Field note
The Teradata proposal that never happened
Performance testing passed on every scenario. Discovery found an undocumented long tail, the RoI model broke, and the programme was stopped.
The most instructive migration I have worked on never happened.
It was a proposal to move a large Teradata platform to Oracle. The major acceptance criterion was performance, and the performance testing went extremely well — the new platform outperformed the old one in every scenario, and the margin was widest exactly where it mattered most, under multi-user and mixed workloads. On the criterion the programme had chosen to be judged by, it was an unambiguous success.
The project never went ahead.
What we had tested was what IT knew about. Underneath the platform IT understood and could describe, there was a long tail of small, discrete line-of-business domains that had never been integrated into the whole and were sparsely documented or not documented at all. As discovery reached them, it became clear that these could not be migrated in any meaningful sense — they would have to be rebuilt from scratch, because there was nothing to migrate against. That was a fundamentally different programme from the one that had been costed. The revised effort broke the return-on-investment model, the costs exceeded the benefits, and the proposal was stopped.
Where this feeds the guide: Do the discovery before you commit to the business case and Performance problems are the rule.