Interactive Application for Product Demand Forecasting
Interactive Application for Product Demand Forecasting James Bird Data Scientist, NXP Semiconductors
Richard Arebalo Supply Chain Forecast Team, NXP Semiconductors #TDPARTNERS16
GEORGIA WORLD CONGRESS CENTER
Outline • • • • • • • • • • 2
NXP overview About me EBI team overview Effective Data Science Motivation for Project Project Overview Teradata in NXP overview Solution Architecture – R / Shiny Demand Forecasting Application Benefits / Conclusion
Outline • • • • • • • • • • 3
NXP overview About me EBI team overview Effective Data Science Motivation for Project Project Overview Teradata in NXP overview Solution Architecture – R / Shiny Demand Forecasting Application Benefits / Conclusion
NXP Overview
✓ #1 Automotive
✓ #1 Broad-Based MCUs1 ✓ #1 Secure Identification ✓ #1 Communications Processors ✓ #1 RF Power Transistors ✓ #1 Small Signal Discretes
Sources for market data: HIS, ABI Research, Strategy Analytics, The Linley Group 1MCU market excluding Automotive 2Excludes memory 3Pro forma revenue resulting from Dec 2015 acquisition of Freescale Semiconductor and Nov 2015 divestiture of Bipolar Power business
✓ 5th Largest semiconductor company2 ✓ 45,000 employees ✓ 11,000 engineers ✓ 9,000 patent families ✓ 50+ year history ✓ $9.8B annual revenue3
NXP SOLUTIONS
Secure, Connected Vehicle • ADAS: Radar, V2X, Vision, Fusion, network processor • Car entertainment • In-vehicle networking • Secure car access • Secure car
End-to-end Security & Privacy • Mobile transactions • E-Government • Smart bank cards • User authentication • Embedded security • Cloud & Infrastructure Security
Smart, Connected Solutions Consumer
Industrial
• Mobile audio • High-speed Interfaces • Smartphone RF • Personal health & fitness • Healthcare
• Smart home & buildings • Smart cities, smart grid • M2M, Industry 4.0 • Intelligent logistics • 4.5G/5G Networks
TODAY: 90% OF AUTO INNOVATION FROM ELECTRONICS ADAS
Security 1. SECURE INTERFACES (SE) 2. SECURE GATEWAY 3. SECURE NETWORKING 4. SECURE PROCESSING (MCU/MPU) (+1) SECURE CAR ACCESS
RADAR FRONTEND & MICROCONTROLLERS V2X COMMUNICATION BASED ON ROADLINK VISION & LIDAR PROCESSING SENSOR FUSION
POWERTRAIN & CHASSIS MICROCONTOLLERS PRESSURE/ MOTION SENSORS BATTERY MANAGEMENT DRIVERS
#1 INFOTAINMENT TUNERS SOFTWARE-DEFINED DIGITAL RADIO MULTIMEDIA PROCESSORS SOUND SYSTEM DSPs & AMPLIFIERS NFC BT PAIRING WIRELESS POWER CHARGING POWER MANAGEMENT
#1 SECURE CAR ACCESS IMMOBILIZER/ SECURITY REMOTE KEYLESS ENTRY PASSIVE KEYLESS ENTRY/ GO BI-DIRECTIONAL KEYS NFC ULTRA WIDE BAND
STANDARD PRODUCTS LOGIC POWER DISCRETES
#1 SAFETY #1 VEHICLE NETWORKING CAN/LIN/ FLEXRAY ETHERNET CENTRAL GATEWAY CONTROLLER SECURITY
Leader in Auto Analog/ RF
#1 BODY MICROCONTROLLERS POSITION/ ANGLE SENSORS SYSTEM BASIS CHIPS
Leader in Auto Processing
MICROCONTROLLERS AIRBAG ANALOG AIRBAG MICROCONTROLLERS BRAKING ANALOG BRAKING SENSORS BRAKING TIRE PRESSURE MONITORING
Leader in Auto Sensors
Outline • • • • • • • • • • 7
NXP overview About me EBI team overview Data Science in Action / Definition Motivation for Project Project Overview Teradata in NXP overview Solution Architecture – R / Shiny Demand Forecasting Application Benefits / Conclusion
About Me
•
data
• Began career as process engineer in semiconductor wafer manufacturing • Always had intense interest in data analysis • Six Sigma Black Belt under Motorola • Worked in Quality organization with focus on data analysis for customer returns • Began coding with R language about 10 years ago • Led me to the Enterprise Business Intelligence team where I lead the Advanced Analytics group BS Material Science / MS in Operations Research & Industrial Engineering (ORIE)
Source: Dr. Eric Siegel, www.predictThis.org 8
About Richard • From Alice Texas – BA in Finance from the University of Texas • Worked for Motorola/Freescale/NXP for 33 years in Finance, Accounting, Marketing, Sales, IT, Supply Chain • Currently works on Demand Forecasting team in Supply Chain • Trained employees all over the world on new software adoption • Worked on Diamond Chip winning project for Freescale supporting disaster responses in Japan following Tsunami • Fluent in Spanish and French 9
Outline • • • • • • • • • • 10
NXP overview About me EBI team overview Effective Data Science Motivation for Project Project Overview Teradata in NXP overview Solution Architecture – R / Shiny Demand Forecasting Application Benefits / Conclusion
EBI Team
11
EBI Team – Data Warehouse
NXP’s EDW Architecture principles are largely adapted from Teradata’s Solution Architecture principles, which have been proven successful for Teradata consultants globally and were defined through many years of expertise. We have adapted these, integrating our own principles developed through our years of experience and corporate culture.
12
Outline • • • • • • • • • • 13
NXP overview About me EBI team overview Effective Data Science Motivation for Project Project Overview Teradata in NXP overview Solution Architecture – R / Shiny Demand Forecasting Application Conclusion
What is a Data Scientist NSF Report (Johnstone & Roberts, 2014) • Computational aspects of carrying out a complete data analysis, including acquisition, management, and analysis of data
Deborah Nolan, Univ of California, Berkeley • A Blend/Integration of computational and statistical thinking when working with data Awesome nerds!!
Effective data science requires • Tight collaboration with business • Understand the business question you want to answer 14
Hilary Mason, Fast Forward Labs
Outline • • • • • • • • • • 15
NXP overview About me EBI team overview Teradata in NXP overview Effective Data Science Motivation for Project Solution Architecture – R / Shiny Project Overview Demand Forecasting Application Benefits / Conclusion
Background / Motivation Sales $$$
• • • •
16
Product A Product B Product C Product D …
Distributor A Distributor B Distributor C
Highly complex monthly forecasting process Thousands of products Products belong to product family hierarchies Monthly Forecasts for each product by Customer/Region
Customer xxx Customer yyy Customer zzz Customer xyz
Old Process • • • • •
manual and highly labor intensive manually pull data from warehouse manual ETL process Excel based 5 person team to support effort
Messy record keeping
Charts built in excel Changes manually noted and tracked 17
Outline • • • • • • • • • • 18
NXP overview About me EBI team overview Teradata in NXP overview Effective Data Science Motivation for Project Solution Architecture – R / Shiny Project Overview Demand Forecasting Application Benefits / Conclusion
Solution Architecture
TM TM
R is a programming language and software environment for statistical computing and graphics supported by the R Foundation for Statistical Computing. The R language is widely used among statisticians and data miners for developing statistical software and data analysis. - Wikipedia
•
Shiny is a web application framework for R created by Rstudio using WebSockets technology
•
Shiny is a reactive programming environment (event driven)
•
Combines the computational power of "R" with the interactivity of modern web pages
•
Fast bidirectional communication between the web browser and R
•
Default UI theme based on bootstrap
•
Complex interactivity / Highly flexible −
Shiny User Showcase (examples)
Outline • • • • • • • • • • 20
NXP overview About me EBI team overview Teradata in NXP overview Data Science in Action / Definition Motivation for Project Solution Architecture – R / Shiny Project Overview Demand Forecasting Application Benefits / Conclusion
Current Project Scope Billings Backlog Current model Inventory Resales Meta data
Save User Selections
Data Warehouse
Disaggregate Forecasts
ETL
Pull journals & configs from db
Build graphs and models
Apply Seasonality & user choices
Outline • • • • • • • • • • 22
NXP overview About me EBI team overview Teradata in NXP overview Data Science in Action / Definition Motivation for Project Solution Architecture – R / Shiny Project Overview Demand Forecasting Application Benefits / Conclusion
Cycle on Cycle Table User selects business group here
The Forecasting App opens to the Cycle on Cycle page. This provides a top level and comparison of prior year to current year.
23.
MAPE Accuracy Forecast accuracies are calculated and tracked.
24.
Journal of Change History Change history is maintained in a searchable, sortable journal.
25.
Seasonal Weights Seasonal decomposition can be set at different levels of the product hierarchy. This allows the forecast team flexibility in setting seasonal weights.
26.
User Configuration Some products are analyzed at the product line level of the hierarchy. The user can easily move a product family up or down the hierarchy for forecasting.
27.
Demand Forecast Dashboard This is the Demand Forecast dashboard.
The graph displays billings, backlog, inventory, resales, and several modeling options. Summary statistics for the selected product family are shown on the right.
Model selection menu. 2nd menu level with detailed options appears after making model choice. 28.
The panel on the left allows the user to experiment with different forecast models. There is a second level menu for Model Choice which is not currently displayed. Seasonality weighting can be turned on/off for the selected model.
Exponential State Space Models Exponential State Space Models In 2000, Robert Hyndman et al developed a state space framework for exponential smoothing methods which incorporates stochastic models, likelihood calculation, prediction intervals and procedures for model selection. A state space model for an N-dimensional time series yt consists of a measurement equation relating the observed data to an m-dimensional state vector at, and a Markovian transition equation that describes the evolution of the state vector over time. Holt’s Linear Method utilizes an additive treatment of the trend component. Alternatively, it is possible to treat the trend component in a multiplicative manner. Seasonal components can also be included in the models and can be either additive or multiplicative. The Exponential State Space models cover all feasible combinations of models where the error term, the trend term, and the seasonal term can be additive or multiplicative. The models are described with a 3-letter naming convention indicating additive, multiplicative, or none for each term in the model. For example, the “MAM” model indicates a multiplicative error term, additive trend term, and multiplicative seasonal term. This is equivalent to the multiplicative Holt-Winters’ seasonal model with multiplicative errors. Dampening terms can also be included in the models. The Demand Forecast R Model uses Hyndman’s exponential smoothing state space model framework. The algorithm optimizes model parameters by minimizing the prediction error for all feasible model choices (models based on the 3-letter exponential state space code). The best model of all the choices is then determined by comparing values of each model’s Akaike Information Criterion (AIC).
29
The algorithm automatically identifies the best model family with the optimal parameter settings
Detailed Model Info If interested, users can delve into the details of the forecast model.
30.
Forecast Disaggregation - part 1 The forecasts must be disaggregated down the lowest level – saleable part by customer. Several algorithms are available to the user. The disaggregation is then calculated automatically.
31.
Forecast Disaggregation – part 2 After the user makes their selections, the final disaggregated results are shown.
32.
Forecast Results Summary Model forecast results are compiled into a table which is sent downstream to the next application in the forecast process.
33.
Outline • • • • • • • • • • 34
NXP overview About me EBI team overview Teradata in NXP overview Data Science in Action / Definition Motivation for Project Solution Architecture – R / Shiny Project Overview Demand Forecasting Application Benefits / Conclusion
Benefits •
Dashboard generation of analytics provides a tremendous improvement cycle time • Generates the MGF model on first possible day instead of taking ~ 1week to build • Model reviews with MBGs can now happen 1+ weeks earlier in the month
•
Business specific seasonal indexes improve accuracy of business forecasts
•
Ability to exclude specific Product Lines improves forecast accuracy at aggregated levels
•
Easier review of model changes improves impact of model reviews
•
Journal tracking improves change management
•
Easy configuration changes enables flexible options for forecasting team
•
Greatly increases the efficiency of the forecasting process
•
Automated model generation and dashboard analytics changes the emphasis from model and data production to • • •
Focus on improved forecast mix Better handling of product transitions (new products / end of life products) More accurate forecasts
Conclusion • Data Science & Machine Learning algorithms are readily available through platforms such as R • Effective application of algorithms can provide significant business benefits • Integrating data science tools & algorithms into an easy to use web application can also deliver tremendous efficiency gains allowing teams to focus energy in improving their business rather than building reports • Tight collaboration with the business client and understanding the business problem are critical to the success of any such project
Thank You Questions/Comments Email: [email protected], [email protected]
Follow Me Twitter @
Rate This Session # with the PARTNERS Mobile App
Remember To Share Your Virtual Passes
37
Richard Arebalo Supply Chain Forecast Team, NXP Semiconductors #TDPARTNERS16
GEORGIA WORLD CONGRESS CENTER
Outline • • • • • • • • • • 2
NXP overview About me EBI team overview Effective Data Science Motivation for Project Project Overview Teradata in NXP overview Solution Architecture – R / Shiny Demand Forecasting Application Benefits / Conclusion
Outline • • • • • • • • • • 3
NXP overview About me EBI team overview Effective Data Science Motivation for Project Project Overview Teradata in NXP overview Solution Architecture – R / Shiny Demand Forecasting Application Benefits / Conclusion
NXP Overview
✓ #1 Automotive
✓ #1 Broad-Based MCUs1 ✓ #1 Secure Identification ✓ #1 Communications Processors ✓ #1 RF Power Transistors ✓ #1 Small Signal Discretes
Sources for market data: HIS, ABI Research, Strategy Analytics, The Linley Group 1MCU market excluding Automotive 2Excludes memory 3Pro forma revenue resulting from Dec 2015 acquisition of Freescale Semiconductor and Nov 2015 divestiture of Bipolar Power business
✓ 5th Largest semiconductor company2 ✓ 45,000 employees ✓ 11,000 engineers ✓ 9,000 patent families ✓ 50+ year history ✓ $9.8B annual revenue3
NXP SOLUTIONS
Secure, Connected Vehicle • ADAS: Radar, V2X, Vision, Fusion, network processor • Car entertainment • In-vehicle networking • Secure car access • Secure car
End-to-end Security & Privacy • Mobile transactions • E-Government • Smart bank cards • User authentication • Embedded security • Cloud & Infrastructure Security
Smart, Connected Solutions Consumer
Industrial
• Mobile audio • High-speed Interfaces • Smartphone RF • Personal health & fitness • Healthcare
• Smart home & buildings • Smart cities, smart grid • M2M, Industry 4.0 • Intelligent logistics • 4.5G/5G Networks
TODAY: 90% OF AUTO INNOVATION FROM ELECTRONICS ADAS
Security 1. SECURE INTERFACES (SE) 2. SECURE GATEWAY 3. SECURE NETWORKING 4. SECURE PROCESSING (MCU/MPU) (+1) SECURE CAR ACCESS
RADAR FRONTEND & MICROCONTROLLERS V2X COMMUNICATION BASED ON ROADLINK VISION & LIDAR PROCESSING SENSOR FUSION
POWERTRAIN & CHASSIS MICROCONTOLLERS PRESSURE/ MOTION SENSORS BATTERY MANAGEMENT DRIVERS
#1 INFOTAINMENT TUNERS SOFTWARE-DEFINED DIGITAL RADIO MULTIMEDIA PROCESSORS SOUND SYSTEM DSPs & AMPLIFIERS NFC BT PAIRING WIRELESS POWER CHARGING POWER MANAGEMENT
#1 SECURE CAR ACCESS IMMOBILIZER/ SECURITY REMOTE KEYLESS ENTRY PASSIVE KEYLESS ENTRY/ GO BI-DIRECTIONAL KEYS NFC ULTRA WIDE BAND
STANDARD PRODUCTS LOGIC POWER DISCRETES
#1 SAFETY #1 VEHICLE NETWORKING CAN/LIN/ FLEXRAY ETHERNET CENTRAL GATEWAY CONTROLLER SECURITY
Leader in Auto Analog/ RF
#1 BODY MICROCONTROLLERS POSITION/ ANGLE SENSORS SYSTEM BASIS CHIPS
Leader in Auto Processing
MICROCONTROLLERS AIRBAG ANALOG AIRBAG MICROCONTROLLERS BRAKING ANALOG BRAKING SENSORS BRAKING TIRE PRESSURE MONITORING
Leader in Auto Sensors
Outline • • • • • • • • • • 7
NXP overview About me EBI team overview Data Science in Action / Definition Motivation for Project Project Overview Teradata in NXP overview Solution Architecture – R / Shiny Demand Forecasting Application Benefits / Conclusion
About Me
•
data
• Began career as process engineer in semiconductor wafer manufacturing • Always had intense interest in data analysis • Six Sigma Black Belt under Motorola • Worked in Quality organization with focus on data analysis for customer returns • Began coding with R language about 10 years ago • Led me to the Enterprise Business Intelligence team where I lead the Advanced Analytics group BS Material Science / MS in Operations Research & Industrial Engineering (ORIE)
Source: Dr. Eric Siegel, www.predictThis.org 8
About Richard • From Alice Texas – BA in Finance from the University of Texas • Worked for Motorola/Freescale/NXP for 33 years in Finance, Accounting, Marketing, Sales, IT, Supply Chain • Currently works on Demand Forecasting team in Supply Chain • Trained employees all over the world on new software adoption • Worked on Diamond Chip winning project for Freescale supporting disaster responses in Japan following Tsunami • Fluent in Spanish and French 9
Outline • • • • • • • • • • 10
NXP overview About me EBI team overview Effective Data Science Motivation for Project Project Overview Teradata in NXP overview Solution Architecture – R / Shiny Demand Forecasting Application Benefits / Conclusion
EBI Team
11
EBI Team – Data Warehouse
NXP’s EDW Architecture principles are largely adapted from Teradata’s Solution Architecture principles, which have been proven successful for Teradata consultants globally and were defined through many years of expertise. We have adapted these, integrating our own principles developed through our years of experience and corporate culture.
12
Outline • • • • • • • • • • 13
NXP overview About me EBI team overview Effective Data Science Motivation for Project Project Overview Teradata in NXP overview Solution Architecture – R / Shiny Demand Forecasting Application Conclusion
What is a Data Scientist NSF Report (Johnstone & Roberts, 2014) • Computational aspects of carrying out a complete data analysis, including acquisition, management, and analysis of data
Deborah Nolan, Univ of California, Berkeley • A Blend/Integration of computational and statistical thinking when working with data Awesome nerds!!
Effective data science requires • Tight collaboration with business • Understand the business question you want to answer 14
Hilary Mason, Fast Forward Labs
Outline • • • • • • • • • • 15
NXP overview About me EBI team overview Teradata in NXP overview Effective Data Science Motivation for Project Solution Architecture – R / Shiny Project Overview Demand Forecasting Application Benefits / Conclusion
Background / Motivation Sales $$$
• • • •
16
Product A Product B Product C Product D …
Distributor A Distributor B Distributor C
Highly complex monthly forecasting process Thousands of products Products belong to product family hierarchies Monthly Forecasts for each product by Customer/Region
Customer xxx Customer yyy Customer zzz Customer xyz
Old Process • • • • •
manual and highly labor intensive manually pull data from warehouse manual ETL process Excel based 5 person team to support effort
Messy record keeping
Charts built in excel Changes manually noted and tracked 17
Outline • • • • • • • • • • 18
NXP overview About me EBI team overview Teradata in NXP overview Effective Data Science Motivation for Project Solution Architecture – R / Shiny Project Overview Demand Forecasting Application Benefits / Conclusion
Solution Architecture
TM TM
R is a programming language and software environment for statistical computing and graphics supported by the R Foundation for Statistical Computing. The R language is widely used among statisticians and data miners for developing statistical software and data analysis. - Wikipedia
•
Shiny is a web application framework for R created by Rstudio using WebSockets technology
•
Shiny is a reactive programming environment (event driven)
•
Combines the computational power of "R" with the interactivity of modern web pages
•
Fast bidirectional communication between the web browser and R
•
Default UI theme based on bootstrap
•
Complex interactivity / Highly flexible −
Shiny User Showcase (examples)
Outline • • • • • • • • • • 20
NXP overview About me EBI team overview Teradata in NXP overview Data Science in Action / Definition Motivation for Project Solution Architecture – R / Shiny Project Overview Demand Forecasting Application Benefits / Conclusion
Current Project Scope Billings Backlog Current model Inventory Resales Meta data
Save User Selections
Data Warehouse
Disaggregate Forecasts
ETL
Pull journals & configs from db
Build graphs and models
Apply Seasonality & user choices
Outline • • • • • • • • • • 22
NXP overview About me EBI team overview Teradata in NXP overview Data Science in Action / Definition Motivation for Project Solution Architecture – R / Shiny Project Overview Demand Forecasting Application Benefits / Conclusion
Cycle on Cycle Table User selects business group here
The Forecasting App opens to the Cycle on Cycle page. This provides a top level and comparison of prior year to current year.
23.
MAPE Accuracy Forecast accuracies are calculated and tracked.
24.
Journal of Change History Change history is maintained in a searchable, sortable journal.
25.
Seasonal Weights Seasonal decomposition can be set at different levels of the product hierarchy. This allows the forecast team flexibility in setting seasonal weights.
26.
User Configuration Some products are analyzed at the product line level of the hierarchy. The user can easily move a product family up or down the hierarchy for forecasting.
27.
Demand Forecast Dashboard This is the Demand Forecast dashboard.
The graph displays billings, backlog, inventory, resales, and several modeling options. Summary statistics for the selected product family are shown on the right.
Model selection menu. 2nd menu level with detailed options appears after making model choice. 28.
The panel on the left allows the user to experiment with different forecast models. There is a second level menu for Model Choice which is not currently displayed. Seasonality weighting can be turned on/off for the selected model.
Exponential State Space Models Exponential State Space Models In 2000, Robert Hyndman et al developed a state space framework for exponential smoothing methods which incorporates stochastic models, likelihood calculation, prediction intervals and procedures for model selection. A state space model for an N-dimensional time series yt consists of a measurement equation relating the observed data to an m-dimensional state vector at, and a Markovian transition equation that describes the evolution of the state vector over time. Holt’s Linear Method utilizes an additive treatment of the trend component. Alternatively, it is possible to treat the trend component in a multiplicative manner. Seasonal components can also be included in the models and can be either additive or multiplicative. The Exponential State Space models cover all feasible combinations of models where the error term, the trend term, and the seasonal term can be additive or multiplicative. The models are described with a 3-letter naming convention indicating additive, multiplicative, or none for each term in the model. For example, the “MAM” model indicates a multiplicative error term, additive trend term, and multiplicative seasonal term. This is equivalent to the multiplicative Holt-Winters’ seasonal model with multiplicative errors. Dampening terms can also be included in the models. The Demand Forecast R Model uses Hyndman’s exponential smoothing state space model framework. The algorithm optimizes model parameters by minimizing the prediction error for all feasible model choices (models based on the 3-letter exponential state space code). The best model of all the choices is then determined by comparing values of each model’s Akaike Information Criterion (AIC).
29
The algorithm automatically identifies the best model family with the optimal parameter settings
Detailed Model Info If interested, users can delve into the details of the forecast model.
30.
Forecast Disaggregation - part 1 The forecasts must be disaggregated down the lowest level – saleable part by customer. Several algorithms are available to the user. The disaggregation is then calculated automatically.
31.
Forecast Disaggregation – part 2 After the user makes their selections, the final disaggregated results are shown.
32.
Forecast Results Summary Model forecast results are compiled into a table which is sent downstream to the next application in the forecast process.
33.
Outline • • • • • • • • • • 34
NXP overview About me EBI team overview Teradata in NXP overview Data Science in Action / Definition Motivation for Project Solution Architecture – R / Shiny Project Overview Demand Forecasting Application Benefits / Conclusion
Benefits •
Dashboard generation of analytics provides a tremendous improvement cycle time • Generates the MGF model on first possible day instead of taking ~ 1week to build • Model reviews with MBGs can now happen 1+ weeks earlier in the month
•
Business specific seasonal indexes improve accuracy of business forecasts
•
Ability to exclude specific Product Lines improves forecast accuracy at aggregated levels
•
Easier review of model changes improves impact of model reviews
•
Journal tracking improves change management
•
Easy configuration changes enables flexible options for forecasting team
•
Greatly increases the efficiency of the forecasting process
•
Automated model generation and dashboard analytics changes the emphasis from model and data production to • • •
Focus on improved forecast mix Better handling of product transitions (new products / end of life products) More accurate forecasts
Conclusion • Data Science & Machine Learning algorithms are readily available through platforms such as R • Effective application of algorithms can provide significant business benefits • Integrating data science tools & algorithms into an easy to use web application can also deliver tremendous efficiency gains allowing teams to focus energy in improving their business rather than building reports • Tight collaboration with the business client and understanding the business problem are critical to the success of any such project
Thank You Questions/Comments Email: [email protected], [email protected]
Follow Me Twitter @
Rate This Session # with the PARTNERS Mobile App
Remember To Share Your Virtual Passes
37
