There are business scenarios where latency in awareness, evaluation and action in response to events has a high impact on business performance. These scenarios are typically around operational activities that are performed by front-line workers. For example the potential cost in both lost revenue and customer satisfaction are high if movement of promotional items from the stock room to the store shelves does not keep pace with customer purchase volume throughout the day.
These types of business scenarios require a shift from the traditional data centric architecture of business intelligence (BI) to an event centric architecture to detect the current state of activities and processes, analyze them against expected states, and embed intelligence into process workflows that helps users determine the most appropriate response to threats and opportunities. Different people or groups use different names to refer to technology solutions for event based intelligence, such as Business Activity Monitoring (BAM, Gartner) and Complex Event Processing (CEP, David Luckham). Regardless of the name used the solutions incorporate technology components from business intelligence, semantics, business rules, and business process management.
The first requirement of event based intelligence is to integrate events from multiple sources that flow across the enterprise service bus. I am defining an “event” as an object that contains information about a change in the state of an operational activity or process. To do this requires a broad spectrum of input adapters for things such as system logs, network protocols, relational databases, API calls, messaging queues and Web services. The input adapters must be able to detect events at different points in the business process workflow, which enables throughput and temporal analysis. Further, event based intelligence must be able to detect events across multiple processes and nonlinear workflows in a single process, which enables correlation of activities to changes in business performance. This last aspect requires workflow modeling capabilities or the ability to integrate with modeling tools using workflow specification languages such as Business Process Execution Language (BPEL).
Event based intelligence also requires the capability to model the attributes, constraints and dependencies of events. As an information object, an event is an entity that has attributes, such as data values, time of occurrence, process ID and workflow sequence that can be used to define queries and constraints for simple pattern-matching. Event entities also have dependencies between each other that can be modeled into hierarchies and used to match against more complex business patterns. Predictive analytics can assist in the creation of an event model when the structures and relationships are not well-understood. Users can employ statistical modeling to estimate the attributes and dependencies when the general event structure can be hypothesized on the basis of domain knowledge and analysis. If the structure is completely unknown, data mining can be employed to infer both the attributes and dependencies of an event model.
The event-processing engine matches sets of events from different workflow streams against patterns in the event model. Filtering reduces the total event set to just the relevant subsets, which are aggregated based on dependencies at different levels within the model hierarchy. Then the event engine uses constraints to detect patterns that should never occur, those that should always occur and changes in state conditions at different hierarchical levels. When constraints are violated, the event processing engine executes response rules, such as delivering information to front-line workers or changing the routing of process workflow.
At a human interface level, event based intelligence uses dashboards to deliver personalized views of process activities and events. Semantic layers take the burden off the user to sort out inconsistencies between the meanings of event information displayed from different applications. Dashboards enable users to drill down through the event hierarchies to understand the underlying causes of complex event patterns. Federated query and search capabilities enable front-line workers to access information in databases, documents, and e-mail messages and on the Web to add context to event information.
While not all decision making processes require event based intelligence, early adoption use cases include;
Financial Services – Algorithmic Trading and Execution, Risk Management, Fraud Detection
Government – Terrorism Surveillance, Systems Intrusion Detection, Battlefield Command and Control
Manufacturing – Procurement, RFID Tracking, Transportation & Freight Management