SlideShare a Scribd company logo

Data science Applications in the Enterprise

•
•
Srinath Perera
Srinath Perera

It explains how enterprise works and common themes for enterprise use cases, including AI use cases within an iPaaS

1 of 20
Download to read offline
Data Science Use Cases in the Enterprise Srinath Perera Chief Architect, WSO2, Apache Member
Context: Understanding Enterprise (ROI) ● It is about Money: long-term Money. ○ If you are looking to make a million once, sometimes, you can get away with exploitation. ○ If you are looking to make a billion every year, you have to care about customers, brand, employees as well as the environment you are operating in ○ E.g., Indra Nooyi and her effort to move Pepsi to healthy food. ● It is a Strategic environment where enterprises compete. ○ “If you know the enemy and know yourself, you need not fear the result of a hundred battles. ” ― Sun Tzu, The Art of War.
Context: Highly valued Outcomes ● Efficiency, Savings ● Improving Customer Experience ● Finding new markets, understanding markets ● Forecasts, Prediction ● Automation and Decision Support I skate to where the puck is going to be, not where it has been. ---Wayne Gretzky
● Examples ○ The effort by the US to use sensor and data analysis to stop infiltration through Ho Chi Minh Trail in 70s ○ Even Nate Silver got Trump's victory wrong ● Reasons ○ History is not always representative of the future (e.g., Trump Elections) ○ Complex systems ( highly interconnected systems where one or few players can significantly change the outcomes) ○ Highly competitive situations such as stock Markets ■ Predictable at stable times, but not with shock ○ Average is affected dramatically by rare events (e,g, Covid) ■ Data can determine "average" outcomes with great accuracy ○ Not enough data or data do not capture critical aspects Nevermind the Press, Data Science does not always work
Use Cases @ Enterprise ● Efficiency, Self Awareness, and Forecasts ● Optimizing the sales funnel ● Predictive Maintenance ● Improving Customer Experience ● Product Use cases from a real-world iPaaS ● Finding new markets, understanding markets, Competitor Analysis ● https://sparktoro.com/ - Instantly discover what your audience reads, watches, listens to, and follows. ● Automate mundane tasks and let people focus on what they are good at ● Automation and Task Assistant Systems ● Decision support systems Often needs Explainability too
Efficiency: Optimize the Sales Funnel ● Each enterprise has a funnel like this ( names may be different) ● KPIs support decisions ● Examples: ○ conversion rates, dropoff - to find bottlenecks ○ cost per conversion - find activities that work well ○ Time spend on each stage ○ Forecasts ○ A/B testing optimizes
Efficiency: Predictive Maintenance ● Often breakdowns have high costs ● We do preventive maintenance to avoid that, but it leaves significant money on the table ● Use telemetry data to predict breakdowns ● We need to manage risk against false negatives (e.g., cost to give customer 100$)
Efficiency: Churn Prediction ● Even small churn compounds significantly to reduce topline, and create negative word of mouth. ● How is the user using the product? ● Has he given up? ● Are there complaints? ● Is there anything we can do if we know before? Need to think through the full story - Ask “so what” until you see $$
User Experience: Understanding Choreo
Choreo Use Cases and Challenges ● Can collect data about everything, clicks, messages, logs etc ● The focus is using AI to improve user experience ● The system will have 10s of thousands of users ○ We can’t run a model per user ● Some use cases have limited data ● The specific user would not have enough data initially, so we have a cold start problem ● Some use cases require personalization
User Experience: Forecasting Performance ● Performance feedback while you write code ● API, service, database calls dominate performance ● Use historical data about each API, service, database call and fit Machine Learning models ● Use queuing theory to model the throughput and latency
Getting a Model to Production is Complicated ● Data Collection ● Model training ● Model deployment and integrating the model into the user experience ○ Acting on results ● Getting user feedback ● Evaluating and improving models
User Experience: Automatic Data Mapping ● Programming with APIs need us to map data between two API calls ( and two systems) ● Automatic data mapping suggest mapping between two data types ● It can maps data types it has never seen
User Experience: Anomaly and Root Cause Prediction ● Detecting Performance anomalies in the system ● The goal is to detect and performance problems and notify the users and supporting them in troubleshooting ● We started with several states of the art papers and eventually beat them ○ 90% precision and 50% recall vs. 98% vs. 81% recall ● Working on attributing anomalies to parts of the system and providing root cause predictions 42
Understanding Markets: Sparktorro
Automation: Extracting information from Images/ Video ● Vidado.ai Using OCR to digitize Data RPA does not work well with paper ● Icetana.com - decision support for video surveillance ● www.dataminr.com detects high impact events from public data ○ E.g., Brand risk, disease outbreaks, potential new stories
Automation: Competitive Adjustments ● Common use cases are adjusting the price ● This leads to curious cases when bots are on both sides A good rule of thumb is to remember AI vs. AI does not work well.
Automation: Automate Mundane Tasks ● Works on top salesforce ● Suggest next Action ● Provides templates for actions ● Full context, connect all information ● Benchmark performance
Parting Thoughts ● If you plan to solve organizational problems with data science, you need to understand how it works and speak their language. ● Make sure there is enough data ● Think through the full lifecycle, including economics (e.g., Choreo) and explain ● Model deployment, evaluation, integration to customer, and evolution is complex ● Harder to build per user custom models, better if you can create value against existing data models and integrate as SaaS Learn to see where Data Science works, but learn to see where it does not also!!
Questions?

More Related Content

Data science Applications in the Enterprise

  • 1. Data Science Use Cases in the Enterprise Srinath Perera Chief Architect, WSO2, Apache Member
  • 2. Context: Understanding Enterprise (ROI) ● It is about Money: long-term Money. ○ If you are looking to make a million once, sometimes, you can get away with exploitation. ○ If you are looking to make a billion every year, you have to care about customers, brand, employees as well as the environment you are operating in ○ E.g., Indra Nooyi and her effort to move Pepsi to healthy food. ● It is a Strategic environment where enterprises compete. ○ “If you know the enemy and know yourself, you need not fear the result of a hundred battles. ” ― Sun Tzu, The Art of War.
  • 3. Context: Highly valued Outcomes ● Efficiency, Savings ● Improving Customer Experience ● Finding new markets, understanding markets ● Forecasts, Prediction ● Automation and Decision Support I skate to where the puck is going to be, not where it has been. ---Wayne Gretzky
  • 4. ● Examples ○ The effort by the US to use sensor and data analysis to stop infiltration through Ho Chi Minh Trail in 70s ○ Even Nate Silver got Trump's victory wrong ● Reasons ○ History is not always representative of the future (e.g., Trump Elections) ○ Complex systems ( highly interconnected systems where one or few players can significantly change the outcomes) ○ Highly competitive situations such as stock Markets ■ Predictable at stable times, but not with shock ○ Average is affected dramatically by rare events (e,g, Covid) ■ Data can determine "average" outcomes with great accuracy ○ Not enough data or data do not capture critical aspects Nevermind the Press, Data Science does not always work
  • 5. Use Cases @ Enterprise ● Efficiency, Self Awareness, and Forecasts ● Optimizing the sales funnel ● Predictive Maintenance ● Improving Customer Experience ● Product Use cases from a real-world iPaaS ● Finding new markets, understanding markets, Competitor Analysis ● https://sparktoro.com/ - Instantly discover what your audience reads, watches, listens to, and follows. ● Automate mundane tasks and let people focus on what they are good at ● Automation and Task Assistant Systems ● Decision support systems Often needs Explainability too
  • 6. Efficiency: Optimize the Sales Funnel ● Each enterprise has a funnel like this ( names may be different) ● KPIs support decisions ● Examples: ○ conversion rates, dropoff - to find bottlenecks ○ cost per conversion - find activities that work well ○ Time spend on each stage ○ Forecasts ○ A/B testing optimizes
  • 7. Efficiency: Predictive Maintenance ● Often breakdowns have high costs ● We do preventive maintenance to avoid that, but it leaves significant money on the table ● Use telemetry data to predict breakdowns ● We need to manage risk against false negatives (e.g., cost to give customer 100$)
  • 8. Efficiency: Churn Prediction ● Even small churn compounds significantly to reduce topline, and create negative word of mouth. ● How is the user using the product? ● Has he given up? ● Are there complaints? ● Is there anything we can do if we know before? Need to think through the full story - Ask “so what” until you see $$
  • 10. Choreo Use Cases and Challenges ● Can collect data about everything, clicks, messages, logs etc ● The focus is using AI to improve user experience ● The system will have 10s of thousands of users ○ We can’t run a model per user ● Some use cases have limited data ● The specific user would not have enough data initially, so we have a cold start problem ● Some use cases require personalization
  • 11. User Experience: Forecasting Performance ● Performance feedback while you write code ● API, service, database calls dominate performance ● Use historical data about each API, service, database call and fit Machine Learning models ● Use queuing theory to model the throughput and latency
  • 12. Getting a Model to Production is Complicated ● Data Collection ● Model training ● Model deployment and integrating the model into the user experience ○ Acting on results ● Getting user feedback ● Evaluating and improving models
  • 13. User Experience: Automatic Data Mapping ● Programming with APIs need us to map data between two API calls ( and two systems) ● Automatic data mapping suggest mapping between two data types ● It can maps data types it has never seen
  • 14. User Experience: Anomaly and Root Cause Prediction ● Detecting Performance anomalies in the system ● The goal is to detect and performance problems and notify the users and supporting them in troubleshooting ● We started with several states of the art papers and eventually beat them ○ 90% precision and 50% recall vs. 98% vs. 81% recall ● Working on attributing anomalies to parts of the system and providing root cause predictions 42
  • 16. Automation: Extracting information from Images/ Video ● Vidado.ai Using OCR to digitize Data RPA does not work well with paper ● Icetana.com - decision support for video surveillance ● www.dataminr.com detects high impact events from public data ○ E.g., Brand risk, disease outbreaks, potential new stories
  • 17. Automation: Competitive Adjustments ● Common use cases are adjusting the price ● This leads to curious cases when bots are on both sides A good rule of thumb is to remember AI vs. AI does not work well.
  • 18. Automation: Automate Mundane Tasks ● Works on top salesforce ● Suggest next Action ● Provides templates for actions ● Full context, connect all information ● Benchmark performance
  • 19. Parting Thoughts ● If you plan to solve organizational problems with data science, you need to understand how it works and speak their language. ● Make sure there is enough data ● Think through the full lifecycle, including economics (e.g., Choreo) and explain ● Model deployment, evaluation, integration to customer, and evolution is complex ● Harder to build per user custom models, better if you can create value against existing data models and integrate as SaaS Learn to see where Data Science works, but learn to see where it does not also!!