AI initiatives create value only when they are prioritised, governed and scaled systematically.
A framework for prioritising AI initiatives across value, feasibility, risk and strategic impact
Many organisations are currently exploring artificial intelligence across a wide range of potential applications. These initiatives often emerge from local experimentation within business units or technology teams rather than from a structured strategic process.
While such experimentation can generate valuable insights, it frequently leads to fragmented portfolios of pilots, prototypes and isolated solutions. As a result, organisations increasingly face the challenge of determining which initiatives should be scaled, which should remain experimental, and which should be discontinued.
One way to address this challenge is to treat AI initiatives not as isolated projects but as part of a managed portfolio of capabilities. In such a portfolio, potential AI use cases are assessed along a consistent set of dimensions that capture both value creation and operational risk.
The framework presented here evaluates AI initiatives across four key dimensions.
The portfolio is structured along two axes that capture the value of AI initiatives and their feasibility in real-world deployment.
1. Value & Scale (x-axis)
The first dimension concerns the economic leverage of an AI application.
Key questions include:
- How frequently does the underlying problem occur?
- How labour-intensive is the process today?
- To what extent does the benefit scale with volume?
Use cases addressing high-frequency, resource-intensive activities typically offer the greatest potential for economic impact. Conversely, use cases with limited scale or marginal efficiency gains may not justify the organisational effort required for implementation.
2. Technical Feasibility & Explainability (y-axis)
The second dimension addresses the practical reliability of AI in a given context. In regulated environments, technical feasibility alone is often not sufficient – models must also be explainable and defensible.
Relevant factors include:
- Availability and quality of relevant data
- Robustness and reproducibility of model outputs
- Maturity of the underlying technology
- Explainability and interpretability of model outputs
Even conceptually attractive use cases may remain difficult to operationalise if data availability is limited, model behaviour remains unstable or results cannot be sufficiently explained. Technical feasibility and explainability therefore determine whether a use case can realistically move beyond experimentation and be deployed in a controlled environment.
3. Governance & Risk
AI systems can introduce new forms of operational, regulatory and reputational risk. As organisations increase the degree of automation in decision processes, governance considerations become increasingly important.
While explainability is a prerequisite for deployment, governance determines how AI systems are controlled, monitored and limited in practice.
A high position on the feasibility y-axis indicates that a use case can be implemented reliably and explainably. However, governance considerations determine whether it can be fully automated, requires human oversight or is subject to deployment restrictions.
Important aspects include:
- Potential worst-case impact of incorrect AI decisions
- Degree of regulatory exposure
- Requirements for human oversight and control mechanisms
Where potential consequences are significant, additional safeguards may be required, such as human-in-the-loop decision processes, monitoring mechanisms or clear accountability structures.
AI multiplies value, but it can also multiply mistakes.
4. Strategic Impact (x-axis)
Finally, organisations should assess whether an AI initiative merely improves existing processes or contributes to the development of new strategic capabilities.
Key considerations include:
- Quality improvements in analysis or decision-making
- Opportunities for differentiation
- Compatibility with the organisation’s target operating model
- Organisational complexity required for implementation
Some AI applications primarily deliver efficiency gains, while others enable entirely new ways of operating or interacting with clients.
From pilots to a managed portfolio
Applying these dimensions allows organisations to structure AI initiatives into a clear portfolio.
While value and feasibility determine the position of a use case in the matrix, governance and risk define the boundaries within which it can be deployed in practice.
Scale
High value, high reliability and manageable risk.
Pilot
Promising potential, but technical or organisational uncertainty.
Assist
AI supports human decision-making rather than replacing it.
Stop
Limited value or excessive risk.
This portfolio perspective helps organisations focus investments, control risk and concentrate on initiatives with the highest strategic impact.