"It was only a matter of time," Roberts says. "Everyone knew this wasa long-term problem and that our business was getting more, not less,complex."

After careful analysis, Roberts determined that the databaseunderlying his systems wasn't inherently flawed. He stuck with theAS/400 platform and is rebuilding his back-office applications in Javaand RPG, an AS/400-specific programming language. He's looking atseveral legacy migration tools that could handle translating the dataanalytics processes hidden within his legacy transaction processingsystem, but he has not decided yet on a particular tool.

PMI's executive board members knew the legacy system was a problem,but they delayed the migration project in order to focus on gettingthe company's e-commerce strategy in place. Once that was complete inlate 1999, Roberts and Kathy Schroeder, vice president of policymanagement systems and the project's business sponsor, got to workconvincing the board that the time was right to tackle the migration.And one of the reasons the time was right, they said, was that thecost of the project, estimated at $20 million in 1996, had come downin 1999 to an estimated $12 million.

"The board knew that the business had changed and that the legacysystem wasn't built to deal with the kind of claims processing andtransaction products we had," Roberts says. "They knew that if wecreated a new product, the old system would limit our ability to rollit out. We had to make it very clear that waiting any longer couldreally hold us back."

The migration was approved, the money found. Roberts went to workpulling apart his legacy applications in October 2000. He plans tofinish by the end of 2003.