Instrument 04 / business case

The ROI is imaginary until failure and review are priced.

Model coverage, failed automations, human review, recurring cost, and implementation payback before an AI agent reaches a budget meeting.

Stress-test one workflow

Price the work that comes back to humans.

Use production observations or a conservative pilot estimate. Coverage is not success, and successful automation is not review-free.

Tasks per monthThe workflow volume available to humans or the agent. tasks Human minutes per task todayUse observed handling time, including normal corrections. min Loaded hourly costSalary, benefits, management, and operating overhead. $ Automation coverageThe share of tasks the agent will actually attempt. % Success rate inside coverageThe attempted tasks that pass the real acceptance test. % Human review per attempted taskInclude checking, correction, approval, and exception routing. min AI run cost per attempted taskModel, retrieval, tool, and other usage-based costs. $ Other recurring monthly costPlatform, monitoring, maintenance, and support. $ One-time implementation costIntegration, evaluation, security, rollout, and change management. $

Sensitivity check

The base case is one story, not the truth.

Conservative reduces coverage and success while increasing review and cost. Upside improves them modestly. A decision should survive more than one row.

Conservative$0—

Base$0—

Upside$0—

Model boundaries

Coverage is not success.
Only completed tasks remove the original labor.

Review is real labor.
Every attempted task may create checking and exception work.

Payback is not certainty.
Validate the inputs in a bounded pilot before approving scale.

Transparent method

What this AI agent ROI calculator actually computes.

Baseline labor is task volume multiplied by current handling time and loaded hourly cost.

Cost after AI keeps every uncovered or failed task in the human queue, adds review labor to every attempted automation, then adds usage-based and recurring costs.

Monthly savings is baseline labor minus cost after AI. Payback divides one-time implementation cost by positive monthly savings. Year-one ROI subtracts implementation cost from twelve months of savings.

This is a planning model, not a forecast or guarantee. It excludes revenue lift, quality changes, taxes, financing, and risk-adjusted discounting. Replace every example input with pilot evidence.

Decision guide

Questions to answer before approving AI agent ROI.

A spreadsheet can make any agent look efficient. A defensible business case keeps failed work, review labor, recurring operations, and uncertainty visible.

Read the AI agent ROI calculation guide →

How do you calculate AI agent ROI?

Compare current labor cost with cost after automation, keeping failed tasks and uncovered work in the human queue. Subtract review, usage, recurring, and implementation costs, then test the result across conservative, base, and upside scenarios.

What costs belong in an AI agent ROI model?

Include implementation, model and tool usage, retrieval, infrastructure, monitoring, maintenance, human review, exception handling, and the labor that remains when tasks fail or sit outside automation coverage.

What is a reasonable AI agent payback period?

There is no universal threshold. Set one that matches the workflow risk, expected lifetime, and confidence in the inputs. Require the project to stay acceptable under conservative pilot assumptions—not only the base case.