Failures · GAIF-FA-001Global AI Forum
The model was not the problem
A cancer centre spent USD 62.1 million on a clinical decision support system that never reached a decision. The audit took no position on whether the technology worked. It did not need to.
Key facts
- MD Anderson Cancer Center spent USD 62.1 million on the Oncology Expert Advisor between June 2012 and October 2016, paying IBM approximately USD 39.2 million and PricewaterhouseCoopers approximately USD 23 million. Source: University of Texas System Administration special review, February 2017.
- The original contract, signed June 2012, was scoped at USD 2.4 million for six months covering one leukemia subtype. It was extended twelve times. No other cancer type beyond the original six was ever added.
- No patient was ever treated on the basis of an Oncology Expert Advisor recommendation. No clinical production deployment occurred at any scale.
- IBM's internal assessment of September 2016 concluded the system was not ready for human investigational or clinical use. IBM withdrew support that month.
- A 2018 peer-reviewed paper in The Oncologist reported OEA information extraction accuracy of 90 to 96 per cent on clear concepts such as diagnosis, but 63 to 65 per cent on time-dependent information such as therapy timelines.
- The primary failure mode was Mode 1, no workflow: the system output had no defined destination inside the clinical decision path. The contributing modes were Mode 2 (no data path), Mode 4 (wrong bottleneck) and Mode 6 (no stop condition).
A treatment advisor that worked was never wired to the moment an oncologist actually decides. A recommendation that arrives beside the decision rather than inside it is not a recommendation. It is a document. USD 62.1 million bought a document.
Section one
What actually happened
In June 2012 MD Anderson contracted IBM to build the Oncology Expert Advisor: a system that would read a patient's record, cross-reference the literature, propose treatment options and match patients to trials. The first version was scoped at six months and USD 2.4 million, covering one leukemia subtype: lower-risk myelodysplastic syndrome.
The contract was extended twelve times over the following four years. Scope grew to five further leukemias and lung cancer. IBM was paid approximately USD 39.2 million and PricewaterhouseCoopers approximately USD 23 million. The final figure was USD 62.1 million, twenty-six times the contracted value. No other cancer type beyond the original six was ever added.
During the same period MD Anderson replaced its record system, moving from the legacy ClinicStation platform to Epic. OEA had been built against ClinicStation. It was never integrated with Epic, and it was never going to be without a rebuild that nobody funded.
The practical consequence is the whole case. To use OEA, a clinician had to leave the record system where the patient's data lived, open a separate interface, enter the data again by hand, read the output, and go back to the record system to act. That sequence is not a workflow. It is the absence of one.
A six-month contract that ran for fifty-two months
Original contracted value against total spend, USD millions
The gap the money never crossed
The clinical path, and where the system was actually connected
In September 2016 IBM withdrew support, noting internally that the system was not ready for human investigational or clinical use. The last contract extension expired that October. In February 2017 the University of Texas System published a special review of the procurement. No patient was ever treated on the basis of an OEA recommendation.
Section two
What produced this outcome
Four structural conditions were present. Only the first is the primary mode. The others made it survivable for four years.
1. No defined clinical workflow
At no point was there a specification of which clinician, at which point in which pathway, would consult the system, and what they would do differently as a result. The output had no destination. This is Mode 1. Removing it would have changed the outcome.
2. No data path into the institution
The migration to Epic was a known, scheduled, institution-wide programme. A decision support tool built against the outgoing system, with no funded integration path to the incoming one, was obsolete on a published timetable. Manual re-entry is not an integration strategy. It is the absence of one. This is Mode 2.
3. Prestige selected the use case
The stated objective, as described in the audit, was to let community oncologists deliver care at the standard of MD Anderson specialists and to elevate the worldwide standard of cancer care. That is an ambition, not a bottleneck. Nothing in the record identifies a measured constraint in the oncology pathway that the system was built to relieve. This is Mode 4.
4. Governance did not stop it
The audit found that procurement standards were not followed, that contracts were structured to sit below oversight thresholds, and that gift funds were committed before receipt. Twelve extensions to a six-month contract is not a series of decisions to continue. It is the absence of a decision to stop. This is Mode 6.
The audit also noted a governance conflict: the project's originator was a senior faculty member married to the institution's then president. The Forum records this as a control weakness, not as a motive.
Section three
What IBM said, and what the evidence settles
IBM's stated position is that the system worked. A spokesperson said the recommendations were accurate, agreeing with expert opinion around 90 per cent of the time, and that the research and development project was a success which could likely have been deployed had MD Anderson chosen to take it forward.
Finding
The model was not the problem. It was never connected to the moment an oncologist actually decides.
A system with a 90 per cent concordance rate and no route to a clinician produces exactly the same clinical benefit as a system with a 40 per cent concordance rate and no route to a clinician. Both produce none.
IBM's position
The system worked. Recommendations agreed with expert opinion around 90 per cent of the time. The project was a successful research and development engagement. Deployment was possible had MD Anderson chosen to proceed. Source: IBM public statements, February 2017.
What the evidence establishes
The University of Texas audit took no position on accuracy or functional capability. It found the system was never integrated into clinical workflows. IBM's own internal assessment of September 2016 concluded the system was not ready for clinical use, contradicting its public defence. The 2018 peer-reviewed paper in The Oncologist reported accuracy of only 63 to 65 per cent on time-dependent information, the harder half of an oncology record.
Section four
The six failure modes
The Failure Atlas assigns each case one primary mode and any number of contributing modes. Modes describe structure rather than technology, which is why they survive changes in the technology.
| Mode | Name | Role in this case | Definition |
|---|---|---|---|
| 1 | No workflow | Primary | The system functioned and was never wired to the moment a human decides. Output arrives beside the decision rather than inside it. |
| 2 | No data path | Contributing | The system cannot reach the data it needs from inside the institution, so humans carry the data to it by hand. |
| 3 | No owner | Not present | No named person is accountable for the decision the system was built to change. |
| 4 | Wrong bottleneck | Contributing | Visibility, prestige or ambition selected the use case rather than a measured constraint. The system solves something that was not costing anything. |
| 5 | No baseline | Not present | Nobody measured what the humans did before, so improvement can never be demonstrated. |
| 6 | No stop condition | Contributing | No agreed circumstance under which the programme would be halted, so it is extended instead of judged. |
Source: Global AI Forum Failure Mode Taxonomy, September 2026. The primary mode is the one whose removal would have changed the outcome.
Section five
What a buyer could have done before signing
This section is not advice about oncology or about IBM. It is a test that applies to any AI programme in any function, and it is designed to be failed early and cheaply rather than late and expensively. Most enterprise AI pilots pass a technical evaluation and fail question three, because question three is the only one that requires somebody to change what they actually do.
-
Name the moment a human currently decides.
Not the process. The moment. Which screen, which meeting, which minute. If the answer is a department or a workflow rather than a decision point, there is no destination for the output. At contract signature in 2012, the stated objective was to elevate the worldwide standard of cancer care. That is not a moment. Fails question 1 before any money moves.
-
Name the person who owns that moment.
A named individual who can change how that decision is made without asking anyone else. If the owner is a committee, the system will be evaluated rather than used. The Epic migration decision moved the named decision moment to a system the tool could not reach. Fails question 2 at the point of the Epic migration decision.
-
Say what they will do differently on the Tuesday after go-live.
In one sentence, in their own words, describing an action they take or stop taking. Not "have better information." An action. By extension 3 or 4, no clinical production use had occurred. No stop condition existed. The absence of an answer to question 3 is itself the finding.
If you cannot name all three, you are not buying a system. You are commissioning a document, and you are building this case again with a better model.
| Checkpoint | What was known at the time | What the test would have surfaced |
|---|---|---|
| Contract signature, June 2012 | Objective stated as elevating worldwide cancer care | No decision moment named. Fails question 1 before any money moves. |
| Epic migration decision | Institution-wide record system replacement scheduled | The named decision moment moves to a system the tool cannot reach. Fails question 3. |
| Extension 3 or 4 | Still no clinical production use | No stop condition exists. The absence is itself the finding. |
Source: University of Texas System Administration special review, February 2017; UT System audit, 2017.
The record
Case GAIF-FA-001 A
- Subject
- Oncology Expert Advisor (OEA) A Watson-powered clinical decision support tool, built with IBM for MD Anderson Cancer Center. Distinct from IBM's Watson for Oncology product, whose clinical content was developed with Memorial Sloan Kettering. The two are routinely confused in secondary coverage.
- Commissioning body
- University of Texas MD Anderson Cancer Center Houston, Texas. Part of the University of Texas System.
- Principal vendor
- IBM With PricewaterhouseCoopers engaged on project support.
- Sector
- Healthcare. Clinical decision support, oncology.
- Jurisdiction
- United States
- Period
- June 2012 to February 2017 Contract signed June 2012. IBM support withdrawn September 2016. Final contract extension expired 31 October 2016. Audit published February 2017.
- Stated objective
- To let community oncologists deliver care at the standard of MD Anderson specialists Described in the audit, paraphrasing the project's originator, as elevating the standard of cancer care worldwide.
- System deployed
- Decision support over curated case data Version 1.0 covered a single condition, lower-risk myelodysplastic syndrome. Scope later grew to five further leukemias and lung cancer. No other cancer type was ever added.
- Quantum
- USD 62.1 million IBM approximately USD 39.2 million. PricewaterhouseCoopers approximately USD 23 million. Original contract: USD 2.4 million for six months, extended twelve times.
- Outcome
- Cancelled. Never used on a patient. The contract was allowed to expire. No clinical production deployment occurred at any scale.
- Primary mode
- Mode 1. No workflow.
- Contributing modes
- Mode 2, no data path. Mode 4, wrong bottleneck. Mode 6, no stop condition.
- Evidence grade
- A. Independent audit by the parent institution, plus first-party statements from both the commissioning body and the vendor, and a peer-reviewed publication.
-
June 2012
Contract signed
MD Anderson contracted IBM for USD 2.4 million over six months to build the Oncology Expert Advisor covering one leukemia subtype.
-
2012 to 2015
Twelve extensions, scope grows
The original contract was extended twelve times. Scope grew to five further leukemias and lung cancer. No other cancer type was ever added.
-
2012 to 2016
Epic migration proceeds in parallel
MD Anderson replaced its record system, moving from ClinicStation to Epic. OEA had been built against ClinicStation and was never integrated with Epic.
-
September 2016
IBM withdraws support
IBM noted internally that the system was not ready for human investigational or clinical use. IBM support was withdrawn.
-
October 2016
Final extension expires
The last contract extension expired. No clinical production deployment had occurred.
-
February 2017
Audit published
The University of Texas System Administration published its special review of procurement procedures. No patient had been treated on the basis of an OEA recommendation.
-
2018
Peer-reviewed accuracy study published
A paper in The Oncologist reported OEA information extraction accuracy of 90 to 96 per cent on clear concepts and 63 to 65 per cent on time-dependent information such as therapy timelines.
What this research does not establish
- Whether the Oncology Expert Advisor would have produced clinical benefit if it had been integrated into the Epic workflow. IBM's assertion that deployment was possible is plausible but untested. The evidence does not settle this question in either direction.
- The total cost to IBM of the engagement, including internal resource not captured in the USD 39.2 million figure. The audit traces payments made by MD Anderson, not IBM's cost base.
- Whether procurement irregularities identified in the audit, including contracts structured below oversight thresholds and gift funds committed before receipt, affected the clinical or technical direction of the programme. The Forum records these as control weaknesses, not as drivers of the primary failure mode.
This case is not an assessment of Watson for Oncology, the separate IBM product whose clinical content was developed with Memorial Sloan Kettering and which was sold internationally. Reporting on that product, including the 2018 investigation into unsafe recommendations, concerns a different system and is filed separately in this Atlas. Conflating the two is the most common error in secondary coverage of this case.
Cite this as
Global AI Forum, The model was not the problem, September 2026. GAIF-FA-001 v1.0. gaiforum.comSources
- University of Texas System Administration, special review of procurement procedures related to the UTMDACC Oncology Expert Advisor project, February 2017. Primary source for the quantum, contract history, twelve extensions and procurement findings.
- IBM, public statements on OEA accuracy and deployability, February 2017. Internal assessment of readiness for clinical use, September 2016, as reported in the UT System audit.
- MD Anderson Cancer Center, public statements, 2017.
- The Oncologist, 2018. OEA information extraction accuracy: 90 to 96 per cent on clear concepts, 63 to 65 per cent on time-dependent information.
- The Cancer Letter and Houston Chronicle, February 2017. First reporting on the audit.
- Forbes, February 2017. IBM's public response.
- IEEE Spectrum, 2019. On the integration failure and the ClinicStation-to-Epic migration.
Global AI Forum · GAIF-FA-001 · v1.0 Research current as of 8 September 2026 gaiforum.com