Case Study 1: Corporate Decision Markets at TechCorp

Overview

TechCorp, a mid-size software company with 2,000 employees and $500M annual revenue, faces a critical strategic decision: choosing between three product strategies for its next fiscal year. The VP of Strategy has proposed using an internal decision market to aggregate employee knowledge and inform the executive committee's decision.

This case study walks through the complete design, implementation, simulation, and analysis of a corporate decision market, comparing its outcomes to traditional committee decision-making.

Background

The Decision

TechCorp must choose among three product strategies:

• Strategy Alpha: Double down on the existing enterprise platform. Low risk, moderate growth. Expected to leverage existing customer relationships and engineering expertise.
• Strategy Beta: Launch a new AI-powered analytics product targeting mid-market companies. Higher risk, higher potential growth. Requires significant new hiring and technology investment.
• Strategy Gamma: Acquire a smaller competitor and integrate their product line. Moderate risk, provides immediate market share gains but carries integration risk.

The Information Landscape

Different groups within TechCorp have different information relevant to this decision:

No single group has all the information needed to evaluate all three strategies. The decision market aims to aggregate this distributed knowledge.

Traditional Decision Process

Without the market, TechCorp's decision process would be: 1. Each functional leader presents their analysis to the executive committee. 2. The executive committee debates and votes (majority rule). 3. The CEO has tie-breaking authority.

This process suffers from several known biases: - Information filtering (each leader presents selectively). - Status and authority effects (the CEO's preference dominates). - Groupthink (dissenting views are suppressed). - Silo effects (cross-functional information is not integrated).

Market Design

Outcome Metric

The primary metric is 12-month incremental revenue (millions of dollars) from the chosen strategy. This is: - Measurable and auditable (from financial statements). - Relevant to all stakeholders. - Sensitive to the strategy choice. - Available within a reasonable timeframe.

Market Structure

We implement three conditional markets, one for each strategy:

• Market Alpha: "Expected incremental revenue if Strategy Alpha is chosen."
• Market Beta: "Expected incremental revenue if Strategy Beta is chosen."
• Market Gamma: "Expected incremental revenue if Strategy Gamma is chosen."

Each market uses an LMSR with liquidity parameter $b = 200$, covering revenue outcomes from $0 to $200M in 20 buckets ($10M each).

Participation Rules

• All employees may participate.
• Each employee receives 1,000 "TechBucks" (play money) to trade.
• Top 10 performers receive cash bonuses ($5,000, $3,000, $2,000 for top 3, then $500 each for 4th-10th).
• Trading is anonymous to prevent retaliation.
• Market runs for 3 weeks before the executive decision date.

Decision Rule

The market is advisory: the executive committee sees the market's recommendation but retains decision-making authority. However, they commit to publicly explaining any deviation from the market's recommendation.

Simulation

Setting Up the Simulation

We simulate the decision market with agents representing different employee groups. Each group has different information about each strategy's likely outcome.

True expected revenues (known to the simulator but not to agents): - Strategy Alpha: $75M (moderate, reliable) - Strategy Beta: $110M (high mean, high variance) - Strategy Gamma: $85M (moderate, acquisition integration risk)

Agent information structure:

Group signals (mean of their private signal about each strategy):

Sales team:     Alpha=78, Beta=95,  Gamma=80  (underestimate Beta)
Engineering:    Alpha=72, Beta=120, Gamma=70  (see AI potential, doubt acquisition)
Product:        Alpha=80, Beta=115, Gamma=85  (good market research)
Finance:        Alpha=75, Beta=100, Gamma=90  (conservative on Beta, optimistic on Gamma)
Marketing:      Alpha=70, Beta=105, Gamma=88  (see mid-market opportunity)
Executives:     Alpha=85, Beta=90,  Gamma=95  (status quo bias, like the acquisition)

Note the executives overestimate Alpha and Gamma (which they favor) and underestimate Beta. This is a realistic bias -- leadership often favors less disruptive strategies.

Simulation Results

Running the simulation (see code/case-study-code.py for full implementation):

Phase 1 (Days 1-5): Initial trading

Early trading is dominated by the most engaged groups (product managers and engineers). The market quickly moves from uniform priors: - Alpha: $50M -> $65M - Beta: $50M -> $82M - Gamma: $50M -> $68M

Phase 2 (Days 6-15): Information incorporation

As more groups participate, prices converge: - Alpha: $65M -> $74M - Beta: $82M -> $98M - Gamma: $68M -> $80M

The sales team pushes Beta down somewhat (they have weaker signals about AI demand), while finance pushes Gamma up (they see the acquisition as accretive).

Phase 3 (Days 16-21): Convergence

Final market prices: - Alpha: $76M (95% CI: $58M - $94M) - Beta: $103M (95% CI: $65M - $141M) - Gamma: $83M (95% CI: $62M - $104M)

Market recommendation: Strategy Beta

The market recommends Beta by a significant margin ($103M vs. $83M for the second-best option). However, Beta also has the widest confidence interval, reflecting genuine uncertainty about the AI product.

Comparison to Committee Decision

In a parallel simulation of the committee process:

1. Each functional leader presents their analysis.
2. Sales argues for Alpha (reliable, existing customers).
3. Engineering argues for Beta (technical opportunity).
4. Finance argues for Gamma (accretive acquisition).
5. Product argues for Beta (market opportunity).
6. Marketing is split between Beta and Gamma.
7. The CEO favors Gamma (strategic positioning through acquisition).

Committee vote: Gamma wins 6-4 (CEO influence tips the balance).

Analysis

The market correctly identifies Beta as the best strategy. The committee process, dominated by executive preferences, selects the inferior strategy Gamma. The $20M+ gap in expected revenue represents the cost of information loss in the committee process.

Information Revelation Analysis

One of the most valuable aspects of the decision market is not just the final recommendation but the information revealed during trading:

1. Engineering's strong signal on Beta: The market revealed that engineering had very high confidence in the AI product's feasibility -- information that might have been discounted in a committee setting ("of course engineers want to build the new thing").

2. Finance's concern about Beta's variance: The wide confidence interval for Beta was partly driven by finance traders who saw the development costs as risky. This is genuine and useful information -- Beta is higher expected value but also higher risk.

3. Sales' skepticism about AI demand: Sales traders pushed Beta down early, revealing that customer demand signals for AI analytics were weaker than product and engineering believed. This is valuable calibration.

4. Executive bias exposed: The executive traders' positions (favoring Alpha and Gamma) were visible in the market as outliers -- their trades moved prices, but were corrected by the weight of information from other groups.

Lessons Learned

What Worked

1. Anonymous participation allowed junior employees with valuable information to participate without fear of contradicting superiors.
2. Play money with bonuses generated significant engagement (65% participation rate) without the regulatory complexity of real-money markets.
3. Three-week trading window allowed time for information to spread and prices to converge.
4. Clear metric definition (12-month incremental revenue) focused traders on a specific, measurable outcome.

What Did Not Work

5. Executive backlash: Despite pre-commitment, the executive committee was uncomfortable being contradicted by the market. The CEO initially dismissed the result as "engineers gaming the system."
6. Low engineering participation in Gamma market: Engineers had strong negative signals about acquisition integration but under-traded in the Gamma market because they focused on the Beta market where they had positive signals.
7. Late-stage momentum trading: In the final days, some traders appeared to follow the trend rather than trade on information, potentially biasing the final prices.

Recommendations for Future Implementations

• Run the market as genuinely binding (or explain deviations publicly) to maintain trader motivation.
• Use targeted questions to extract specific information ("If we pursue Gamma, what is the probability that integration takes longer than 12 months?").
• Implement market-maker adjustments to ensure all conditional markets receive adequate liquidity.
• Provide training to all employees on how prediction markets work before launching.

Discussion Questions

1. If TechCorp had followed the market's recommendation of Strategy Beta, and Beta had turned out to yield only $60M (the low end of the confidence interval), would that mean the market was "wrong"? How should we evaluate decision quality vs. outcome quality?

2. The market used play money with bonuses. Would real-money trading have changed the outcome? In what direction?

3. How would you modify the market design if TechCorp needed to choose not just which strategy but how much to invest in each (e.g., 60% Beta, 30% Alpha, 10% Gamma)?

4. The executive committee was uncomfortable being contradicted by the market. How would you manage this organizational challenge in a real implementation?

5. If a competitor could observe TechCorp's prediction market prices, what competitive intelligence could they extract? How should TechCorp manage this information security risk?