Introduction to Prescriptive Analytics: Solving Real World Optimization Problems using IBM ILOG CPLEX Optimization.

Size: px

Start display at page:

Download "Introduction to Prescriptive Analytics: Solving Real World Optimization Problems using IBM ILOG CPLEX Optimization."

Winifred Oliver
5 years ago
Views:

1 Introduction to Prescriptive Analytics: Solving Real World Optimization Problems using IBM ILOG CPLEX Optimization

2 Housekeeping Link to Webinar Recording and Presentation Slides will be shared after the presentation. Comments and Questions in the GoToWebinar Control Panel. Additional questions should be directed to Nabeel Nazeer - nnazeer@newcomp.com 2

Open Source Trusted Analytics Advisors Newcomp Analytics is a leading Analytics partner that provides software, services, support,

3 About Newcomp Analytics A dedicated analytics practice since 1997 with over 400 successful project implementations and satisfied clients. Focus on 5 Analytics Pillars 1. Business Intelligence 2. Planning & Forecasting 3. Advanced Analytics 4. Information Management 5. Open Source Trusted Analytics Advisors Newcomp Analytics is a leading Analytics partner that provides software, services, support, renewals, training and education for IBM Solutions IBM Platinum Business Partner 2017 North American IBM Strategic Partner of the Year 3

4 Today s Presenter Alkis Vazacopoulos Optimization Expert 4

5 Introduction to Prescriptive Analytics: Solving Real World Optimization Problems using IBM ILOG CPLEX Optimization Alkis Vazacopoulos Optimization Expert 5

6 Why Optimization? Plans are nothing; planning is everything Dwight D. Eisenhower 6

7 What is the difference between industry leaders? vs. vs. vs. vs. vs. vs. Source: The Optimization Edge, Steve Sashihara (New York, NY: McGraw Hill, 2011) p. 3 7

8 Optimization solutions documented ROI 2 Chilean Forestry firms Timber Harvesting $20M/yr + 30% fewer trucks UPS Air Network Design $40M/yr + 10% fewer planes South African Defense Force/Equip Planning $1.1B/yr Motorola Samsung Electronics Procurement Management Semiconductor Manufacturing $100M-150M/yr 50% reduction in cycle times SNCF (French RR) Scheduling & Pricing $16M/yr rev + 2% lower op ex Continental Airlines Crew Re-scheduling $40M/yr AT&T Network Recovery 35% reduction spare capacity Grantham Mayo van Otterloo Portfolio Optimization $4M/yr Source: Edelman Finalists, or 8

Hard Benefits of Optimization - Calculable ROIs, paybacks within months, sometimes even weeks - Capital expense avoidance or deferral - Operating expense reductions - Total revenue, revenue mix, and

9 Hard Benefits of Optimization - Calculable ROIs, paybacks within months, sometimes even weeks - Capital expense avoidance or deferral - Operating expense reductions - Total revenue, revenue mix, and margin improvements - Improved customer satisfaction - Provide better and more customized customer service - Reduce costs; improve customer service - Improved operations - Increase productivity - Better planning and scheduling processes - Minimize Costs, Maximize Profits 9

10 What is Decision Optimization? 10

11 The advanced analytics picture IBM s Advanced Analytics Portfolio Descriptive Analytics Predictive Analytics Prescriptive Analytics What is happening in my business today? What will happen in the future? What should I do about it? 11

The advanced analytics picture IBM s Advanced Analytics Portfolio Decision Optimization is the secret sauce of Prescriptive Analytics Descriptive

12 The advanced analytics picture IBM s Advanced Analytics Portfolio Decision Optimization is the secret sauce of Prescriptive Analytics Descriptive Analytics Predictive Analytics Prescriptive Analytics What is happening in my business today? What will happen in the future? What should I do about it? 12

Business Analytics and Optimization Prescriptive Stochastic Optimization Optimization Predictive Modeling Competitive Advantage Predictive Forecasting/ Extrapolation

13 Business Analytics and Optimization Prescriptive Stochastic Optimization Optimization Predictive Modeling Competitive Advantage Predictive Forecasting/ Extrapolation Statistical Analysis Alerts Query/ Drill Down Ad hoc reports Standard Report Descriptive 13 Complexity From the book Competing on Analytics by Thomas H. Davenport, Jeanne G. Harris 13

Decision Support Applications Key steps for a mature decision support application leveraging advanced analytics Historical & Master Data ETL Capture price, product, location and date for each

14 Decision Support Applications Key steps for a mature decision support application leveraging advanced analytics Historical & Master Data ETL Capture price, product, location and date for each transaction. Predictions and Insights Determine important variables, predict trends, seasonality etc. Rules & Process Management Set policies, promotions etc. Allow reviewers and auditors to have a say. Collaboration & What-if Allow multiple users to experiment with multiple scenarios. Decision Making Automatically generate decisions, allow user interaction with decisions. Descriptive Predictive Prescriptive 1 14

15 Prescriptive Analytics by Industry 15

16 Indeval (Mexican Central Securities Depository) Challenge Process security transactions in real-time rather than daily. Provide a better service to the Mexican Stock Exchange. Solution Decision Optimization for assignment and scheduling Profile A private securities depository organization in Mexico. Benefits/ROI Real time reconciliation and completion of trading operations for more than USD$250 B in average, every day Reduced liquidity requirements for trading partners by 52 percent Increased the volume of operations by 26 percent Reduced the costs of each trading transaction for electronic trading facilities, the Stock Exchange and trading brokers Testimonial By building a unique technology solution for our securities services, we now better serve the Mexican Financial Community and trading partners. We are very proud that this solution has played a key role in helping elevate the economy of Mexico. Jaime Villaseñor Chief Risk Officer, INDEVAL 16

Auditing Tax Returns at New York State Challenge Questionable tax refunds can total about $400-500M. 98,000 exception returns are processed. 800-1200 people analyze exceptions.

17 Auditing Tax Returns at New York State Challenge Questionable tax refunds can total about $ M. 98,000 exception returns are processed people analyze exceptions. Solution BPM for process automation SPSS predictive analytics for scoring Decision Optimization for assignment and scheduling Benefits/ROI Increased collections of outstanding debt by $83M Average age of cases when assigned to field agents decreased by 9.3% Dollars per staff day increased by 15% 35,000 fewer taxpayers had serious enforcement actions taken against them. Collections process is more productive, efficient and fairer to all taxpayers. Profile New York State 20M total population ~$200B total state tax revenue (third highest) 17

Constant changes in the market, technology and suppliers Solution Excel spreadsheets were replaced by dedicated optimization solution for supplier planning Solution allowed for freeze period, limited

18 Flash Memory Supplier Management Challenge Place orders for flash memory from manufacturers Determine how memory will be used. Which product line, which market and which assembly shop will utilize provided raw materials. Initial 4 week freeze period. 4-8 weeks of limited flexibility week of full flexibility. Constant changes in the market, technology and suppliers Solution Excel spreadsheets were replaced by dedicated optimization solution for supplier planning Solution allowed for freeze period, limited flexibility and full flexibility periods. Benefits/ROI Planning team can now collaborate on a global plan Consistency in decision making Audit capability on old plans Customer Profile Global consumer electronics retailer Multiple markets, multiple product lines, multiple suppliers Multiple global suppliers

Detailed Scheduling, Monitoring and Re-Scheduling Challenge Increasing complexity in the Final Assembly Line Volatile market demand Tough competition, pressure on costs & on time deliver Need to

19 Detailed Scheduling, Monitoring and Re-Scheduling Challenge Increasing complexity in the Final Assembly Line Volatile market demand Tough competition, pressure on costs & on time deliver Need to Increase productivity, utilization and efficiency. Ability to manage unforeseen event during production Capitalize very experienced planners expertise Solution IBM ILOG optimization solution to schedule, monitor and reschedule production. Complementing SAP Delivering understandable and executable schedule in minutes with the capability to plan activities, outstanding work, and manage unexpected events that can build a detailed schedule in a few minutes. Ability to monitor progress and re-schedule on demand Benefits/ROI One unique tool for all Final assembly lines for all AIRBUS aircraft families Starting deployment for aircraft component Planning effort reduced from 7 days to 3-4 hours. Re-scheduling in minutes Customer Profile Airbus is a leading aircraft manufacturer whose customer focus, commercial know-how, technological leadership and manufacturing efficiency have propelled it to the forefront of the industry

20 IBM Decision Optimization portfolio Market Leading MP & CP Optimization Engine Model complex business problems. Solve with IBM CPLEX Optimizer. Prescribe precise and logical decisions. SaaS Delivery via Decision Optimization on Cloud Prescriptive analytics as a service. No install, no setup. Embed in other applications via Rest API. 20

Decision Optimization in end-to-end IBM Analytics Example:

data and master data TM1 Automate capacitated plan generation

key correlations SPSS Allow planners to create multiple whatif

21 Decision Optimization in end-to-end IBM Analytics Example: Advanced S&OP Lifecycle Import historical sales transaction data and master data TM1 Automate capacitated plan generation Decision Optimization Determine demand trends, seasonality and key correlations SPSS Allow planners to create multiple whatif scenarios TM1 Let planners review and edit forecast and master data TM1 21

22 Optimization for LoB typical use cases Marketing Sales Finance IT Operations HR Descriptive & predictive analytics ROI analysis Customer retention Prioritizing accounts Receivable Helpdesk case Analysis Warranty analysis Employee retention Prescriptive analytics (optimization) Campaign planning Territory optimization Portfolio optimization Staff assignment Sales & operations planning Compensation & training planning 22

Descriptive analytics How should I create territories to maximize quota

23 Bridget s story Bridget Hart Sales Manager What is the revenue breakdown by region? Descriptive analytics What are the predicted sales per region? Descriptive analytics How should I create territories to maximize quota achievement fairly? Descriptive analytics Descriptive Analytics Predictive Analytics Prescriptive Analytics 23

24 Bridget s story Bridget Hart Sales Manager What is the revenue breakdown by region? Descriptive analytics What are the predicted sales per region? Descriptive analytics How should I create territories to maximize quota achievement fairly? Descriptive analytics Descriptive Analytics Predictive Analytics Prescriptive Analytics 24

25 Combining Machine Learning with Prescriptive Analytics Used Cases 25

26 Decision support : Don t stop at the insight level Historical Data Insight Forecast Decision Data ML model Score Data DO model Plan Customer A is about to churn (score = 95%) Customer B might need product C (score = 20%) Send a proposal X to customer A Send a proposal Y to customer B 26 IBM Analytics University 2018

27 Transportation Machine learning Forecasted demand Travel times Traffic Weather Past trips 27

28 Transportation Machine learning Forecasted demand Travel times Traffic Weather Past trips Decision optimization Policies Capacities Actual demand Routing 28

29 Insurance and fraud Machine learning Decision optimization Weather patterns Past claims Actual claims/ cases Skills Claim cases Forecasted claims and fraudulent cases Schedule Agents Customer class Claim/case type (house, car, etc.) Capacity Routing 29

30 Finance Machine learning Scenario 1 Market Scenario 2 Scenario 3 30

31 Finance Machine learning Decision optimization Scenario 1 Budget Risk Market New portfolio Scenario 2 Scenarios Scenario 3 Rules diversification Commission fees 31

32 Manufacturing Machine learning Customers Output: expected orders, durations 32

33 Manufacturing Machine learning Decision optimization Customers Lines and recipes Line 1 R1 R2 Output: expected orders, durations Line 2 R1 R2 R3 33

34 Manufacturing Machine learning Decision optimization Customers Lines and recipes Line 1 R1 R2 Output: expected orders, durations Production schedule Line 2 R1 R2 R3 34

35 Use Case - Retail 35

36 Pricing using promotions, markdowns, and clearance strategies 36

37 Retail optimization use case - Vertical: Retail - Products: Apparel & Accessories - Objective: Maximize Revenue, Maximize margin, Reduce Inventory - Decisions: Dynamic Pricing - What do I have: Initial Plan - Status: Review the week - Decisions: Pricing - Dynamic Pricing - Markdowns - Price Points - Clearance - Promotions 37

38 Our sales plan for last week was: PLAN Last week Sales $ Units Sold Margin $7,689, , % REVENUE TARGET 38

39 How did we do? Plan vs. Actual PLAN Last week Sales $ Units Sold Margin $7,689, , % Actual Last Week Sales $ Units Sold Margin $7,083, ,390 51% REVENUE TARGET ACTUAL Revenue 39

40 How did we do? Plan vs. Actual PLAN Last week Sales $ Units Sold Margin $7,689, ,000 54% Actual Last Week Sales $ Units Sold Margin $7,083, ,390 51% We missed both in sales revenue, units sold and margin 40

41 How did we do? Plan vs. Actual PLAN Last week Sales $ Units Sold Margin $7,689, , % Actual Last Week Sales $ Units Sold Margin $7,083, ,390 51% Which season/s was the problem? 41

42 Where we miss? SPRING 2016 is the problem! PLAN Last week SPRING SEASON Sales $ Units Sold Margin $5,515, , % Actual Last Week Sales $ Units Sold Margin $4,571, , % We missed on Revenue and on units 42

43 What can we do? - Using TM1/IBM Planning Analytics, we can analyze the data and identify Variance in the Plan vs. Actual - How can we affect the demand? - Promotions - Markdowns - Clearance - How do we decide which products, groups, when to act? 43

44 Technology - We use Predictive analytics - To predict the sales for next week/s - To identify slow and fast moving products - To identify products that react well in markdowns and promotions - We use Prescriptive analytics optimization - To decide optimal prices that maximize our revenue - To decide when to offer promotions to maximize our revenue 44

45 What are the data we need for each SKU? SKU ID Price Cost Days on the Floor SKU999 $ $ days Total Quantity Ordered Revenue Cost of Sold Current Margin Total Sold Units Total Length of Selling period 1794 $9969 $ % weeks Liquidation price $6.85 Average Price $16.07 Total Sold * Price IS NOT EQUAL to REVENUES SO FAR Avg. Sales through 3.14% 45

46 What is the output of the optimization? SKU ID Price Cost Days on the Floor SKU999 $24.04 $ days Total Quantity Ordered Revenue Cost of Sold Current Margin Total Sold Units Total Length of Selling period 1794 $9969 $ % weeks Liquidation price $6.85 Average Price $16.07 Avg. Salesthrough PROMOTE 30% 3.14% NEXT WEEK 46

47 $12,000 Markdown & Promotion strategy for a slow-moving product 50% off Revenues $10,000 $8,000 $6,000 $4,000 20% off 30% off 40% off 50% off $2,000 $ Weeks 47

48 $12,000 $10,000 Markdown & Promotion strategy for a fast-moving product 30% off 10% off 20% off 50% Revenues $8,000 $6,000 $4,000 $2,000 $ Weeks 48

49 Revenue $240,000 $210,000 $180,000 $150,000 $120,000 $90,000 $60,000 Effect of Markdowns & Promotions on Revenue Very slow $79,000 $100,608 Slow moving $133,000 $174,846 $187,000 Fastmoving $204,279 Revenue without Markdown & Promotion 49

50 70.0% 55.0% 40.0% Effect of Markdowns & Promotions on Margin FAST SLOW 69.4% 66.6% 64.3% VERY SLOW 53.0% 37.9% 25.0% 10.0% 20.9% Margin without Markdown & Promotion Margin with Markdown & Promotion Original Sales Rate 50

51 Learn more Product Slow moving Fast Moving Bad Sales Lift Good Sales Lift Bad Sales Lift Good Sales Lift 51

52 $7,000 $6,000 50% off 60% off Revenues $5,000 $4,000 $3,000 $2,000 40% off 50% off 40% off 30% off 50% off 60% off 40% off $1,000 $ Weeks 52

53 $210,000 $180,000 $174,846 Revenue $150,000 $120,000 $100,608 $133,000 $142,379 $90,000 $79,000 $81,064 $60,000 SLOW Revenue without Markdown & Promotion FAST MOVING Revenue with good sales lift 53

54 70.0% 55.0% 53.0% 64.3% 56.1% 40.0% 37.9% 25.0% 10.0% 20.9% 22.4% SLOW MOVING PRODUCT Margin without Markdown & Promotion Margin with bad sales lift FAST MOVING PRODUCT Margin with good sales lift 54

55 Call to Action - Connect with Newcomp Analytics to discuss your current use cases - Where can you see Decision Optimization adding value in your organization? - Proof of Concept with your data showcase the art of the possible 55

56 56 Thank You! 56

IBM ILOG Optimization and Supply Chain Applications Overview

IBM ILOG Optimization and Supply Chain Applications Overview Mozafar Hajian, Ph.D, CITP Client Technical Professional, ILOG Optimization and Supply Chain Successful companies Use Optimization To Make better