Why this is a framework
Big-bang AI projects kill themselves.
The default shape of an AI transformation engagement is: six months of integration work, three months of model work, one big cutover, followed by a month of on-the-ground firefighting while operators learn the new system. That shape fails, consistently, because it puts every risk in the last mile.
The framework inverts this. We sequence so the least-risky, most-visible wins ship first — data model, pipeline stages, intake migration — and the AI work comes after the ground is clean. If the program gets defunded in month four, the client still has a working pipeline. Every phase ends with a hard usable-after milestone we'll sign our names to.
Contract clause — Every phase has a written "usable after" line in the statement of work. If what shipped doesn't deliver that outcome, we fix it before starting the next phase. No exceptions.