Putting Predictive Planning to Work
- 1. Integrated Business Analytics Solutions Putting Predictive Planning to Work Ron Moore June 12, 2018
- 2. 2 Ron Moore • Principal Architect at Ranzal • Over 20 years Essbase consulting and training experience • Certified in Essbase, Planning and R programming • Many webcasts and KScope sessions • 19 Oracle University Quality Awards Intro
- 3. 3 Comprehensive Business Solutions Ranzal’s solutions drive improved business performance through better decision making, strong customer engagement and optimized operations Deep Oracle Partnership Drives Customer Value Adaptable Deployment Models Diverse Client Portfolio & Industry Expertise Bio Tech and Pharma Medical Supplies Team Highlights Multiple Oracle ACEs Seasoned delivery team with avg ~6 yrs serving Ranzal clients Experienced mgmt team with avg 12 yrs leading Ranzal
- 4. 4 8 Speaker Sessions Monday, 6/11: • 10:45am – 11:45am: Baha Mar's All In Bet on Red - The story of integrating data and master data with PBCS, FCCS and ARCS • 2:30pm - 3:30pm: Visual Approach to Essbase Calcs: 2018 • 4:15pm - 5:15pm: Integrated Planning Using Enterprise Planning and Budgeting Cloud Service at Sims Metal Management Tuesday, 6/12: • 9:00am - 10:00am: FDMEE versus Cloud Data Management - The Real Story • 10:15am - 11:15am: Edgewater Ranzal: Winning Strategies for Oracle Cloud Adoption: Should You Test Drive, Lease, or Buy? • 2:15pm - 3:15pm: Why Should I Care About DVD? Blu-Ray is the New Thing, Right? Wednesday, 6/13: • 11:45am - 12:45pm: Putting Predictive Planning to Work • 2:15pm - 3:15pm: EPM Automate - Automating Enterprise Performance Management Cloud Solutions Visit us at Booth # 407
- 5. 5 Visit us at Booth # 407
- 6. 6 • Overview • Potential Predictive Planning objectives • Walkthrough of a prediction • Understanding the prediction methods • Lessons from the field Agenda
- 7. 7 • Smart View • Open forms or use ad-hoc views • Web interface for Planning Cloud • Tight integration with Planning forms • Automatically tests different forecasting methods, chooses the best and creates the forecast • 12 time series based methods including, moving averages, exponential smoothing, seasonal and non-seasonal and ARIMA Features Overview
- 8. 8 • Automatically handles missing data and outliers • Flexible control of forecast granularity • Optionally source history from an alternate plan type • “Paste” predictions for Most Likely, Best Case and Worst Case • Comparative views • Predefined results report • Extract results to Excel Features Overview - continued
- 9. 9 • Crystal Ball • Additional features such as Monte Carlo, correlation and regression • Essbase Calc Scripts/Business Rules • @TREND includes exponential smoothing and regression • @CORRELATION • Oracle R Enterprise (ORE) • Part of Oracle Advanced Analytics Related functionality
- 10. 10 • While the brass ring is creating more accurate forecasts, there is a lot of value in a “second opinion” • Automatically create “seed” forecasts • Easily calculate and apply seasonality • Identify trends human forecasters might miss • Save time • Sanity check Potential Predictive Planning Objectives
- 11. 11 • Forecast revenue and costs for P&L forecasts • Forecast walk-in traffic • Forecasts clicks for online sales • Forecast re-stock requirements for large number or low/medium costs parts Examples
- 12. 12 • Manual forecasts for a large number or low cost parts would require a lot of manual effort for low to moderate return • Seasonality is easy to capture statistically and labor-intensive manually • Geographic distribution multiples the number of forecasts required Example
- 13. 13 • Twice as much data as you want to predict • 2 full cycles for seasonality • Predictive Planning will interpolate missing values and normalize outliers • Too aggregated will lose definition and you may not have a place to store it • Too granular may be too sparse or too volatile Data Considerations
- 14. 14 Smart View Predict Toolbar
- 15. 15 • Period and Year • Scenarios: e.g. Actual, Forecast and optionally transformations • Versions for forecast Best, Worst and Middle cases • “Business Units” : what levels do you want to store • Accounts : what levels do you want to store Outline Design
- 16. 16 • Time Axis • Period, Year or both • Optionally Scenario and/or Version • No others • Series Axis • e.g. accounts or entities • Actual and Forecast Scenarios in rows • Not needed if you copy Actual to Forecast. Form Design
- 17. 17 • Create Versions for Prediction and difference from control data (actual or comparison data) • Create dynamic difference formula Form Design – analyze error
- 18. 18 • Lowest level of period determines granularity of prediction • Prediction end date is independent of the form • You can predict form members that are read only, but can’t paste them. Form Design (continued)
- 19. 19 Walkthrough of a Prediction using Smart View
- 20. 20 Open a Form
- 21. 21 Set Up Prediction Menu Purpose Data Source Select data source plan type and Date Range Map Names Select scenarios and versions for comparison and prediction destination Member Selection Select which members to predict Options Select prediction options
- 22. 22 • Optional connection to an alternate plan type with additional years • Select date range Set Up Prediction – Data Source
- 23. 23 • Select scenario/version combinations • Historical data • Comparison views • Pasting predictions Set Up Prediction –Map Names
- 24. 24 • Which members on the form to predict • Skip read-only Set Up Prediction – Member Selection
- 25. 25 • Seasonality • Data “clean-up” • Which methods • Error measure • Confidence Interval Set Up Prediction - Options
- 26. 26 Set Up Comparison Views
- 28. 28 Predict Button - continued
- 29. 29 Prediction Results • Results in the Predictive Planning Panel but not yet pasted to form • Select member prediction to view from drop down
- 30. 30 Predictive Planning Panel – Data and Statistics tabs
- 32. 32 • Choose which prediction to paste • Current member • All members • Filtered members • Selected members • Choose the destination to store the predictions • Submit Data!! Paste Results
- 33. 33 Submit Data
- 34. 34 Create Report
- 37. 37 Extract Data
- 38. 38 Extract Data - Output
- 39. 39 • Open the form • Actions | Predictive Planning Planning Web Interface
- 41. 41 Settings
- 44. 44 Components of a Prediction Component Parameters Level Alpha : smoothing parameter between 0 and 1 not inclusive Trend Beta : smoothing parameter for the second pass between 0 and 1 not inclusive Cycle/Seasonality Gamma: smoothing parameter for seasonality between 0 and 1 not inclusive Trend & Seasonality All of the above Error/Noise Source: Adapted from Predictive Planning documentation
- 45. 45 Understanding the Prediction Methods Non-seasonal Method (Non-seasonal) Description Best for Forecast Type Parameters Single Moving Average Simple moving average Volatile data with no trend Straight flat line Period. 1 to ½ the number of data points Double Moving Average Applies moving average twice Trend but no seasonality Straight sloped line Period. 2 to 1/3 the number of data points Single Exponential Smoothing Weights more recent data more heavily Volatile data no trend No seasonality Straight flat line Alpha Double Exponential Smoothing Applies SES twice Trend , no seasonality Straight sloped line Alpha and beta Damped Trend Smoothing Trend is damped Trend , no seasonality Trend flattens over time Alpha, beta and phi Source: Adapted from Predictive Planning documentation
- 46. 46 Method Description Best for Forecast Type Parameters Seasonal Additive Exponentially smoothed forecast + seasonal adjustment No trend and seasonality Seasonal cycle without trend Alpha, Gamma Seasonal Multiplicative Exponentially smoothed level and seasonal adjustment * seasonal adjustment No trend and seasonality increases or decreases Seasonal cycle without trend Alpha, Gamma Holt-Winter’s Additive Exponentially smoothed level, trend and adjustment + seasonal adjustment Trend and stable seasonality Trend and seasonal cycle Alpha , beta, gamma Holt-Winter’s Multiplicative Exponentially smoothed level, trend and adjustment * seasonal adjustment Trend and increasing seasonality Trend and seasonal cycle Alpha , beta, gamma Damped Trend Additive Seasonal Projects seasonality, damped trend and level separately and reassembles - additive Trend and seasonality Flattening with seasonality Alpha, Beta, Gamma, Phi Damped Trend Multiplicative Seasonal Projects seasonality, damped trend and level separately and reassembles - multiplicative Trend and seasonality Flattening with seasonality Alpha, Beta, Gamma, Phi Understanding the Prediction Methods Seasonal Source: Adapted from Predictive Planning documentation
- 47. 47 Non Seasonal Seasonality Stable Increasing or Decreasing No Trend Single Moving Average Single Exponential Smoothing Seasonal Additive Seasonal Multiplicative Trend Double Moving Average Double Exponential Smoothing Damped Trend Non-seasonal Holt-Winter’s Additive Damped Trend Additive Damped Trend Multiplicative Holt-Winter’s Multiplicative Methods by Trend and Seasonality Source: Adapted from Predictive Planning documentation
- 48. 48 Error Measurements Error Abbr. Error Measure Description RMSE default Root Mean squared error Square root of the average of the squared errors MAD Mean Absolute Deviation Average of the absolute value of the errors MAPE Mean Absolute Percentage Error Average of the sum of the average errors
- 49. 49 • Hold out a few months to compare to actuals • How much data can you afford to hold out before it affects the forecast, in particular seasonality? • Hold out some business units • Do the Business Units behave the same way from a forecast point of view? Back-Testing Approaches
- 50. 50 • Data quality especially at lowest levels • New company with many new service introductions created a lot of cannibalization • Missing data at lowest levels. Need to pick a level with stable data. • Logically inconsistent sets of forecasts. e.g. Revenue and costs. Costs should probably be driven by revenue not forecasted independently. • “Events” that affect the actual outcome e.g. advertising, weather Lessons From the Field
- 51. 51 • Identifying and quantifying seasonality alone can be a big improvement and time savings • Consider Predictive Planning for selected “seed” forecasts • Predictive Planning may identify trends that aren’t obvious looking at numbers in a spreadsheet. • Consider Predictive Planning where you need speed and low cost • Consider Predictive Planning as a sanity check or second opinion Lesson form the Field - continued
- 52. 52 Let’s Connect on LinkedIn! • Open the LinkedIn app on your phone • Click My Network • Select Find Nearby • Connect with me and your peers!