This morning I watched and listen to two old friends of mine – Wayne Eckerson of TDWI and Dan Graham of Teradata as they gave a webinar on “Approaches to operational BI” (PDF of slides, Webinar Recording). Wayne and Dan did a nice tag team discussing the principles involved and giving some Teradata-specific examples.
Wayne began by discussing the operational BI continuum as he sees it – from analyze process to monitor process to facilitate process (composite applications) to execute processes (event-driven analytic platforms). This is a nice way to think about the different approaches, though I worry that people will think you must go through each level across the board before moving to the next one. I believe that different problems can be approached at different levels very effectively.
Wayne discussed some of the problems with analyzing processes and their historical performance, particularly that of multiple data sources, and went through various solutions – Multiple tools to access them (a pain), using EII to link a single BI tool to multiple operational systems (can be effective but it can be very hard to integrate and requires direct access to operational systems), using an ODS (typically duplicates the data in data warehouse and tends to be separate from the BI enviroment) and finally an active data warehouse (needs mixed workload support as now doing both analytic and operational reporting, no data redundancy or inconsistency but has issues). Clearly the last is Teradata’s POV
As you move into monitoring processes, dashboards become the tool of choice, focused on exceptions and on issues with process execution. Dashboards are becoming more dynamic, more integrated with Microsoft Office and more mobile-friendly. Collaboration and modern UIs are coming too. Facilitating processes means composite applications and this tends to mean either portals or embedded reports in applications. Executing a process makes for an analytically driven process and can use everything from stored procedures to custom event-driven applications to BAM/CEP to rules. While I don’t think event-driven is the only kind of analytical process, I do agree that this is a main driver for adopting analytic processes – it is much easier as part of a move to event-driven.
Dan had some nice examples. One of a portal for a call center with database information (recent calls), dashboard stuff (the particular call center representative’s KPIs), customer information and decision automation for offers – different trigger words linked to “right” offer based on analytics. He used a nice example of returns where a fairly dumb POS system needs decisions about valid returns that can detect fraudulent returns, calculate the refund etc. without requiring any understanding of the process on the part of the sales clerk (classic unassisted decision automation). Finally he used one of my favorites, Harrahs, with their nice use of ILOG’s rules engine for event filtering and business logic. Gaming events come in, The Tibco service bus is used along with an Active Data Warehouse and various local stores and responses are based on rules and on analytics run against the data in real time. He also had a nice hook in to show how visualization can show where someone is, once a decision is made.
It is worth watching for some good introductory material on terminology and some nice examples. I am still not sure I would lump decision automation and management under Operational BI though – I think enterprise decision management or EDM is complementary to but different from Operational BI.