jee oun JIGSAW ACADEMY ‘Analyiies for Prafecionale Analytics MethodologyANALYTICS METHODOLOGY = * Attack the “right” problem Goi) * Define problem specifics * Available data will sometimes drive solutions == * Complex analytical models not always best solutions x * Majority of time spent here Hae - Iterative process * Ongoing “4 # Assess effectiveness over time coy4 ANALYTIC METHODOLOGIES = Overview of Analytics Methodologies Four Stages + Problem Identification + Solution Identification « Aackihe ght” problem *ineproblem spectes « Avalable data ll sometimesdveolriens Conplexanalyicalmotelsnotahaysbestsolutons + Solution implementation + Solution Monitoring Difference between Analytical Problem Solving and Standard Problem Solving? + Reliance on data to drive solution identification + Use of analytical techniques based on statistical and numeric theories + Understanding of application issues of theoretical concepts to business situations + Conversion of solution results into actionable business steps * Nejortyo tire penthere * torative process + Ongcing + Asessellectvenes overtimePROBLEM IDENTIFICATION Problem Identification How do | identify my most valuable customers? How do | identify the best promotional tactic for my products? How do | ensure that | minimize losses from product not being available on shelf? How da | optimize inventory? How do | detect customers that are likely to defaulton bill payment? How can linerease my revenues by mi customers buy more of my products? Overall business statement may not ‘always directly lead to clear problem identification Understand underlying needs and obtain context all issue into sub-issues to + Available data wil sometimes drive solutions «+ Complorarlytal madl nat anastessouions + Majortyof ine spent here ‘ heratireprocess “Ongoing «Assess efectveness over ineSolution Identification *Ateckthe gt pclen *bsfne podem ec The “best” solution based on theory may not be the “best” solution in practice Lack of good data or specific variables «+ Avalable data wil sometimes dive solutions Solution implementation constraints + Majorty of time spent here «iterative process Other business constraints Need to understand how solution will be implemented to finalize best solution Ongoing © Assess efectiveness overtime * Compl anal model nat evas best solosSOLUTION IMPLEMENTATION + Solution implementation ‘Typically is the longest phase Starts with data collection Data Preparation and Exploratory Data Analysis will typically form a significant part of total project effort and time Initial modeling results and calibration Finalization of models post calibrations Interpretation and framing of results in business context * Atekthee pblen «Def role pecs + alae til sometimes ve suo *Conpleranace mtlsnot das bests ‘+ Majrtyoftie ent here ‘heratire process Ongoing + Assess effectiveness over time aee SOLUTION MONITORING Solution Monitoring Very important phase that is sometimes ignored Once solution is implemented, need to ensure that original assumptions/constraints, still hold Constant monitoring of solution to ensure tat the ht ren *Defneprolen sedis changes are accurately captured and “Majority afte spet here assessed erie process Solutions age — need to track reliability over time + Ong + Asesefetieres oertire i +d onines de sos ingame tas betasPROBLEM DEFINITION Problem Definition Itis critical to frame problem accurately and realistically Here’s a “problem” statement We want to only accept credit card applications from good customers Problems with the “Problem” statement: What are “good” customers? How can one control applications to be “only good”? Is itrealistic to assume that one can build a static algorithm that will never allow acceptance from a customer with a “bad” record?PROBLEM DEFINITION — CONTD. How could this problem statement be modified to allow useful analytic solution development? Example: Use SIMART technique: Specific Measurable Attainable Realistic Timely Understand the bu: problem ess context thoroughly to ensure you are tackling the right4 SOLUTION IDENTIFICATION STAGE e + Solution Identification ‘There could be multiple ways of approaching a problem, and not all analytical based In general, if you want to use data to drive decisions, you need to first make sure that you have the right data For example, you want to predict the potential trial rate of an ultra-luxury brand that you want to launch, but you have so far launched brands that are more value conscious ~ no data for an analytical approach + You could try to obtain data on competitor launches or run some qualitative research in this case, but itis not a good idea to develop a trial % based on a different segment of populationSOLUTION IDENTIFICATION - CONTINUED = + Solution depends on end use: Reporting for insights . Use historical data to understand performance and identify patterns in data and use as basis for future planning Forecasting: Use past data collected over time for actual forecasts of future performance Predictive Analytics: Similar to forecasting, but not just time related Optimization: Develop strategies to derive op’ constraints ial use of resources given4 SOLUTION IMPLEMENTATION 6 Once the best solution approach has been identified, the actual solution building and implementation phase will start This phase includes all the data exploration and preparation phases, the analytical techniques application, and the final implementation of the derived solution This phase typically accounts for anywhere between 60 ~ 80% of total project timeSOLUTION IMPLEMENTATION £ + Typical Steps within Solution Implementation Data Exploration Data Preparation Data Partitioning Model Building — Test Data Set Initial Model Model Iterations and Tweaking Final Model Validation Model More Iterations If required based on Validation Model Results Interpretation of results into business friendly context Solution implementation ae 4SOLUTION MONITORING Solution monitoring is required to ensure that solution supplied is continuously providing expected results For example, if we have devised a model to predict probability of default, if the characteristics of the underlying applicant pool changes or there are other external factors affecting population behavior, you will need to modify the solution or potentially come up with a new one ee 4-. ANALYTICS BEST PRACTICES € Models should be as simple as possible Model results and recommendations need to be actionable Models need to be rigorously validated on multiple sample sets if possible The most critical three things to remember: Domain, Domain, and Domain ee 4