What would you have done? MD Anderson launched a $2.4M AI project — it cost $62M and was never used

In 2012, the MD Anderson Cancer Center signed a six-month, $2.4M contract with IBM Watson to build an oncology advisor meant to analyze records and recommend treatments. Three options were on the table: (A) aim straight for the most ambitious system; (B) start with one bounded use case, integrated with existing tools and evaluated, before scaling; (C) wait for the technology to mature.

MD Anderson chose (A). The scope kept drifting: twelve amendments, four years, $62M spent — and the system never entered clinical use, because it was never integrated with the hospital’s new patient-record system. The lesson holds for any sector: the decisive risk in an AI project is rarely the technology, but scope creep and deferred integration. Before committing an AI budget, define the first use case narrowly, require proof of integration with your existing systems, and stage the work in phases with quantified exit criteria rather than an open-ended contract.

Source: The Cancer Letter, University of Texas System audit of the MD Anderson–IBM Watson project, February 2017.