AI agents: how to build a business case your CFO will approve

Start with the work, not the technology

An agent is worth exactly the value of the work it removes or improves. So the first question is not "which model?" but "which process?". The best candidates share four traits:

• High volume. Hundreds or thousands of similar items a month, such as invoices, applications, tickets or reports.
• Rules plus judgement. Enough structure for an agent to follow, enough variation that simple automation has failed.
• Digital inputs. Email, documents, forms and system records rather than phone calls and paper.
• Measurable outcomes. Cycle time, cost per item, error rate or backlog that someone already tracks.

Measure the baseline before anything is built

Without a baseline there is no business case, only a story. Capture, for at least two to four weeks of normal operations:

• volume per week and how it varies
• average handling time and the people involved
• cycle time from arrival to completion
• error, rework and exception rates
• the cost of delay, such as late fees, lost revenue or customer churn

Size the value honestly

Value usually comes from four places: hours returned to skilled people, faster cycle times, fewer errors and better decisions. Model a conservative case, assuming the agent handles only the straightforward share of items and people still review exceptions. If the conservative case does not pay back within a year, choose a different process.

Count the full cost

Include build, integration, testing, security review, change management and the running costs of models, hosting and monitoring. Agents are software that needs owners, so include the ongoing effort to evaluate and improve them.

Agree the stop criteria up front

The single most useful line in an agent business case is the one that says when to stop. Agree, before building, the target metric, the threshold that counts as success, and the date you will decide. A pilot that misses its target is not a failure if it is stopped on time and the lessons are kept.

The question for the board is not "should we use AI agents?" but "is this agent, on this process, worth more than it costs?"

Present it on one page

1. The process and why it matters
2. The baseline, measured
3. The conservative and expected value
4. The full cost to build and run
5. Risks and controls, including human approval and security testing
6. The pilot plan, target metric and stop criteria