IBM Watson for Oncology
Watson for Oncology was discontinued after a decade of commercial struggle when IBM sold Watson Health assets to Francisco Partners in 2022.
“In our view, IBM marketed a game-show trivia champion as a world-saving oncologist, only to deliver a digital binder of New York treatment habits that bewildered doctors worldwide.”
Our takeIBM pitched Watson for Oncology as a cognitive supercomputing breakthrough that would democratize expert cancer care by analyzing medical literature in minutes, but the system struggled with real-world patient diversity, relied on manually scripted training, and was ultimately broken up and sold off.
- Hype · our take
- 5/5 promised → 2/5 delivered
- Timing · our take
- too early
- Dates
- announced · discontinued
- Where
- United States · Incumbent Product · Internal
The Pitch
IBM developed Watson for Oncology as a flagship enterprise healthcare offering designed to revolutionize cancer care through cognitive computing E4. Intended for oncologists, hospitals, and multidisciplinary cancer centers, the cloud-based system promised to address the overwhelming deluge of clinical trials and scientific publications E4E2. Instead of clinicians spending weeks searching medical databases, Watson was promoted as able to analyze tumor genomic fingerprints, assess patient records, and recommend tailored therapies in mere minutes E2E4.
Clinicians interacted with the system by uploading a patient's genetic sequence and tumor mutations directly to the cloud interface E4E2. Watson would then sift through thousands of gene variations to identify driver mutations and match them against approved cancer therapies, experimental regimens, or off-label pharmaceuticals E2. To train the underlying recommendation pathways, IBM partnered extensively with Memorial Sloan Kettering Cancer Center, ingesting roughly 15 million pages of medical literature, 200 textbooks, and 300 medical journals alongside institution-specific care protocols E4. The product was sold to medical centers via subscription agreements and per-patient fees ranging from $200 to $1,000 E4.
What Happened
IBM began marketing Watson for Oncology to clinical institutions globally with high-profile partnership announcements across dozens of cancer centers E4. Early press and announcements celebrated the platform as the next frontier for precision medicine E2. By 2017, IBM claimed the technology was supporting treatment across multiple major cancer types and expanded into more than 55 hospitals worldwide E4E4.
However, operational cracks emerged quickly as the software encountered the messy realities of clinical practice. The University of Texas MD Anderson Cancer Center halted its high-profile Oncology Expert Advisor collaboration after spending approximately $62 million without delivering a viable tool E4. Shortly thereafter, independent investigative reporting revealed that Watson was not deducing novel insights on its own, but instead reciting hypothetical scenarios scripted by human doctors at Memorial Sloan Kettering E4E2.
International clinicians reported that recommendations were heavily biased toward American practices and unavailable therapies E4E4. Concordance rates fluctuated wildly; an unpublished study in Denmark recorded Watson agreeing with oncologists only 33 percent of the time E4. Furthermore, genetic sequencing frequently yielded no actionable therapeutic targets, limiting the utility of automated matching E2. With clinical adoption faltering and no peer-reviewed trials demonstrating improved patient outcomes E4, IBM retreated. In 2022, IBM sold its core Watson Health assets to private equity firm Francisco Partners, formally dismantling the venture E4E4E4.
Why It FailedOur take
In our view, Watson for Oncology failed due to a fundamental model capability gap combined with an aggressive marketing disconnect. The core premise presumed that rule-based natural language processing and synthetic training cases could replicate the nuanced clinical reasoning required for multifaceted oncology care. Instead of generating genuine clinical discoveries, Watson functioned as an expensive, fragile expert system that mirrored only the local prescribing patterns of its training hospital. When exposed to disparate global treatment protocols, fragmented health records, and ambiguous genomic variants, the product could not maintain clinical accuracy or earn the trust of physicians.
Lessons and What Would Have Worked
- In our opinion, founders building clinical decision support must publish rigorous, peer-reviewed prospective outcomes before launching enterprise commercialization.
- Synthetic or single-institution training sets will consistently fail when deployed across international healthcare systems with varying formularies and protocols.
- Never let marketing promises outpace baseline model capabilities in safety-critical domains where errors compromise physician trust and patient safety.
Timeline
What happened to IBM Watson for Oncology?
- What was IBM Watson for Oncology?
- IBM pitched Watson for Oncology as a cognitive supercomputing breakthrough that would democratize expert cancer care by analyzing medical literature in minutes, but the system struggled with real-world patient diversity, relied on manually scripted training, and was ultimately broken up and sold off.
- What happened to IBM Watson for Oncology?
- IBM began marketing Watson for Oncology to clinical institutions globally with high-profile partnership announcements across dozens of cancer centers .
- Why did IBM Watson for Oncology fail?
- In our view, Watson for Oncology failed due to a fundamental model capability gap combined with an aggressive marketing disconnect.
Receipts
- E2IBM's Watson supercomputer to speed up cancer care - BBC News
Supporting quote
IBM's supercomputer Watson will be used to make decisions about cancer care in 14 hospitals in the US and Canada, it has been announced.
- E4IBM Watson Lost MD Anderson, but Has Plenty of Momentum - Voicebot.ai
Supporting quote
In addition, IBM made an announcement in October 2016 with Memorial Sloan Kettering Cancer Center, MIT and Harvard about using Watson with genomic tumor sequencing. IBM also announced an agreement earlier this month with Jupiter Medical Center to “adopt Watson for Oncology trained by Memorial Sloan Kettering.” The solution is scheduled to go live in March.
- E4At ASCO 2017 Clinicians Present New Evidence about Watson Cognitive Technology and Cancer Care
Supporting quote
Watson for Oncology is now available to support the multi-disciplinary care of prostate cancer patients. In Watson for Oncology’s ongoing training with Memorial Sloan Kettering Cancer Center in New York, Watson has now been trained and released to help support physicians in their treatment of breast, lung, colorectal, cervical, ovarian, gastric and prostate cancers. By the end of the year, the technology will be available to support at least 12 cancer types, representing 80 percent of the global incidence of cancer.
- E4IBM pitched Watson as a revolution in cancer care. It's nowhere close
Supporting quote
But three years after IBM began selling Watson to recommend the best cancer treatments to doctors around the world, a STAT investigation has found that the supercomputer isn’t living up to the lofty expectations IBM created for it. It is still struggling with the basic step of learning about different forms of cancer. Only a few dozen hospitals have adopted the system, which is a long way from IBM’s goal of establishing dominance in a multibillion-dollar market. And at foreign hospitals, physicians complained its advice is biased toward American patients and methods of care.
- E4IBM shifts Watson from drug discovery to the clinic
Supporting quote
IBM has struggled to adapt its Watson computing system to drug discovery. It had to halt a marquee installation of Watson for Oncology at the University of Texas MD Anderson Cancer Center in 2016. Both MD Anderson and IBM said they scrapped the project because of procurement improprieties.
- E4Case Study 20: The $4 Billion AI Failure of IBM Watson for Oncology - Henrico Dolfing
Supporting quote
The trajectory of Watson for Oncology can be understood through a sequence of events that illustrate how the program moved from promise to retreat over the course of a decade. The initial collaboration with Memorial Sloan Kettering in 2012 established the foundation, while the launch of Watson Health in 2015 marked the transition into commercialization and global expansion, supported by acquisitions and early deployments that created the appearance of rapid progress. At this stage, the narrative remained coherent, combining technological innovation, institutional credibility, and growing market presence into a story that aligned with both internal strategy and external expectations ( [MSKCC IBM Collaboration Announcement](https://www.mskcc.org/news-releases/mskcc-ibm-collaborate-applying-watson-technology-help-oncologists), [Reuters – Watson Cancer Centers](https://www.reuters.com/article/us-health-watson-idUSKBN0NR2K4)). The first major disruption occurred in 2016, when the University of Texas MD Anderson Cancer Center halted its Oncology Expert Advisor project, another high-profile collaboration with IBM that had consumed approximately $62 million without producing a clinically usable system. This event represented more than a project failure, as it highlighted the difficulty of translating Watson’s capabilities into real-world clinical environments where data quality, workflow integration, and patient variability created challenges that could not be resolved through controlled training alone. While the project was discontinued, the broader program continued to expand, suggesting that the signal was not fully integrated into strategic decision-making ( [IEEE Spectrum – Watson Health Analysis](https://spectrum.ieee.org/ibm-watson-health), [UT System Audit](https://www.utsystem.edu/sites/default/files/offices/board-of-regents/board-meetings/agenda-book-finance-planning/2-2019FAPCpp15-177.pdf)). Between 2017 and 2018, the issues became public, as investigations by STAT reported cases of unsafe or incorrect treatment recommendations and highlighted the gap between IBM’s claims and actual adoption levels. These reports did not introduce new problems but made existing ones visible, shifting the evaluation of Watson from its potential to its performance in real clinical settings. In 2022, IBM sold key Watson Health assets to Francisco Partners for reportedly more than $1 billion, marking the final stage of a strategic retreat in which a program once positioned as a core growth engine was reduced to a set of assets that could be separated and operated independently ( [STAT Investigation – Watson Recommendations](https://www.statnews.com/2018/07/25/ibm-watson-recommendations/), [IBM Watson Health Sale](https://newsroom.ibm.com/2022-01-21-Francisco-Partners-to-Acquire-IBMs-Healthcare-Data-and-Analytics-Assets)).
- E4IBM Watson Health GM: Partnerships set us apart - MedCity News
Supporting quote
Memorial Sloan Kettering Cancer Center [in New York] worked with IBM Watson Health to teach Watson the care protocols the center uses. So in a place that does not have the same level of diagnostics that we have in this country, they can use Watson for oncology. It will say, OK, if you have a patient with this type of cancer, here is the care protocol that MSK uses and the best one for your patient at this time. With MSK, Watson ingested something like 15 million pages of medical content, looked at 200 medical textbooks, read 300 medical journals, along with care pathways from MSK. We have 25 cancer institutes working with us and using Watson for these applications
- E2IBM's Watson to guide cancer therapies at 14 centres | Reuters
Supporting quote
NEW YORK (Reuters) - Fourteen U.S. and Canadian cancer institutes will use International Business Machines Corp's Watson computer system to choose therapies based on a tumour's genetic fingerprints, the company said on Tuesday, the latest step toward bringing personalized cancer treatments to more patients. Oncology is the first speciality where matching therapy to DNA has improved outcomes for some patients, inspiring the "precision medicine initiative" President Barack Obama announced in January. But it can take weeks to identify drugs targeting cancer-causing mutations. Watson can do it in minutes and has in its database the findings of scientific papers and clinical trials on particular cancers and potential therapies. Faced with such a data deluge, "the solution is going to be Watson or something like it," said oncologist Norman Sharpless of the University of North Carolina Lineberger Cancer Center. "Humans alone can't do it." It is unclear how many patients will be helped by such a "big data" approach, however. For one thing, in many common cancers old-line chemotherapy and radiation will remain the standard of care and genomic analysis may not make a difference. Cloud-based Watson will be used at the centres – including Cleveland Clinic, Fred & Pamela Buffett Cancer Center in Omaha and Yale Cancer Center – by late 2015, said Steve Harvey, vice president of IBM Watson Health. The centres pay a subscription fee, which IBM did not disclose. Oncologists will upload the DNA fingerprint of a patient's tumour, which indicates which genes are mutated and possibly driving the malignancy. Watson, recognised broadly for beating two champions of the game show Jeopardy! in 2011, will sift through thousands of mutations and try to identify which is driving the tumour, and therefore what a drug must target. Distinguishing driver mutations from others is a huge challenge. IBM spent more than a year developing a scoring system so Watson can do that, since targeting non-driver mutations would not help. "Watson will look for actionable targets," Harvey said, matching them to approved and experimental cancer drugs and even non-cancer drugs (if Watson decides the latter interfere with a biological pathway driving a malignancy). But Watson has trouble identifying actionable targets in cancers with many mutations. Although genetic profiling is standard in melanoma and some lung cancers, where drugs such as Zelboraf from the Genentech unit of Roche Holding AG target the driver mutation, in most common tumours traditional chemotherapy and radiation remain the standard of care. "When institutions do genetic sequencing, only about half the cases come back with something actionable," Harvey said, often because it is impossible to identify the driver mutation or no targeted therapy exists. The other collaborating centres are Ann & Robert H. Lurie Children's Hospital of Chicago; BC Cancer Agency in British Columbia; City of Hope, in Duarte, California; Duke Cancer Institute in North Carolina; McDonnell Genome Institute at Washington University in St. Louis; New York Genome Center, Sanford Health in South Dakota; University of Kansas Cancer Center; University of Southern California Center for Applied Molecular Medicine, and University of Washington Medical Center.
Sentiment
At launch
No launch coverage quote recorded.
At death
No death coverage quote recorded.
No sentiment quotes recorded.
Last updated · Phil · How we source entries
Changelog
- — Published IBM Watson for Oncology.