all you wanted to know about analytics in e commerce- amazon, ebay, flipkart
TRANSCRIPT
Retail Analytics – E-Commerce
Group 9IIM LucknowAnju R Gothwal PGP28250Animesh PGP29181Malory Ravier IEP15003Mayank Khatri PGP29220Richa Narayan PGP29207Shashank Singh Chandel
PGP29493
Tushar Gupta PGP29197
AGENDA1) RETAIL ANALYTICS
Industry Practice – Types of Analytics Information Providers
2) ANALYTICS IN ECOMMERCE INDUSTRY Web analytics – basic metrics, top tools Data Handling – Software in Trend- HADOOP Major Analytics Applications in Ecommerce
3) ANALYTICS IN ECOMMERCE COMPANIES Amazon Flipkart Ebay
4) RESEARCH PAPER STUDYCustomer Segmentation and Promotional Offers RFM Lifetime Value
5) RECOMMENDATIONS
1) RETAIL ANALYTICS
2) ANALYTICS IN ECOMMERCE INDUSTRY
3) ANALYTICS IN ECOMMERCE COMPANIES
4) RESEARCH PAPERS STUDY
5) RECOMMENDATIONS
Industry Practices - Types of Analytics – RETAIL ANALYTICS
CUSTOMER ANALYTICSCustomer AcquisitionCustomer LoyaltyBehavioral SegmentationGeneral Merchandiser - TESCO
MARKETING ANALYTICSMarketing MixBrand HealthMultichannel Campaign OptimizationApparel Chain – SEARS CANADA
MERCHANDISING AND PLANNINGShelf space optimizationProduct PricingStore Location DecisionsFashion Retail – BELK
RISK ANALYTICSDetecting Fraudulent activityDetecting Process ErrorsDetecting Store TheftOnline Retailer - AMAZON
DEMAND AND SUPPLY CHAINInventory PlanningDemand ForecastingProduct Flow OptimizationDepartment Store – METRO GROUP
PREDICTIVE ANALYTICSDetermining Customer LTVRevenue forecastingProduct RecommendationsTrend Analysis
Information Providers -RETAIL ANALYTICS
Market research companies providing retail intelligence IRI: Information Resource Inc. Leader in delivering powerful market and shopper information, predictive analysis and the
foresight Keeps systems on big retailers, collect info, sell data and trends, simplifies and supports
manufacturers and allServices Provided
Market, consumer and shopper intelligence Retail tracking information Online and offline marketing ROI strategy and effectiveness Predictive analytics and modeling Enterprise-class business intelligence software platforms and solutions Pricing, trade promotion and brand portfolio maximization Store level and merchandising insights Strategic consulting and thought leadership
AC Neislen: Another Player in the arena
1) RETAIL ANALYTICS
2) ANALYTICS IN ECOMMERCE INDUSTRY
3) ANALYTICS IN ECOMMERCE COMPANIES
4) RESEARCH PAPERS STUDY
5) RECOMMENDATIONS
Web Analytics – E Commerce Web Analytics involves mainly studying consumer behavior and traffic online Ecommerce applications – study consumer purchase to boost sales, attract more customers,
build brand BASIC METRICS TO TRACK
TOP ANALYTICS TOOLS FOR ECOMMERCE:
TOOL CAPABILITIES APPLICATIONSGoogle Analytics Monitors traffic from social media, emails Measures effectiveness of marketing
programAdobe Site Catalyst Real time segmentation Increase checkout conversion ratesIBM Corementrics Enterprise level Solution, provides
actionable information Know how website affects visitors, advertisement ROI
Webtrends Digital marketing intelligence Increase Conversions, Search and social advertising, visitors segmentation and scoring
MEASURE DESCRIPTIONVisitors No of visitors tells how business is doingPage Views Maximum viewed Tells the popular contentReferring Sites Tells the interests of customerBounce Rates Tells why people leave the siteKeywords and Phrases Tells about customers requirements
DATA HANDLING - Software in trend - HADOOP HADOOP: Open source software project Accomplishes two tasks: massive data storage , faster processing
ADVANTAGES:• Handle huge amount of data - great volumes and varieties – esp. from social media and
automated sensors• Low cost - the open-source framework is free and uses commodity hardware to store large
quantities of data• Computing power - distributed computing model can quickly process very large volumes
of data• Scalability - can easily grow your system simply by adding more nodes. Little
administration is required.• Storage flexibility - can store as much data as you want and decide how to use it later.• Inherent data protection and self-healing capabilities - Data and application
processing are protected against hardware failure. If a node goes down, jobs are automatically redirected to other nodes to make sure the distributed computing does not fail. And it automatically stores multiple copies of all data.
Other S/W involved – Tableau, TeraData etc.
Major Analytics applications – E Commerce
• Personalization helps to increase conversion rates• HBR say personalization increases ROI by 8 to 10 times• Ex: Gilt Group ecommerce company uses targeted emails to give
offers matching customer searchPersonalization
• Analyzing buying pattern to make online purchase seamless process• Optimizing services like customer call
Improving Customer Experience
• Develop models for real time pricing of millions of SKU’s• Parameters considers are competition, inventory, required margins
etc.Pricing
• Used to predict consumer behavior ex. Used by Amazon to predict customer purchase
• Vendors like Atterix, SAS, Lattice provide such servicesPredictive Analysis
• Supply chain intelligence for real time communication between different stakeholders like vendors, warehouses, customer etc.
• Helps achieve faster delivery, higher fulfillment, low inventoryManaging Supply
Chain
Platforms for Predictive Analytics Platforms
Predictive Tools that integrate with e-commerce platform
• Tools and Plugins• No headache of integration• Springbot, Custora, Canopy
Labs• $199-$300/month
Open Source Product
• Suitable for an analytics team
• Hiring the right skilled resources a challenge
• R, KNIME, PredicitionIO• Free
Full Featured Site
• Most functionality• Point solutions for
various areas• Consulting options
provided• SAS, SAP, Predixion• Approx. $10,000 for
single user license
1) RETAIL ANALYTICS
2) ANALYTICS IN ECOMMERCE INDUSTRY
3) ANALYTICS IN ECOMMERCE COMPANIES
4) RESEARCH PAPERS STUDY
5) RECOMMENDATIONS
Analytics Practices – Amazon
ATTRIBUTES PRACTICESIn-house/ outsourced All analytics done in- house
Major Tools Open Source Tweaked to Amazon’s needs Amazon uses its native analytics platform – Hadoop with Elastic Map
Reduce and S3 database Amazon also uses Glacier for archiving data and Kinesis for stream
processing of high volume real time data streams
Major Metrics One of the most Metrics driven company almost everything measured and evaluated
Analytics major heads 1. Customer Analytics2. Seller Analytics3. Trust Analytics4. Supply Chain Analytics
Notable attributes They also monetize the platform by offering it to other companies
Customer Analytics - AmazonPRODUCT RECOMMENDATIONS Hybrid Recommender Systems – a mix of both content and collaborative filtering Main metrics analyzed are –
1) Customer’s past purchases2) Items customers have rated and liked3) Purchases compared to similar purchase by other competitors4) Items in virtual shopping carts
Generates approximately 29% sales from recommendationsCUSTOMER SERVICE No attempts to up sell over customer service calls Data network allows Amazon to call the customer in under a minute after he places a
service request Reports and Views are extensively used to have selected customer information on screen Customers are only last name and address to fetch all their data Customer service reps are well informed due to big data analytics; leads to
individualized and human
Seller Analytics - Amazon Amazon treats its over 2 million sellers as its customers, provide all the technology and
services sellers need to run their business Personalization with sellers, proactive, data driven recommendations to each and every seller
on the platform Tens of millions of recommendations to entire seller base in a day through emails and the
native platform ‘Seller Central’ Business reports are also available for purchase for in depth insights Examples of some recommendations 1) Almost out of stock – Recommendation on how much to add to inventory based on forward looking demand for the product adjusted for seasonality and festivals 2) Search Results – When customer encounters no search results or results of low relevancy, the results are surfaced back to the seller and recommend to carry products customers
are looking for 3) Fulfillment by Amazon – Recommendations based on the characteristics of how difficult the
products are to fulfill 4) Performance Feedback – Metrics on satisfying customers, serving their needs and getting products to them fast and easily 5) Sharpness of Pricing – Surface up the sellers of all different products a seller is carrying on Amazon, determine whether it makes sense to lower prices for customers
Supply Chain Analytics - Amazon Monitors, tracks and secures 1.5 Billion items laying around 200 fulfillment centers 50 million updates are made to the database per week Entire data is crunched every 30 minutes and the results are transmitted to all the terminalsINVENTORY CONTROL Amazon uses ‘non-stationary stochastic model’ for optimizing inventory Has developed algorithms for joint and coordinated replenishments Algorithms also support fulfillment, sourcing and capacity decisions Forecasting is done at an SKU level for each fulfillment centerDEMAND Analytics on customer wish lists, gift registries and pre-orders to anticipate demand apart from usual
forecasting techniques Wish lists are publicly visible, software crawls wish lists to aggregate data about customer demandLOGISTICS Patented ‘Method and System for Anticipatory Package Shipping’ Anticipates customer needs before they express them Analyzes a) Customer Ordering History d) Feedbacksb) Wish-lists e) Searchesc) Average Shopping Cart Content f) How long a cursor hovers over a product page Results in very fast delivery, sends off packages to a shipping hub or a truck near the customer’s address
and waits to receive a go ahead to deliver
Control and Trust - Amazon
CREDIT CARD FRAUD DETECTION Uses a scoring approach to identify the most likely fraud situations Some of the situations analyzed are
1) Purchase of easily resold goods on gray market such as electronics
2) Use of different billing and shipping address3) Use of fastest shipping option
WAREHOUSE THEFTS Constantly Updates database of high ticket, most likely to be stolen
items
Software Used - Flipkart QLIKVIEW – Parent Company: Qlik, based at Pennsylvania
Improved Inventory Management tool to optimize Stock Levels
CHALLENGES Integrate Complex Data from disparate sources Deliver Analytical data to staff in various departments Improve inventory utilization
Initial Usage: Open source Business Intelligence (BI) but the problem faced – ScalabilityADVANTAGES Provided transparent and up-to-date information for analysis Embedded data-driven decision making at Flipkart Improved Inventory Utilization
Information gathered over telephonic conversation with IIM L alumnus working in Flipkart
Software Used - FlipkartBIGFOOT - Computerized Maintenance Management Software (CMMS) 1) Managing the maintenance operational needs of organizations 2) Bigfoot CMMS' full functionality paired with its intuitive design allows to implement the solution
and get results quickly.
KEY FUNCTIONS preventive and predictive maintenance inventory management, work order asset, and equipment management purchasing built-in reporting and analysisADVANTAGES The system can support any number of facilities and multiple languages Increases staff productivity and reduce maintenance costs today Support integration with other systems like ERP, bar code, custom interfaces, advanced reporting
solutions building Automation solutions, and Active Directory Bigfoot CMMS can be configured for different user types, security settings, site and location
details, and user access settings
Analytics Practices – eBay
ATTRIBUTES PRACTICESIn-house/ outsourced Most of the analytics done by the in- house analytics team
Few practices are outsourcedMajor Tools SAS
ExcelMajor Metrics Exit Rate, Transactional and operational metrics
Analytics major heads 1. Buyer Analytics2. Seller Analytics3. Trust Analytics
Notable attributes Analytics used by Marketing team for segmentation of customers or predicting churn rate for customers is handled differently
AB Testing for measuring efficiency of new feature
Information gathered over telephonic conversation with IIM L alumnus working in eBay
Major Metrics - eBayEXIT RATE
Which is the page which marks the termination of user’s session Find the dissatisfying elements of the page if the page is not meant for user to exit the
session Improve the elements from pages in order to increase the length of session and reduce
chances for abrupt end of user sessionsTRANSACTIONAL METRICS
Number of bought items Revenue from bought items Frequency of transaction
OPERATIONAL METRICS Conversion from home page or search results to cart due to some features Easy payment options increasing number of sales One click payment option or reach cart at least steps Customer engagement and avoid exit rates
Buyers Analytics- eBay
ANALYTICS FOR HOMEPAGE Arrange the homepage according to the purchase history, likes and comments of customers Analyze the increase in number of clicks on home screen and difference in navigation flow Analyze the increase in number of visits on home page during one session Analyze number of items listed on homepage to be selected for wishlist or cart
ANALYTICS FOR SEARCH Add a pop up/layer when clicked on an item from search result Give multiple options on pop up: Checkout, check details, compare Analyze increased or decreased number of clicks and conversions to cart in order to see
efficiency of the new feature and hence decide on whether to continue with the feature or not.
BUYERS ANALYTICS deals with the analytics used to design or experiment with the process flow related to purchase of a productE.g. Homepage, Search, View Item window, Checkout, Cart, Wish list etc.
Seller Analytics - eBay
ASSORTMENT ANALYTICS What are the suggested assortments for a seller Which sellers to be listed so as to maintain the assortments Major trends like most number of clicks for an item and most selling items Analyze if the most clicked items is most selling or not? If No, why not?
RATING OF SELLERS Categorize sellers into groups and hence decide on what types of deals to be done with the
sellers Analytics used for recommendation of established and flourishing practices of high rated
sellers to the less performing sellers Categorize sellers as High and low trusted or performing enabling recommendation and
listing of items from good sellers to enhance customer experience
SELLER ANALYTICS include1) Assortment Analytics 2)Ratings of Sellers
Trust Analytics - eBay
FRAUD ANALYTICS Which are the sellers or Buyers who are included in fraud For Example A Buyer may buy a product but deny paying multiple times suggesting fraud A seller may claim shipment but actually delay the shipment and increase customer waiying time
reducing their customer experience Such accounts for Buyers/ Sellers needs to be blocked for significant duration Model allow to create a new account Analyze the fraud accounts either new or old to unlist /block them
CREDIT CARDS ANALYTICS Analyze the credit rating history of customers Identify the exposure of the card and decide on highest allowed purchase amount. The allowed exposed
amount is at risk Analyze the probability of loosing this money if the customer defaults
PRODUCT HEALTH MANAGEMENT Analytics on products categories to increase customer’s experience and hence loyalty by fostering trust
for the product, seller or e-bay as whole
TRUST ANALYTICS include1) Fraud Analytics 2) Credit Cards Analytics 3) Product Health Management
Notable Practices- AB TESTING -eBay
DIVISION OF CUSTOMERS INTO TWO SEGMENTS Control Group (30% customers) Test Group (30% customers)
STEPS IN AB TESTING Introduce a feature - Eg. Increase the size of a button Enable the feature for Test Group and keep it disabled for the control Group Notice the change in behavior - Had the number of clicks increased significantly to
measure the positive response of the introduced feature. If yes continue with the feature to enhance customer experience
Decision Making - If the result in not significantly better then retract the introduced feature
AB testing is to check the efficiency of the introduced eBay product or feature is widely used by Ebay and probably the only major player using it
Notable Practice - RFM Analysis -eBayRecency | Frequency | Monetary
for Customer segmentation and Promotional Offers
Recorded data in form:Customer ID | Category of purchase | Date of purchase | Quantity of purchase | Amount of purchase
Recency Frequency Monetary
Get Recency, Frequency &
Monetary score out of 5
Calculate the combined score
Decide number of clusters & segment customers according to score.
Apply promotional schemes.Influence of
category is not considered
Frequency outweighs other
two factors
Ideal number of segments-
Managerial Decision
Which parameters should be focused for the target customer segments
Current Scenario Recommendations
Analytics used to segment customers and then direct suitable promotional in order to increase the overall revenue generated by each customer
Recency – last visit to site Frequency – how frequent is purchase and in what quantity Monetary – amount of money spend
1) RETAIL ANALYTICS
2) ANALYTICS IN ECOMMERCE INDUSTRY
3) ANALYTICS IN ECOMMERCE COMPANIES
4) RESEARCH PAPERS STUDY
5) RECOMMENDATIONS
Customer Segmentation and Promotional offersRFM Analysis : Suggested Improvements
Instead of rating similarly for all the product for Recency, Frequency and Monetary.Ratings can be done differently for different category. For E.g.
This is so because a customer buying apparel 3 month back may not be term as recent but buying cell phone 5 month back may be termed as recent because of difference in life cycle of the product or category of product
Assign weights to Recency Frequency and Monetary instead of equal weights
Home & Kitchenn_Bought_Item n_GMV n_months* score
0<=n<0.35 n<2.5 n<3 10.35<=n<0.5 2.5<=n<3 3<=n<5 20.5<=n<0.75 3<=n<3.75 5<=n<7 30.75<=n<1 3.75<=n<4.5 7<=n<10 4
1<=n 4.5<=n 10<=n 5
Appareln_Bought_Item n_GMV n_months* score
0<=n<0.35 n<2.5 n<3 10.35<=n<0.5 2.5<=n<3.25 3<=n<5 20.5<=n<0.75 3.25<=n<3.7
5 5<=n<7 30.75<=n<1 3.75<=n<4.5 7<=n<10 4
1<=n 4.5<=n 10<=n 5
Techn_Bought_Item n_GMV n_months* score
0<=n<0.35 n<2 n<2 10.35<=n<0.5 2<=n<2.5 2<=n<4 20.5<=n<0.75 2.5<=n<3.5 4<=n<6 30.75<=n<1 3.5<=n<4.2
5 6<=n<9 41<=n 4.5<=n 9<=n 5
Home & KitchenFactor Weight
Recency 1Frequency 2Monetary 3
ApparelFactor Weight
Recency 2Frequency 1Monetary 3
TechFactor Weight
Recency 2Frequency 1Monetary 3
Depending on the category one may want customer to be more recent, or more frequent or more revenue generator per purchase
Ideal Clusters based on RFMRecency Frequency Monetary Clusters
H H H BESTH H L VALUABLEH L H SHOPPERSH L L FIRST TIMESL H H CHURNL H L FREQUENTL L H SPENDERSL L L UNCERTAIN
Customer Segmentation and Promotional offersRFM Analysis : Suggested ImprovementsRate the Recency, Frequency and Monetary as High or Low for each customers and then define the segments based on the combination of these values
Divide your customers into these 8 segments Now if one wants to convert his valuable customers into best customers he knows that he can target the Monetary value of the customers and direct promotional which would increase the per purchase spending of the customers.
Customer Segmentation and Promotional offers- based on Customer Lifetime Value
THREE APPROACHES1) Segmentation by using Lifetime Value2) Segmentation by using Lifetime Value components3) Segmentation by using Lifetime Value & other information
Eg: socio-demographic factors or transaction analysis
APPROACH I (LIFETIME VALUE) Customers are sorted in descending order of LTV Percentile score is generated Target customers (constraints usually financial budgeting determines how many customers to be
targeted)
Customer Segmentation and Promotional offers- based on Customer Lifetime Value
APPROACH II (LIFETIME VALUE COMPONENTS)Three components
1) Current Value2) Potential Value3) Customer Loyalty
Three axis is derived Scoring of each customer for each component on a scale of 0 to 1 Segments based on scoring Eg: A customer with High Current value, Potential Value & Customer loyalty must be retained
Internal Data: Customer Profile; Behavior Data; Survey
DataExternal Data: Acquisition data; Co-operation data
Current Value; Potential Value; Customer Loyalty
Customer Segmentation and Promotional offers- based on Customer Lifetime Value
APPROACH II (LIFETIME VALUE COMPONENTS) Calculation of Present value
Present Value= Amount paid by customer – cost Calculation of Potential value
Probij : Probability that the customer i uses service/product j out of n services/productsProfitij : Profit that the company has when customer i uses product/service j
Calculation of Customer LoyaltyCustomer Loyalty = 1- Churn rate
Probij and Customer loyalty can be calculated through models like decision tree, neural networks and logistic regression (Training data set : Validation data set :: 30 : 70)
Customer Segmentation and Promotional offers- based on Customer Lifetime Value
APPROACH III (LIFETIME VALUE & OTHER COMPONENTS) Behavioral segmentation in terms of usage volume
Heavy users Medium users Light users
Brand buying behavior Brand loyal Brand switchers
Customer profitability
Marketing Strategy based on the segments
Customer Segmentation and Promotional offers- based on Customer Lifetime Value
CROSS SELLING AND UPSELLING Segmentation based on current value and Customer Loyalty
SEGMENT I (Loyal but less profitable) Companies may have large opportunity for upselling
SEGMENT II (Unattractive) SEGMENT III (Loyal and profitable)
Best for Cross selling of products SEGMENT IV (profitable but likely to Churn)
Unfit for cross selling but company would like to retain them
Current ValueChurn probability Low High
High II IVLow I III
1) RETAIL ANALYTICS
2) ANALYTICS IN ECOMMERCE INDUSTRY
3) ANALYTICS IN ECOMMERCE COMPANIES
4) RESEARCH PAPERS STUDY
5) RECOMMENDATIONS
Tactics for Building and Sustaining a Data Analytics TeamAs per our study we have found that the companies doing major analytics
work have in house teams hence we suggest in- house centralized analytics team
One core analytics team located at one spot in the organizational chart
Ability to allocate resources as needed Team gets exposure and experience on multiple parts of the company
Jack of all Trades, Master of None
Expertise can be built once the analytics practices have been set
In the long run, the company should move to decentralized analytics team to leverage expertise in each of the domains
Building an Analytics Culture Make intellectual curiosity a priority
Technical skills alone are insufficient Find techies who also can communicate visually
Express ideas about how a business use can best consume the output of data analysis Business Savvy Analytics Focus on important and the right level of granularity Ensure Cross-Training
Expert doing a lunch and learn with the team or writing documents with tips and tricks Look for domain expertise in your industry
They add the perspective of reality Keep top talent in steady rotation
Domain experts gain a stronger understanding of the impact of actionable insights on a company’s day-to-day decision-making
Cultivate a touch of conflict Biggest breakthroughs come from disagreement
References• Customer segmentation and strategy development based on customer
lifetime value: A case studySu-Yeon Kim a, Tae-Soo Jung b, Eui-Ho Suh c, Hyun-Seok Hwang d,*
• Realizing the Potential of Retail Analytics Plenty of Food for Those with the Appetite – Thomas H Davenport
• Explore RFM Analysis using SAS® Data Mining ProceduresRuiwen Zhang, Cary, NC; Feng Liu, University of North Carolina at Chapel Hill, NC
• How Predictive Analytics Is Transforming eCommerce & Conversion Rate Optimization (http://conversionxl.com/predictive-analytics-changing-world-retail/?hvid=352IDw)
• http://techcrunch.com/2013/08/31/how-amazon-is-tackling-personalization-and-curation-for-sellers-on-its-marketplace/
• http://www.ecommercebytes.com/pr/?id=794560• http://www.infoworld.com/article/2619375/big-data/amazon-cto--
big-data-not-just-about-the-analytics.html• http://blog.sqreamtech.com/2013/12/how-retailers-are-using-big-data-to-improve-sales-and-custo
mer-service/
• http://aws.amazon.com/elasticmapreduce/• https://gigaom.com/2011/10/18/amazon-aws-elastic-map-reduce-hadoop/• https://datafloq.com/read/amazon-leveraging-big-data/517• http://www.predictiveanalyticsworld.com/patimes/amazon-knows-what-you-want-before-you-buy-it
/
Thank you