Bottom line
The standard AI ROI model goes like this: estimate the hours saved per employee per week, multiply by the number of employees, multiply by the average hourly rate, arrive at an annual productivity saving that justifies the investment several times over. This model is almost always wrong — not because the hourly estimates are incorrect, but because hours saved do not reliably translate into value created.
The productivity illusion.
When an AI tool saves two hours per employee per week, three things can happen. First, those hours are genuinely redirected to higher-value work — the intended outcome. Second, those hours are absorbed by existing workload without any observable change in output — the common outcome. Third, those hours are invisible to the organization because knowledge work is not managed at hourly precision — the most common outcome. In most cases, the productivity saving is real but the value creation is not, because the organization has not designed the workflow change that converts saved time into incremental output.
Where AI value is real and measurable.
The AI applications that produce real, measurable value share a specific characteristic: they either directly produce valuable outputs (replacing a task that previously required hours of human work with an AI output that takes minutes) or they directly improve the quality of high-value human outputs (improving the decisions that shape revenue, cost, or risk). The first category — direct output substitution — is where most AI productivity measurement focuses. The second category — decision quality improvement — is where the larger value typically lives, but it is harder to measure and rarely appears in standard ROI models.
More on Business Value — every week.
Join 2,400+ leaders getting the decision intelligence briefing.
The decision quality ROI model.
For organizations that use AI to improve strategic decision quality, the value case is built differently. It requires quantifying the value at stake in the decisions being improved — the revenue decisions, the investment decisions, the hiring decisions — and applying a realistic estimate of how much the decision quality improvement reduces the error rate. A 5% improvement in the quality of decisions that control €100M in annual revenue is worth €5M, without any hourly productivity measurement required. This model is harder to construct but it reflects where the value actually is.
The honest measurement agenda.
Organizations that want to measure AI value honestly should ask: which decisions does AI influence, what are those decisions worth, and what evidence do we have that the decision quality has improved? This is a different measurement agenda from time-and-motion productivity analysis — but it is the one that captures the real value of AI investment and creates the organizational clarity needed to direct that investment to where it matters most.
Research context
Among surveyed organizations using AI, 71% reported revenue gains in marketing and sales, although the most common increase was below 5%.
Stanford Institute for Human-Centered AI: The 2025 AI Index Report: Economy (2025)This independently published statistic provides context for the topic; it does not substantiate every claim in this essay.
More on Business Value, weekly.
One essay. Every week. No noise.