From recommendation to realisation: closing the loop on Return on Decision

Most organisations can tell you what an AI or decision intelligence tool recommended. Far fewer can show what value was generated because of it. 

Return on Decision closes that gap. Faculty Frontier™, our Decision Intelligence Platform, is built around this idea from the ground up. Value doesn't come from a recommendation on its own. It comes from helping a business improve a specific decision, act on it, measure the result, and use what it learns to improve the next one. That loop is not a feature we bolted on. It is the design principle behind how Frontier works.

The missing metric for enterprise AI

Return on Investment captures the overall financial return from an investment. Return on Decision shows which decisions contributed to that return, and how. It links the recommendation to what the business accepted, changed or rejected, who acted on it, and how the result compared with the original baseline.

That matters to anyone accountable for enterprise performance. A CFO wants to know which decisions contributed to the financial result and how much expected value was realised. An operational leader needs to see whether an agreed change actually reached the business. A CIO or AI leader needs evidence that their investment in technology is improving decisions that matter, not just generating more questions to answer. 

The need for that evidence is growing. Gartner reported average spending of $1.9 million on GenAI initiatives in 2024, while fewer than 30% of AI leaders said their CEOs were happy with the return [1]. The Global AI Forum recently cited MIT NANDA's finding that 95% of enterprise GenAI pilots return no measurable P&L impact. [2].

Decisions as the unit of value

As Tom Oliver, Head of Product at Faculty, explored in his recent blog, decisions - not models or analyses - are the fundamental unit of value in an enterprise. Return on Decision starts by making those decisions visible and explicitly defined. “Improve supply chain planning” is too broad to measure. “Decide how next year’s production volume should be allocated across the factory network” identifies the decision, its owner, the baseline and the outcome to measure.

The same logic applies to capital investment, pricing, clinical development and workforce decisions. Without a defined decision, expected value stays attached to a broad use case or initiative rather than a business decision that is owned and assessed.

Usage metrics do not provide that evidence. Logins, sessions and scenario runs show that people interacted with the product. They cannot tell leaders whether a plan changed, capacity shifted, costs came down, or risk was reduced.

From recommendation to realised value

Closing the loop requires a clear process around each decision. Operations, finance and the relevant decision owners review the recommendation, test its assumptions, weigh the trade offs and agree what to take forward. Governance creates a regular executive review of the decisions supported, the value expected, what the business committed to and what has since been delivered, giving finance, operations and technology leaders a shared view of where value came from and where it fell short.

Planning, finance, ERP and operational systems hold the execution and transactional data needed to show whether the agreed action was implemented and some of what followed. Bringing that data back into the decision process gives leaders a consistent view of expected, committed and realised value.

Take the production allocation decision described earlier. Frontier might recommend moving production volume between factories to reduce overall cost while still meeting customer demand. Producing more at a lower cost factory could increase transport costs or create capacity pressure elsewhere.

Operations and finance therefore need to assess the recommendation across the whole network. They decide which changes to accept, what value to commit to and who will carry them out. Execution data then shows whether those changes reached the operating plan, and how the result compared with the baseline.

Compounding value, one decision at a time

The learning step turns each outcome into better inputs for future decisions. A capacity limit identified during execution can become a constraint in the next scenario. A decision to accept a lower saving to ensure products remain available can be reflected in future recommendations. Measures that failed to capture the outcome that mattered can also be changed. This is what makes the loop compound. Frontier learns. The business learns. And every decision after that gets a little more intelligent than the one before it.

This feedback loop will become more important as AI supports more business decisions. Gartner predicts that half of business decisions will be augmented or automated by AI agents using decision intelligence by 2027 [3]. At that scale, value will depend on whether organisations learn from each decision and use its outcome to improve the evidence, constraints and measures behind the next recommendation. That is what Return on Decision measures, and why I believe it is about to become the number that separates leaders who can prove their AI investment from those who can only describe it.

References

[1] Gartner, The Latest Hype Cycle for Artificial Intelligence Goes Beyond GenAI, 8 July 2025, section “Gen AI enters the Trough of Disillusionment.”

[2] MIT Project NANDA, The GenAI Divide: State of AI in Business 2025 (2025), as cited in Global AI Forum, The State of Enterprise AI 2026, https://gaiforum.com/state-of-enterprise-ai-2026.”

[3] Gartner, Gartner Announces the Top Data & Analytics Predictions, 17 June 2025.