AI agents and business automation
How to measure AI agent ROI without inventing the savings
Count released capacity, avoided delay, quality and operating cost with enough discipline to support a decision.
Establish the current unit economics
Measure volume, touch time, wait time, rework, error, escalation and the blended cost of the people involved. This matters because AI agent ROI decisions rarely fail through a lack of possible technology. They fail when the business problem, operating context and responsibility for the outcome remain implicit. Bring evidence from the people doing the work, the systems supporting it and the leaders accountable for the result. Test assumptions about time, behaviour, data quality and implementation effort before treating them as facts. Use observed samples where system data cannot describe the real workflow. Record the choice, the evidence still required and the person who will return with it. Keep the mechanism proportionate: the purpose is better judgement and follow-through, not additional ceremony.
Separate time saved from money saved
Released hours create value only when the business can redeploy capacity, absorb growth, improve service or remove cost. This matters because AI agent ROI decisions rarely fail through a lack of possible technology. They fail when the business problem, operating context and responsibility for the outcome remain implicit. Bring evidence from the people doing the work, the systems supporting it and the leaders accountable for the result. Test assumptions about time, behaviour, data quality and implementation effort before treating them as facts. Do not present every minute saved as cash returning to the profit and loss account. Record the choice, the evidence still required and the person who will return with it. Bring specialist judgement into the decision where required while retaining business ownership of the outcome.
Include the new operating cost
Count infrastructure, licences, usage, integration, monitoring, assurance, support, review and continuing improvement. This matters because AI agent ROI decisions rarely fail through a lack of possible technology. They fail when the business problem, operating context and responsibility for the outcome remain implicit. Bring evidence from the people doing the work, the systems supporting it and the leaders accountable for the result. Test assumptions about time, behaviour, data quality and implementation effort before treating them as facts. Compare steady-state cost as well as the attractive cost of a small pilot. Record the choice, the evidence still required and the person who will return with it. Keep the mechanism proportionate: the purpose is better judgement and follow-through, not additional ceremony.
Measure quality and consequence
Track accuracy, completeness, customer impact, correction effort and the cost of errors that travel downstream. This matters because AI agent ROI decisions rarely fail through a lack of possible technology. They fail when the business problem, operating context and responsibility for the outcome remain implicit. Bring evidence from the people doing the work, the systems supporting it and the leaders accountable for the result. Test assumptions about time, behaviour, data quality and implementation effort before treating them as facts. Faster poor work can destroy more value than slow reliable work. Record the choice, the evidence still required and the person who will return with it. Bring specialist judgement into the decision where required while retaining business ownership of the outcome.
Attribute value cautiously
Use control groups, before-and-after samples or matched workflows where practical, while recording other changes that affected performance. This matters because AI agent ROI decisions rarely fail through a lack of possible technology. They fail when the business problem, operating context and responsibility for the outcome remain implicit. Bring evidence from the people doing the work, the systems supporting it and the leaders accountable for the result. Test assumptions about time, behaviour, data quality and implementation effort before treating them as facts. A credible range with stated assumptions is more useful than a precise figure nobody trusts. Record the choice, the evidence still required and the person who will return with it. Keep the mechanism proportionate: the purpose is better judgement and follow-through, not additional ceremony.
Agree the value-realisation action
Name how released capacity will be used and who is accountable for turning operational improvement into a business outcome. This matters because AI agent ROI decisions rarely fail through a lack of possible technology. They fail when the business problem, operating context and responsibility for the outcome remain implicit. Bring evidence from the people doing the work, the systems supporting it and the leaders accountable for the result. Test assumptions about time, behaviour, data quality and implementation effort before treating them as facts. The agent creates potential value; management choices determine whether the business captures it. Record the choice, the evidence still required and the person who will return with it. Bring specialist judgement into the decision where required while retaining business ownership of the outcome.
What to carry into the work
- Use observed samples where system data cannot describe the real workflow.
- Do not present every minute saved as cash returning to the profit and loss account.
- Compare steady-state cost as well as the attractive cost of a small pilot.
- Faster poor work can destroy more value than slow reliable work.