"It wasn't a terribly scientific analysis, and we didn't have a good understanding of the overall response time for customers," says Brian Lawton, vice president of Web support services for KeyCorp, an $80 billion financial institution based in Cleveland.
The only other ways to measure Web-site performance at that time were through outsourcing deals, where an outside service provider or vendor monitored the site using customer simulations, not the real thing. That wasn't an option for KeyCorp given the secured transactions on its site. Now the bank runs a monitoring tool on its Web servers that measures the response time of an online banking transaction--everything from how long it takes to sign on to transferring funds or paying a bill. "We needed something that could provide performance measurement from the customer's perspective--24 hours a day, seven days a week," Lawton says. "We felt it was just as important to know what was happening at 2 a.m. as at 11 a.m."
KeyCorp's goal is for banking customers to wait no more than three to six seconds for a transaction. After running Candle Corp.'s eBA*ServiceMonitor tool, it streamlined some transactions it found weren't performing as efficiently.
The monitoring technology tags a customer's HTML transaction with a Java applet, which is then launched from the KeyCorp customer's desktop and runs during the session. The applet records the response time, and feeds that information into the eBA*ServiceMonitor. Lawton says this adds a negligible bit of overhead while the applet is downloaded--about one to one-and-a-half seconds. KeyCorp preferred the Java applet to the standard cookies polling mechanism that comes with most HTTP transactions and is less secure. "We didn't want to leave cookies behind cluttering the customer's PC, and we didn't want to be in a position where a customer might perceive the response-time monitoring as information 'mining,'" Lawton says.
The downside of this technology, however, is that it's not reporting performance in real time. The software logs the response-time performance, and KeyCorp accesses it through Candle's secured Web site. "There's a 24-hour turnaround on the data," Lawton says. "I'd like to see performance issues on a real-time basis instead of the next day when we check the Web site."
And Candle's software doesn't feed its data into KeyCorp's other monitoring tools, such as BMC Software Best/1 and Tivoli TME 10. But APIs are available that send the data to TME so either tool can generate alerts when response-time problems arise on the Web site. KeyCorp plans to implement these real-time alerts later this quarter, Lawton says.