Delivering Personalized Experiences using the Power of Data
1 like588 views
The document discusses how a retailer used data and machine learning to address declining sales in their Seattle market. They built a sales driver model that explained 89% of the variations in sales. This identified that a large competitor price gap was negatively impacting sales. The retailer analyzed scenarios showing that reducing the price gap by 15% could lead to $4,500 more in profit. Increasing social media engagement or advertising spending could also boost sales and profits. The recommended actions were to decrease the price gap and increase marketing to reverse the sales declines.
Delivering Personalized Experiences using the Power of Data
- 2. Listening to the voice of the customer 0 5 10 15 20 25 30 1995 1996 1997 1998 1999 2000 2001 2002 Sales Forecast in 1999 Traditional Cameras Digital Cameras 0 2 4 6 8 10 12 14 16 18 20 1995 1996 1997 1998 1999 2000 2001 2002 What Really Happened Traditional Cameras Digital Cameras
- 3. The Challenges of Limited Data Person 1 Person 2 Born 1948 Born 1948 Grew up in England Grew up in England Married Twice Married Twice Two Children Two Children Successful in Business Successful in Business Wealthy Wealthy Spend Winter Holidays in the Alps Spend Winter Holidays in the Alps Likes Dogs Likes Dogs Like Music Like Music
- 4. Retail has a multitude of devices that generate petabytes of potential insights Monitoring and mining social media data enables retailers to enhance customer insights Open data sources and internal sources enable retailers to better understand customers Democratization of data
- 6. Cortana Intelligence Information Management Big Data Stores Machine Learning & Analytics Visualization Democratization of Tools
- 8. Furnishings importer delighted customers with the right offers Challenge • Give customers a better experience and selection. • Understand what customers are looking for based on online search. Strategy • Combined online and in-store transactional and behavioral data to predict what products customers would be most likely to purchase next. Results • Provided more personalized choices to customers. • Enabled targeted campaigns. • Improved inventory management. “We are continually getting better at identifying what our customer wants, using Microsoft Azure Machine Learning and resulting data insights.” — Sharon Leite, EVP Sales and Customer Experience, Pier 1 Imports
- 9. 10
- 10. Foodservice retailer built brand loyalty with a cloud-based mobile app Challenge • Improve relationships with customers. • Build upon a successful brand app to reach more customers. • Generate upsales and cross-sales. Strategy • Added an analysis and insights system based on Microsoft Azure. • Gathered data on customers’ behaviors and preferences through their app use. Results • Expanded app use to millions of customers in five countries. • Enabled personal, relevant, contextual content that adds value for the users. • Improved customer loyalty.
- 13. Random Selection Offers given out randomly, but you can see each segment prefers a different offer
- 14. Pre-learned ML Model Activated ML gives the right segment the right offer to maximise redemptions
- 15. Impressions The ML system gave Cheeseburgers to Students and Sundaes to the Employed segments
- 16. Adaptive MWT MWT adapts to the new condition it hasn’t seen before. Everyone gets Sundaes, sales are back up!
- 17. Back to Cloudy - Redemptions Summary MWT can manage both cloudy and sunny conditions, sales lift is maintained
- 18. Impressions Summary ML only knows how to change based on segment, MWT changes with a new variable
- 33. Big data boosts business for beverage company “Over time, this advanced analytics solution with its statistical approach to big data will transform the way we take business decisions through all business processes” - Ruben Dario Torres Martinez, Arca Continental, IT Manager Solution • Provide executive decision makers with an understanding of what drives sales of their beverage products. • Understand influencers & answer “Why did/do we sell?” • Explain sales variance so that decision makers know exactly what caused the variance and what needs to be optimized • Analyze SKU profitability and recommend which SKU’s should be discontinued Results • improved way to measure the success of marketing efforts. • can easily view a graph of the market DNA, which explains how each different market responds to each variable. • revenue growth managers can now slice and dice the effects of multiple variables, practically in real time
- 34. Key Driver Analysis • It's important to identify and understand the drivers of key business outcomes • Which aspects of your service influence how likely a customer will be to recommend you to others. • A key driver analysis investigates the relationships between potential drivers and customer behavior • A key driver analysis is often performed using multiple linear regression to model the primary outcome as a linear combination of the potential drivers.
- 36. 37 Sales for Northwest Territory retailer Problem: Seattle is not selling to forecast 1 2
- 37. 38 When we drill down to Seattle, we can see a problem in soft drinks Click and see further details of Seattle sales 1 2
- 38. 39 Sales driver analysis – build a model that explains what drives sales Sales delta analysis – use the model to see problems 3. How can we fix sales? – apply the model to fix the problems 21 3
- 39. 40 25.6% variations explained Internal transaction and marketing data include variables as: - Stock Up - Price Elasticity - Radio Advertising - TV Advertising - SKU presence Transaction dataset in AML experiment 1 2 3
- 40. 41 Variations explained improves to near 50% External weather, demographic, and competitor data include variables as: - Temperature - Precipitation - Household size - Annual Income - Competitor Price Gap Transaction dataset in AML experiment External dataset enters the model in AML experiment 2 1 3 4
- 41. 42 IoT dataset enters the model in AML experiment Variations explained improves to 89% New IoT, research and online activity data include variables as: - Survey research - Web traffic - Social media traffic - Mobile traffic - Store traffic - Shelf traffic Transaction dataset in AML experiment External dataset enters the model in AML experiment 2 4 1 3 5
- 42. 43 Monthly ∆ by sales driver Let’s first zero in on the sales impact of price gaps, as they are the biggest problem Competitor price gap caused 7,598 less units sold than previous month Click one of the controllable variables to see what would happen if we take some actions 21 3 4
- 43. 44 See the impact on physical sales if we reduce the price gap by different levels See the impact on profit if we reduce the price gap by different levels. When it is reduced by 15%, we would be able to achieve 4.5K incremental profit. Select competitor price gap as it is a controllable variable It would be recommended to decrease the competitor price gap 2 1 3 4
- 44. 45 See the impact on physical sales if we increase social media engagement by different levels See the impact on profit if we increase social media engagement by different levels. When it is increased by 20%, we would be able to achieve 12.6K incremental profit. Select Social Media Engagement as it is a controllable variable It would be recommended to increase social media engagement 2 1 3 4
- 45. 46 See the impact on physical sales if we increase advertising by different levels See the impact on profit if we increase advertising by different levels. When it is increased by 20%, we would be able to achieve 7.1K incremental profit. Select Own Brand Advertising as it is a controllable variable It would be recommended to increase our own brand advertising 2 1 3 4
- 46. 47 See the overall impact on physical sales if we take the recommended actions The sales will continue sliding down if no actions are taken 1 3 Within the budget constraints, select the recommended actions 2
Editor's Notes
- #3: On the left is forecast from an electronics retailer. In 1999, the retailer wanted to determine what products to continue investing in and what products to stop investments for. This chart shows two of the categories: film cameras and digital cameras. The chart on the left shows the result of regression for forecasting sales over the next few years. The chart on the right shows what actually happened. This is one of the examples indicating that no amount of large volumes of data can help predict the future if it is proprietary data. Gartner talks about the 3 Ps of Data: Proprietary, public and purchased. When combined with external data, you can get a better picture of what customers want.
- #6: In the retail ecosystem, we have devices in the store for RFID, NFC, Bluetooth, Wireless data, Video Analytics, POS data, Kinect in the Store collecting huge amount of data on a daily basis. This information can be stored by retailers and used for analyzing customer behavior, predicting demand, assortment, store and shelf layout, upselling, customer recommendations, web page personalization, delivering personalized ads. There is an unprecedented amount of data about people, places, products, companies, brands, and pretty much anything we can think of. Companies that can mine this treasure trove of data and glean insights to gain a huge competitive advantage
- #8: The Cortana analytics suite provides information management, big data stores, machine learning and analytics tools, and visualization options to extract value from data. There’s a host of technologies to fill each of these roles.
- #9: First let’s start by level setting what we mean by Advanced Analytics. We often get asked whether Advanced Analytics is just “BI” with fancier branding. This chart helps to illustrate why that is not the case. Put simply, BI is a tool that is designed to show you what has happened so you can make your own decisions about what to do next. The next level – Advanced Analytics – allows you to take that data to the next level by having the computer predict what will happen next. This is more accurate than a human can ever hope to be, as the computer can reason over far more variables than a human can on a BI dashboard. But predictive is only the first step, the next step is once you are accurately predicting the future you can program the computer further to anticipate those occurrences and react accordingly.
- #10: 9
- #12: 11
- #15: [Click the Play Button] Offers are given out randomly (equally) to the Student and Employed customers It becomes obvious that Students prefer Cheeseburgers and Employed prefer Sundaes So targeting the right segment with the right offer should improve results
- #16: [Activate ML] Offers are given out now based on previously known behaviour Redemptions are now going up on both offers. Great!
- #17: [Show Impressions Tab] Students – prefer Cheeseburgers so given more of these Employed – prefer Sundae so given more of these
- #18: MWT is adaptive to an unexpected change in behaviour. [Activate MWT] Total Redemptions come back up again because the system has adapted to change in preference when it is hot!
- #19: Cloudy is activated again with the ML
- #35: https://customers.microsoft.com/Pages/CustomerStory.aspx?recid=26082
- #36: With a 'key driver analysis', statistical modelling can be used to quantify the relationships between multiple variables. This can help you to understand what drives customer behaviour and ultimately how to improve your performance.