The Easy Way (And Why It's Usually Wrong)
"We built an agent. It should save us 10 hours a week. 10 hours x $75/hour x 52 weeks = $39,000 annual savings. ROI: 300%."
Clean math. Probably wrong.
The agent probably saves you some hours. But maybe not 10. Maybe it saves 6 and you still need someone to review the output. Maybe your team doesn't actually have 10 hours of free time — they're already overloaded doing other things.
The math that feels good is usually the math that's wrong.
How to Actually Measure It
Start with a baseline. "Right now, this task takes Sarah 8 hours a week. She spends 2 hours pulling data, 4 hours processing it, 2 hours reviewing." Write that down.
Deploy the agent. "Now the agent does steps 1-2 in 30 minutes. Sarah still spends 2 hours reviewing." New total: 2.5 hours instead of 8 hours.
That's the real impact: 5.5 hours saved, not 8. At $75/hour, that's $21,450 per year, not $39,000.
Still worth it? If the agent cost $12,000, yeah. You pay for it in seven months.
The Metrics That Actually Matter
Time Savings
Measure this directly. Have someone track their hours before and after for two weeks. Don't estimate. Actually track.
Error Reduction
If the agent catches mistakes, measure that. "We used to miss 5 errors per month. Now we catch them." What's an error worth? If it causes a customer complaint, it might be worth thousands.
Throughput Improvement
"We used to process 20 reports per month. Now we process 30." The extra capacity is valuable even if the agent doesn't reduce total hours.
Quality Improvement
"Reports are more consistent. They follow our standards better. Clients say they're clearer." This is harder to measure but can be worth more than time savings.
What Not to Measure
Don't count freed-up capacity unless you actually plan to use it. If Sarah saves 5 hours a week but you don't have other work for her, that's not savings. That's just... her having less to do.
Don't make up "strategic value." "The agent gives us more time to focus on strategy." Maybe. But unless you actually spend that time on strategy, it's not real value.
The Honest Conversation
If an agent saves 4 hours a week instead of the projected 8, that's not failure. That's data. You learned something about how the work actually happens.
The companies I've worked with that measure honestly are the ones that build successful AI programs. They see what works and what doesn't. They adjust. They iterate.
The companies that lie to themselves about ROI keep building agents that sound good on paper and don't actually move the needle.