AI Project ROI: How to Measure, Calculate & Justify AI Investment

Q: What is a realistic ROI for an enterprise AI project in Year 1?

For the majority of enterprise AI initiatives, Year 1 ROI is negative or marginally positive. The investment in infrastructure, integration, and model development typically exceeds the returns visible within a 12-month window. This is expected and not a signal of project failure. The relevant question in Year 1 is not "what is our ROI?" but "are we on track with the leading indicators that predict Year 2 and Year 3 returns?" — adoption rate, accuracy metrics, integration completeness, and operational stability.

Q: How do you calculate ROI for AI projects that replace human work indirectly rather than directly?

Indirect displacement — where AI augments human workers rather than replacing defined tasks — requires tracking time reallocation. If a knowledge worker spends 30% less time on retrieval and synthesis tasks, document how that time is redirected. If it moves to higher-value strategic work, the ROI is the value of that output (which itself must be quantified). Structured time studies before and after deployment are the most defensible methodology for indirect displacement cases.

Q: How should we handle AI ROI when the benefits span multiple departments?

Cross-departmental AI systems require a platform ROI model rather than a project ROI model. Calculate the total cost of the platform and the aggregate returns across all business units using it. Allocate platform costs to business units by usage volume. Each business unit presents its own return calculation against its allocated cost. This approach makes the platform amortisation effect explicit — a key driver of Year 2 and Year 3 returns.

Q: What is the difference between measuring AI ROI and measuring AI success?

AI success is a technical and operational concept — the model is accurate, reliable, and performing as designed. AI ROI is a business and financial concept — the system is generating measurable value that exceeds its cost. A technically successful AI system can have poor ROI if the use case does not generate sufficient value. Measuring both separately is important. Conflating them leads to defensive conversations with finance where technical metrics are presented as evidence of financial return.

Q: At what point should an AI project be stopped if ROI is not materialising?

Establish explicit review gates before the project starts. At Month 6, assess whether deployment milestones and leading indicators are on track. At Month 12, assess whether Category 1 returns are beginning to materialise. At Month 18, if returns are still not visible and leading indicators remain weak, stopping is the rational decision. Build the stop criteria into the original governance documentation — the most expensive AI projects are failed ones that nobody wants to be responsible for ending.

Q: How do we justify AI investment when the primary value is competitive parity rather than net-new returns?

Competitive parity is a legitimate justification that requires a different framing. Present it as the cost of inaction: what market share, customer satisfaction differential, or talent acquisition disadvantage does the organisation accept by not investing? This turns the ROI question from "what return does this generate?" to "what is the cost of the alternative?" — which is often a much larger number and a more compelling board conversation.