Database Marketing Fall 2016 Web analytics (incl real-time data) Collaborative filtering Facebook advertising Mobile marketing Slide set 8 1 Web analytics: Data Collected via the Internet Customers can state preferences on the Internet and this data can be used to help establish relationships. Reputable Internet marketers will protect customer data and clearly outline privacy policies. Individual-level customer data collected on the Internet can be classified as either registration, behavior, or source data. NOTE that it is also possible to manage the behavior using cookies Web analytics 2 1
Registration Data Many e-commerce sites require consumers to register before they can make purchases, get information, or enter the Web site. Examples of Internet registration data typically collected include Name Title Business title (if the site is a b-to-b site) E-mail address Postal address (business or home) Phone number Fax number Age Income level Gender etc Web analytics 3 Behavior Data Types of customer behavior data that can be captured includes Visits Total page views Specific page views Time spent at the site Products purchased Customer service requests Access to personal account information Discounts used Web analytics 4 2
Source Data It is valuable for the marketer to know what media convinced a customer to visit the site. Possible sources are e.g.: Promotional e-mail, Article or site link Search engine Word of mouth Online ad Some are familiar with the site s offline store Commercials or print ads Direct mail etc Web analytics 5 Use of cookies Cookies are small text files that are initiated in the web page visitor s computer when the visitor visits a certain web-page first time When the visitor makes a re-visit he/she can be recognized Allow to consider web page visitor s behavior without registration Web analytics 6 3
Challenges in the use of cookies Cookies are deleted by users to protect privacy People have several devices to access internet (no combining of data) Different family members use the same device All these mentioned sources of inaccuracy are problems in the use of cookies Web analytics 7 Targeting Online Customers Targeting online customers is very similar to the off-line world and consists of two major steps: Defining the target markets Executing a contact strategy The strategy will dictate the offer and message and the manner in which both are extended Web analytics 8 4
Online Analysis of Customers For E-commerce applications, clickstream data and traditional direct marketing data can be used in predictive modeling. Marketers can also define your online target markets with regression, neural network etc modeling. Internet data can be used to predict customers likely to click through. The same methods are used when realtime data is employed but the data architecture is different there. Web analytics 9 Real-time data and analytics Five Sources of streaming data Real-time data 10 5
1) Operational monitoring Datacenters house thousands of discrete computer systems that record data about their state (processor temperature, state of disk drives, processor load, network activity). This data is collected and aggregated in real time. Real-time data 11 2) Web analytics Website activity tracking. There is the need to have a short feedback loop and to collect the data in real time. Streaming infrastructure currently in use was developed to safely move data from the webservers to processing and billing systems. Also to recommendation systems. Real-time data 12 6
3) Online advertising 1/2 When a visitor arrives on a web page bidding agencies (perhaps 30-40 at the same time) place bids on the page view in real time via an advertising exchange. An auction is run, the advertisement(s) of the winner(s) displayed. Elapsed time is less than about 100 milliseconds. All the parties (exchange, bidding agent, advertiser..) collect the related data. The data collected for the advertising exchange are e.g. for billing, data monitoring the fraudulent traffic and data for other risk management activity. Real-time data 13 3) Online advertising 2/2 Advertisers publishers and bidding agents are collecting data to optimize the campaigns: to set the bid (advertiser) / to set the reserve price (publisher). A bidding agent represents often many advertisers or publishers and will will often be collecting on the order of hundreds of millions to billions of events per day. All the data is collected, analysed and stored as it happens. Real-time data 14 7
4) Social Media The data is text data and thus, unlike web or online advertising, unstructured. It must be processed to be understood by automated systems. Social media data is challening for the real.time data sources. Real-time data 15 5)Mobile data and the internet of things Smart phones and their number is growing fast. They are communicating with nearby objects using e.g. Bluetooth. Most versatile real time applications (sleep activity activates coffee maker, biometrics, sensor measurements collected by smart phones Internet of Things). IFTTT If This Then That (a web-based service that allows users to create chains of simple conditional statements) Real-time data 16 8
The special characteristics of streaming data Always on. Processing times need to be short or the data is lost, shorter than real time. Loosely structured. High-cardinality storage. Large set with a high number of dimensions. Storage space is restricted because very fast main memory storage needs to be used. STREAMING DATA REQUIRES SPECIAL ARCHITECTURE Real-time data 17 Sampling from a Streaming Population Basically analogous to sampling from a fixed population. Use specific algorithms. Real-time data 18 9
Collaborative Filtering A technique used in recommendation engines We discuss two kinds : q MEMORY BASED q MODEL BASED Pioneer: Amazon Now the use is spreading outside normal retailing to financial services, travel agencies etc Collaborative filtering 19 How to predict the missing values Customer Movie1 Movie2 Movie3 Movie4 Movie5 A 5 2-4 1 B 1-1 2 - C 5 3 4-1 The figures are individual preferences of A, B and C for films. We try and find the missing preferences. When we have evaluated the missing values we may start recommending movies according to the preferences. Collaborative filtering 20 10
Memory based methods They are neighborhood based meaning they attempt to find a set of users that have preferences similar to the target user. Once similar users, called the neighborhood have been identified, the neighborhood preferences are combined to predict the preference of the target user. Thus we have to define the similarity between users (much in the same way as in clustering). Collaborative filtering 21 Model based methods Examples: 1) Methods based on clustering Customers are simply clustered with the basis of clustering being e.g. how many items of each product the customer has bought. Once the clusters are determined cluster models assign a target user to the segment including most similar customers. Collaborative filtering 22 11
Model based methods (cont.) 2) Item-based collaborative filtering Consider the items the customer has rated, then calculate how similar they are to the target item (which we recommend or not). Then thosesimilar items rates are be used in the prediction. Collaborative filtering 23 Combining Content-Based Information with Collaborative Filtering Content based information filtering recommends items for users by analysing the content of items that they liked in the past. Here other users are not employed but instead some kind of similarity measure between two product items. Example: if you liked Manga comics we recommend other Manga comics. Collaborative filtering 24 12
Recommendation engines Interact with the customer while the customer is in (suggesting related products while filling the customer basket in the e-shop) Nowadays also text mining used (evaluations of books in Amazon etc.) Collaborative filtering 25 Facebook advertising 1/3 Facebook offers you targeting possibilities without having to have your own database They can target e.g. on the basis of location, demographics, interests, behavior, connections. The behavior needs knowledge about your browsing history and can display advertisements on that basis in e.g. in the news feed (cookies use). Cookie is data file placed on your computer by the website that you visit. Third party cookie is a datafile placed by another website than which you visit. Facebook advertising 26 13
Facebook advertising 2/3 Advertiser may also provide a list of their customers and seek them in Facebook (custom audiences) Additionally they can seek Facebook members that are similar to their best customers (lookalike audiences) Factor Analysis 27 Facebook advertising 3/3 Facebook tracks your browsing activity if you re logged in to the network but not actually using it and the modern advertisers show you ads on third-party sites and apps. Facebook tracks the third-party websites people visit on mobile devices through the Like button, assuming that they are logged into the social network. Based on that advertisers can refine an audience to target ads around likes and interests for third-party websites and apps without having to depend on cookies. Facebook advertising 28 14
Mobile Marketing marketing on a mobile device Advertisements in apps (Facebook, games) Geofencing Location based marketing QR codes, scanning a code takes you to a website Mobile search ads Mobile image ads SMS Mobile Marketing 29 15