Data centered decision making
LEVEL-3
Build Enterprise Data Model/ Production Algorithms
OurAmbition
StrategicPillars
DataRoadmap
DESCRIPTIVE
Gain insights from historical data with reporting & visual
dashboards.
DIAGNOSTIC
Build complex Data Models to evaluate
business hypothesis
PRESCRIPTIVE
Recommend decisions & course of action based
on the predictive algorithms
PREDICTIVE
Data Models able to predict patterns and behaviors through machine learning.
LEVEL-5
Prescriptive analytics drives key strategies
Data Strategy
….But before we move further
Hadoop is not the answer to everything…Not Yet…… Right Data Architecture is
….We need to talk about the Elephant in the room…..
AnalyticalDataProcessing&ModellingCenterofExcellence
TM1
AnalyticPresentation&EnterpriseDistribution
MicrosoftAX- ERP
CDSS– ERP
AD(HR)
MasterDataFiles
SalesForce SFDC
PDWArchivedData
MyCanon
SocialBakers&SocialNetworks
GoogleAnalytics
EnterpriseDWH/DimensionalModel
Cubes/TabularModels
PBIDesktop&PBIServiceportal
DataDiscovery
ReportsandDashboards
SSRSReportPacks
OperationalDataStore
MetricsandotherFacts
DimensionalAttributesandHierarchies
BusinessDataGlossary
ReportTaxonomy
MasterDataHub
DataGovernance/DataStewardship
DataLineageandProcessingControl
Microsoft Analytics Architecture
Microsoft Data Science R Server Architecture
Python Production code in SQL Server - 2017
Rapid Prototyping - Analytical Lifecycle
DecisionMakingonInsightsNewlyestablisheddatareportormodel
empowerbusinesswithactionableinsights.Businessevaluatesdecisionsbasedonthe
insights.
DataVis.OrMLProductionModel
BasedontheDWHschemaandbusinessrequirementsproductionmodelisshared
withthebusinessstakeholders..
ProcessAreaWorkshopRunworkshopswithbusinessstakeholderspertainingtheindividualprocessareas
BuildAnalyticalPrototypeCreateabusinessprototypeoftherequirementsandusethattovalidatewithstakeholders
ScaleatEnterpriseAnalyticalModel
OncevalidatedbythebusinesspasstheprototypeforascalableEnterprise
AnalyticalModel
Putting the Prototyping Structure to Work
RapidPrototyping&BusinesslogicDesignAcceleratedbusinesslogictest,sampledatatestandvalidatebusinessrequirement
SolutionIntegration/Optimisation&testing
Solutionintegrationwithrestoftheanalyticalpieces,comprehensivetesting&required
optimisation.
ProductionDeployment&Presentation
Basedonthesolution,productiondeploymentwithsecuritymatrix
appliedbuildsoutthepresentationlayerwithautomation.
MachineLearningExpert
Developsprimarydatatesting,buildsanddevelopstestingalgorithm.ValidatesBusiness
Hypothesis
SolutionDevelopmentArchitectscalabledevelopmentoftheprototype,DataEngineeringforODS/DWH
VisualisationExpertBuildsdatamodelsafterdatadiscoveryandbuildsrequiredbusinesslogicincloseconsultationwithBusiness
Owners
BusinessOwnerInitializestheprocessbysubmittingthebusinessrequirementandlogic.Evaluatesthefinalreportthathas
beenproduced
Journey that we have covered…. • Usingrapidprototypingwithbusinessunitstovalidatebusiness
hypothesis• Bridgedatasourcesandcreateprototypeswithconsolidatedview• Datagovernanceframeworkacrossgrouptofirmupprocedural
changes• ScalingthesolutionsinSQLserveraftertheprototypevalidation• PowerBIasthetooltoconsumeanalyticalcontentacross
organisation• Democratising notonlythedatabutalsotheabilitytodraw
powerfulinsights
Way Forward….//>>
• InsteadofthemaintainingthewebAPIinPowerBI,pushtheminSSASlayer
• Bettermanagementintermsofthesecuritymodelaswellnomenclaturedefinitionandmaintenance
• UsethepowerBIqueryfunctionalityforrapidprototypingbutSSASlayerforproductionlayer
• UsebaselevelRtestingthroughRvisualsinpowerBIinsteadoflocalmachineRlibraries
• Continuebuildingprototypesusinglocalmachinelibraries• ContinuescalingusingSQLStoredProceduresforproduction
versionofthealgorithms
Crawl
Walk
Run
What we could have done differently…• Insteadofconnectingeachflatfileconnectionin
PowerBI,weshouldhaveusedthefolderconnections
• InstalltwoinstancesofEnterpriseGatewayonefordevelopmentandanotherforproductionreleases
• UsethepowerBIqueryfunctionalityforrapidprototypingbutSSASlayerforproductionlayer
• UsebaselevelRtestingthroughRvisualsinpowerBIinsteadoflocalmachineRlibraries
• Buildingworkspacedoesnotallowwhopublishedwhat….withintheteam
Thank YouQuestions?