Blog
Measuring the ROI of AI Automation: Four Numbers, Before and After
"It feels like it's helping" is how most automation projects get evaluated, and it's worthless: feelings follow enthusiasm, and enthusiasm fades on schedule whether the system works or not. ROI is measurable here, but only if you capture the baseline before you launch. This post is the checklist.
Before launch: capture four baselines
Spend one week recording these, honestly, before anything goes live:
1. Response numbers. Median time from a new lead or call to first real response, and the share of calls that go unanswered. (The secret-shopper test from speed to lead: inquire on your own website and time it.)
2. Conversion numbers. Of inquiries in a typical month, how many became booked appointments? Of quotes sent, how many closed? Rough counts beat no counts.
3. Leak numbers. No-shows per week. Invoices overdue and average days-to-payment. Missed calls per week (the worksheet).
4. Hours. Ask each person, including yourself: hours per week on phone answering, follow-up, data entry, reminders. People underestimate; count anyway.
Without these, you'll never know what changed, and, just as bad, you won't know which automation to fix or kill.
After launch: measure the same four, plus the system's own health
At 30, 60, and 90 days, re-run the identical numbers. Then add the operational ones your system reports: what share of conversations the assistant resolved end-to-end versus escalated, and where it got stuck. A rising resolution rate means the grounding is improving; a stuck pattern (everyone asks about X, assistant punts on X) is a to-do list, not a failure.
The attribution honesty rule
Not every improvement is the automation. Seasonality, a new ad campaign, a price change, all move the same numbers. Two disciplines keep you honest. Compare against the same period last year where you can, not just last month. And weight the mechanical metrics (response time, answer rate, follow-ups actually sent) over the outcome metrics when attributing: if response time fell from four hours to two minutes and bookings rose, the causal story is strong; if bookings rose while response time didn't change, thank your ad campaign, not your bot.
The verdict math
Total the monthly gain conservatively: recovered bookings at average customer value, hours returned at loaded cost, faster payments at your working-capital value. Set it against the all-in monthly cost. Healthy small-business deployments typically clear several times their cost, and they do it visibly in these numbers, not in vibes. If yours doesn't clear cost by day 90, something specific is wrong (usually grounding gaps or a broken handoff), and the operational metrics will point at it. Fix or kill; don't drift.
We bake this measurement plan into every build, and the baseline capture is part of our free assessment, so the before-photo exists no matter who you end up hiring.