The AI ROI we miss: better decisions, not only hours saved
Measuring AI only through hours saved excludes much of its value and can reward the wrong work. Returns also appear in quality, decision speed, learning, avoided risk and new possibilities.
“How many hours does it save?” is a reasonable question. It can also be a trap. It is easy to explain, fits neatly in a spreadsheet and promises to turn technology into money. But not all released time becomes a cost reduction, and not all value appears as time.
Someone may prepare an analysis in half the time and use the remainder to review it more carefully, handle more cases or explore an opportunity that previously never reached the table. The economic outcome depends on what happens after the saving.
Seven forms of value
I propose broadening the conversation with seven categories:
- Time: less effort or less waiting.
- Quality: fewer errors, greater consistency or better deliverables.
- Decision speed: evidence arrives while it can still change an action.
- Capacity: more cases handled without growing at the same rate.
- Learning: hypotheses tested and knowledge shared more frequently.
- Risk: earlier detection, traceability or reduced exposure to failure.
- Possibility: products, services or decisions that were not viable before.
Not every case needs to score across all seven. The point is to prevent one metric from making the rest invisible.
Measure the workflow, not the demonstration
A demo can produce an answer in seconds. The real process includes preparing information, reviewing, correcting, integrating, communicating and handling exceptions. ROI should compare the complete cycle before and after.
It should also include new costs: licences, consumption, integration, governance, training, review and operations. If a tool creates more volume than the team can validate, the assumed saving may move work instead of removing it.
Baseline, change and consequence
To measure a case, I need three snapshots:
- Baseline: how it works today, with what variation and which problems it produces.
- Intervention: which part changes and what it costs to sustain.
- Consequence: which business outcome or capability appears afterwards.
The third is often missing. Saving twenty minutes when preparing a proposal creates value only if it improves the customer response, increases capacity, reduces errors or frees attention for something more valuable.
The quality of a decision
We cannot always know immediately whether a decision was right. We can measure whether it arrived with better information, made its assumptions explicit, considered more alternatives and left a trail from which we can learn.
In strategy and product, that improvement may be worth more than the time saved. A bad idea executed quickly remains a bad investment. A prototype that allows us to discard it early may save no hours in the specific task and still prevent months of work.
A portfolio, not a magic number
Early AI cases should form a portfolio: some seek efficiency, others quality and others new options. Demanding the same type of return from every case removes precisely the experiments that might transform the business.
Discipline does not mean abandoning ROI. It means declaring which form of value we expect and what evidence we will accept before we begin.