answers
How do I measure ROI on an AI agent?
Measure first-year AI agent ROI as (12 x net monthly benefit - build cost) divided by build cost, where net benefit is hours saved plus added revenue minus run and review costs, against a baseline taken before launch. Pick one unit of business value, such as cost per qualified lead or hours per client update. Metrics table and worked example.
Measure first-year ROI on an AI agent as (12 × net monthly benefit - build cost) ÷ build cost, where net monthly benefit is the value of hours saved plus any added revenue, minus run and review costs, all compared with a baseline you measured before the agent went live. Pick one unit of business value the agent moves, such as hours per client update, time to first reply, or cost per qualified lead, and track it before and after.
Without a baseline, you do not have an ROI, you have an opinion. Measure the process for one to two weeks before launch.
Updated 2026-09-24 by Esteban Valencia, EV Advisory. Written for founder-led businesses in Canada and the United States.
What should you measure?
| Metric | How to measure it | Baseline needed |
|---|---|---|
| Hours spent on the process | Time tracking or a simple log for one to two weeks | Yes, before launch |
| Volume handled | Count of runs (enquiries, updates, tickets) per week | Yes |
| Speed | Time from trigger to completion, for example time to first reply | Yes |
| Quality | Correction rate: share of agent outputs a person changes | Measure from launch; compare with past error rate if known |
| Revenue effect | Conversion rate or revenue from the affected work, for example leads to booked calls | Yes, ideally a month or more |
| Run cost | Model usage, subscriptions, and support per month | Estimate before launch, actual after |
Choose one primary metric that matches the agent’s job. The others are guardrails.
How do you set a baseline before launch?
- Pick the metric the agent should move.
- Log the current process for one to two weeks: how many times it ran, how long each took, who did it.
- Note the quality level today, for example how often replies were late or reports had errors.
- Write it down with the date so the after measure is compared with the same thing.
If the process is too irregular to baseline in two weeks, it is probably not the right first agent.
Which costs belong in the ROI calculation?
- Build or setup cost. The implementation fee, or DIY hours at a realistic rate.
- Monthly run cost. Model usage, tool subscriptions, and maintenance. Published Canadian maintenance arrangements run about $300 to $800 a month.
- Review time. The hours a person spends checking the agent’s output. This is real and often forgotten.
- Change cost. Time spent on adjustments when your process or a connected tool changes.
A worked illustration (made-up numbers, to show the arithmetic): an agent saves 6 hours a week of account manager time at a loaded cost of $60 an hour, which is about $1,560 a month. It costs $400 a month to run and takes 1 hour a week of review ($260 a month). Net monthly benefit: about $900. With a $9,000 build, it pays back in about 10 months, and year-one ROI is about 20% ($10,800 net benefit on $9,000). Your numbers will differ; the method is what matters.
How long before the return shows up?
For a narrow, single-workflow agent, time savings usually show within the first month of daily use, once the early tuning is done. Payback on the build cost commonly takes 6 to 12 months for a small business, depending on volume. Revenue effects, such as faster replies turning into more booked calls, take longer to read because you need enough volume to see a real change.
If the primary metric has not moved after three months of daily use, stop and diagnose. Either the workflow was the wrong pick or the agent needs redesign. Do not keep paying for an agent because it is already built.
Sources
- Planwebbies, AI Automation Cost in Canada: Small Business Pricing (2026), published August 5, 2026.
- Innovation, Science and Economic Development Canada, Toolkit for SMEs deploying artificial intelligence.
Next step
Every EV Advisory project includes a baseline before the build and the same measure after launch, so the return is visible rather than assumed. See how on the AI agent implementation page, or book a free 20-minute Live Audit to find a first workflow with a measurable payoff.
FAQ
Questions operators usually ask
What if the benefit is time saved, not revenue?
Price the time. Multiply hours saved per month by the loaded hourly cost of the people who did the work (wage plus overhead). Then ask the honest follow-up question: what did those hours go to instead? Time saved that goes into billable or sales work is worth more than time that simply disappears.
Should I count the cost of my own time?
Yes, on both sides. Count the hours you and your team spent scoping, testing, and reviewing as a cost, and count founder hours saved as a benefit at a realistic hourly value. Founders who leave their own time out usually overstate the return of DIY projects.
How often should ROI be reviewed?
Monthly for the first three months, while the agent is being tuned and your team is adjusting, then quarterly. Review it again whenever the underlying process or a connected tool changes.