AI Strategy: How to Choose What AI Product to Implement

Organizations frequently struggle to distinguish between promising AI projects and those destined to fail. Traditional ROI estimates often fail because they conflate two distinct problems: the business value of a successful outcome and the technical uncertainty of achieving that outcome. This leads to a catch-22 where teams cannot estimate ROI without knowing if a project will work, yet they cannot determine if it will work without investing in it. This paper introduces the expected ROI (eROI) framework to resolve this by decoupling these variables.

The eROI framework forces decision-makers to evaluate three pillars separately:

1. Value if Successful: This assesses the strategic alignment, direct product value, option value (future opportunities created), and understanding value (insights gained) of the project.
2. Likelihood of Success: This evaluates the probability of achieving a functional model, accounting for product, technical, data, and scientific certainty. Crucially, this pillar is gated; a low rating in any one of these areas caps the overall likelihood of success.
3. Investment Required: This accounts for the total cost, including team effort, data acquisition, and the often-underestimated costs of integration, productization, and business process reengineering.

By rating these components on a coarse scale (e.g., Low, Medium, High, Very High), teams can create a profile for each project. This allows leaders to identify 'quick wins' (low investment, high likelihood), 'moonshots' (high value, low likelihood), and 'strategic bets' (high value, high investment) without forcing them into a misleading single-number calculation.

Most AI projects fail not because they lack potential value, but because they are technically infeasible or poorly scoped. By explicitly separating 'scientific certainty'—the risk that a model simply cannot be built to the required accuracy—from business value, the eROI framework prevents organizations from over-committing to high-risk projects that lack a clear path to production. It transforms AI strategy from a guessing game into a structured portfolio management process, ensuring that resources are allocated to projects with the best risk-adjusted profiles.