Target Behavior in Real Time for Effective Outcomes: Part 2
Drive marketing and business management decisions using a real-time, adaptive architecture
The first installment in this series examines the growing role of behavioral targeting by leveraging big data, Internet of Things data, and advanced analytics across myriad application domains. This concluding installment presents deploying a real-time adaptive architecture for effective behavioral targeting. Effective targeted actions require analytics and process infrastructures for detecting, focusing, and designing an action. In many cases, the action must be taken in real time while a customer is interacting with a provider or walking past a point-of-interest location. The solution requires tight integration and performance management across a number of its components. The D4 framework offers an integrated solution for these targeted actions using four components that discover, detect, decide, and drive targeted actions (see figure). A real-time, adaptive architecture comprising discover, detect, decide, and drive components
Analytics target behavior in a retailer use case
An analyst uses analytics techniques to discover historical behavioral patterns. Through rigorous quantitative and qualitative analytics processes, the analyst may establish normal as well as abnormal behaviors. For example, a marketer working for the New Store retailer may discover a microsegment that likes to collect Foursquare points for visiting grocery stores and also regularly shops at the Good Neighbor department store. In addition, the marketer observes Lisa, a customer who regularly visits the Good Neighbor department store every Thursday evening when she is not traveling out of town for work. The associated parameters and related thresholds are now passed on to the detect component, in which big data sources are scanned on a near-real-time basis. This component discovers when a collection of observed parameters match a desired pattern. In the campaign example provided in part 1 of this series, this matchup could be the Thursday when Lisa is not traveling for work and should be offered an incentive to visit the New Store. In doing so, the real-time analytics system may ignore all of Lisa’s neighbors, who do not match the specific pattern. The selected customers are then passed on to the decide component, in which a rule engine configures a specific action to handle a particular situation. In the case of the marketer campaign example, the marketer may use a variety of rules covering saturation—how many offers were made in the recent past—along with response likelihood based on past offer redemptions and predisposition to location-based offers. As a result, two members in a microsegment may receive radically different offers to maximize the likelihood of redemption. The chosen decision is now conveyed to the drive component for the business process action. This process is a campaign management system that engages with the customer, delivers the action, and records the customer response to fine-tune the parameters for future offers. The architecture requires a tight integration across four divergent tools. The discover component is mainly conducted using statistical and machine learning tools such as IBM® SPSS® statistical analysis software and the Big R built-in analytics feature in the IBM InfoSphere® BigInsights™ big data management and analysis software. The detect component uses real-time analytics products such as IBM Now Factory Sourceworks Data Collector, and InfoSphere Streams streaming data analytics software, which can ingest large-scale data and filter it based on known criteria. The decide component uses rule engines—for example, IBM Operational Decision Management (ODM)—that use a combination of real-time data and historical data to configure a specific action. And the drive component uses business process automation and workflow tools such as IBM Unica® marketing solutions, business process management (BPM), and IBM Counter Fraud Management to drive the decision.
The adaptive architecture depends on advanced skills
The solution requires an intricate combination of three underlying advanced skill sets: data engineering, data science, and application and domain knowledge. Data engineering is required for high-volume, high-velocity data. Given the level of automation, the entire decision-making cycle should be completed during a transaction. It requires a system that can react to changes in near-real time ranging from a couple of seconds to a couple of minutes, depending on the use case. At the same time, both the discover and decide components represent large amounts of data to establish patterns and for historical analysis of a chosen focus area. Data engineering also needs to institute the appropriate data flows without choke points to help facilitate smooth interworking across these components. Data science is required to handle patterns that look for exceptions. Working with averages is no longer sufficient. In many use cases—including those mentioned in the first installment of this series—exceptions rule. Data scientists should establish ways to identify micropatterns and methods of detecting and deciding to find and act on these micropatterns. Application and domain knowledge is necessary because the pattern covers the entire gamut of the decision component. Someone who has a very good understanding of the use case and the associated industry domain knowledge can easily weave an intricate interplay across elements. While an outsider may develop surface-level algorithms and rules for each component, a domain-savvy expert can identify an underlying domain model to help drive all four components. Please share any thoughts or questions in the comments.
|[followbutton username='voraconnect' count='false' lang='en' theme='light']|
|[followbutton username='IBMdatamag' count='false' lang='en' theme='light']|