Building Watson: Not So Elementary, My Dear! Custom Case Solution & Analysis

Evidence Brief

Financial Metrics

Research Investment: IBM allocated significant capital to the DeepQA project over four years. While the exact dollar figure remains internal, the project utilized 25 to 30 full-time research scientists (Source: Paragraph 4).
Hardware Infrastructure: The system ran on a cluster of ninety IBM Power 750 servers. This configuration included 2,880 POWER7 processor cores and 15 terabytes of RAM (Source: Exhibit 4).
Market Value: IBM shares traded at approximately 125 USD at the start of the challenge in 2007 and rose to over 160 USD by the 2011 match (Source: Financial Market Data).

Operational Facts

Content Volume: The system ingested 200 million pages of structured and unstructured content, including dictionaries, encyclopedias, and movie scripts (Source: Paragraph 12).
Processing Speed: The DeepQA architecture processed thousands of simultaneous language analysis algorithms to find answers in less than three seconds (Source: Paragraph 15).
Success Rate: In 2007, the system could only answer 15 percent of questions correctly. By 2010, the accuracy matched the best human players at approximately 85 percent (Source: Exhibit 2).
Geography: Development occurred primarily at the Thomas J. Watson Research Center in Yorktown Heights, New York.

Stakeholder Positions

David Ferrucci: Principal Investigator. Focused on the scientific integrity of Natural Language Processing. He maintained that the system does not think but rather calculates probabilities (Source: Paragraph 8).
John Kelly: Senior Vice President of IBM Research. Viewed the project as a vehicle to demonstrate the viability of the Smarter Planet initiative (Source: Paragraph 3).
Samuel Palmisano: CEO of IBM. Positioned the challenge as a successor to the Deep Blue chess match to revitalize the brand of the firm (Source: Paragraph 2).
Ken Jennings and Brad Rutter: Competitors. Represented the benchmark for human cognition in broad-domain retrieval (Source: Paragraph 20).

Information Gaps

Commercial Pricing: The case does not provide a projected price point for a commercial Watson unit.
Operating Costs: Ongoing energy consumption costs for the 90-server cluster are not specified.
Data Acquisition Costs: The cost to license proprietary medical or financial databases for future use is absent.

Strategic Analysis

Core Strategic Question

The primary dilemma involves the transition of a specialized game-playing architecture into a profitable, scalable enterprise solution.
IBM must determine if the DeepQA architecture can provide superior utility in narrow, high-stakes fields like healthcare compared to broad-domain information retrieval.

Structural Analysis

The Value Chain analysis reveals that the primary strength of IBM lies in R and D and brand signaling. However, the outbound logistics and service components for a commercial AI product are underdeveloped. The Jobs-to-be-Done lens suggests that customers in the medical field do not need a trivia bot; they need a decision-support tool that manages uncertainty and cites evidence. The current system provides probabilities, which aligns with diagnostic needs but requires a shift from speed to precision.

Strategic Options

Option 1: The Healthcare Specialist. Focus exclusively on oncology and chronic disease management.
- Rationale: High willingness to pay and massive volumes of unstructured clinical data.
- Trade-offs: Requires long regulatory cycles and high liability risks.
- Resource Requirements: Deep partnerships with medical institutions and hiring of clinical experts.
Option 2: The Horizontal API. Offer the DeepQA capabilities as a cloud service for various industries.
- Rationale: Rapid scaling and lower entry barriers per segment.
- Trade-offs: Dilution of the brand and difficulty in maintaining accuracy across unrelated domains.
- Resource Requirements: Significant investment in cloud infrastructure and developer support.

Preliminary Recommendation

IBM should pursue the Healthcare Specialist path. The margins in enterprise search are commoditized by existing web players. High-stakes medical diagnostics utilize the core strength of the system: processing vast unstructured data to find obscure connections. This path justifies the high price point required to recoup R and D costs.

Implementation Roadmap

Critical Path

Month 1-3: Transfer the DeepQA core team from the Research division to a newly formed commercial business unit. This prevents the loss of institutional knowledge.
Month 4-6: Initiate data ingestion of PubMed and clinical trial records. The system must move from general knowledge to specialized medical literature.
Month 7-12: Launch a pilot program with a major cancer research center. The goal is to achieve a 90 percent correlation between Watson recommendations and expert board decisions.

Key Constraints

Data Quality: The system is only as effective as the corpus it reads. Inconsistent electronic health records will degrade performance.
Talent Retention: Research scientists may prefer academic freedom over the rigid requirements of commercial software development.

Risk-Adjusted Implementation Strategy

The plan assumes a staggered rollout. Rather than a full diagnostic launch, the first version will act as a research assistant for clinicians. This reduces liability while the system learns. Contingency includes a secondary focus on the financial services sector if medical regulatory hurdles prove insurmountable within 24 months.

Executive Review and BLUF

BLUF

IBM should immediately pivot Watson from a marketing showcase to a specialized medical diagnostic tool. The Jeopardy victory proved the technical capability of the system to handle unstructured data at scale. However, the long-term value of the firm depends on solving high-value problems that commoditized search engines cannot touch. Healthcare offers the necessary density of unstructured data and a high willingness to pay. Success requires moving beyond the speed-based requirements of a game show to the precision-based requirements of clinical practice. Failure to narrow the focus will result in a tool that is interesting but not indispensable.

Dangerous Assumption

The most consequential premise is that the ability to win a trivia game translates directly to professional expertise. Trivia relies on the existence of a known answer within a static corpus. Medicine requires reasoning through incomplete, contradictory, and evolving patient data. The system may struggle when there is no single correct answer to find.

Unaddressed Risks

Regulatory Barrier: The FDA or equivalent bodies may classify the system as a medical device, leading to years of delays and unforeseen compliance costs.
Economic Viability: The cost to maintain the massive hardware requirements may exceed the efficiency gains for the average hospital, limiting the market to only the largest institutions.

Unconsidered Alternative

The team did not fully evaluate the option of a pure IP licensing model. Instead of building products, IBM could license the DeepQA architecture to existing industry leaders in legal research, finance, and medicine. This would offload the execution risk and capital expenditure of building a sales force while generating high-margin royalty streams.