Day 13 · "AI for Business" series
"Okay, sounds cool. But how much will I actually save?"
It's the most frequent question I hear from business owners. And it's the right one — because AI isn't magic, it's a tool. And a tool has a cost and a return.
Let's do the math. On three real examples.
#How to calculate ROI for an AI agent
The formula is simple:
ROI = (Savings − Cost) / Cost × 100%
Where savings is the money or time the agent frees up, and cost is setup plus the monthly running fee. If ROI > 100% in the first month, the agent has paid for itself. If it's > 500%, you're obligated to deploy it.
#Example 1: A coffee shop
Say "Beans & Beavers." 3 baristas, 200 orders a day.
Task delegated to the agent: taking orders via Telegram.
Before: a barista gets pulled to the phone, scribbles the order in a notepad, passes it to the kitchen. 3 minutes per order × 50 phone orders = 2.5 hours of barista time a day.
Now: the bot takes the order, logs it to Google Sheets, pings the kitchen. The barista just hands over the finished coffee.
Savings: ~2.5 barista-hours a day ≈ $250/month in freed labor.
Cost: $110 one-time setup + $5/month for the server.
| Period | ROI |
|---|---|
| Month 1 | (250 − 115) / 115 = 117% |
| Month 2+ | (250 − 5) / 5 = 4,900% |
Yes. Nearly five thousand percent — because the ongoing cost is just a $5 server.
#Example 2: An online store
A pet-supply shop, "Tails." 50 orders a day, 2 support managers.
Task: answer routine questions (where's my order, how to return, sizes).
Before: a manager handles ~80 tickets a day. 60 of them are routine. 5 minutes per ticket = 5 hours a day.
Now: the AI agent handles the routine ones. Complex cases get escalated to a human.
Savings: the manager now handles only hard cases and can take on 30% more orders without a second hire ≈ $650/month.
Cost: $320 one-time setup + $180/month (subscription + API).
| Period | ROI |
|---|---|
| Month 1 | (650 − 500) / 500 = 30% — modest |
| Month 2+ | (650 − 180) / 180 = 261% |
Pays off in the second month. After that — pure savings.
#Example 3: A freelancer
Me. The exact case this blog started with.
Tasks I delegated: client support, auto-posting, and server monitoring.
Savings: 40 hours/month (see the Day 3 article).
Here's the nuance: I didn't earn an extra $900 out of thin air. I freed up time I then spent on other projects. Indirect savings work differently from direct ones. For a freelancer it's more accurate to count "how much I earned thanks to the freed time" rather than "how much I saved."
Extra income: 40 hours × ~$22/hour (average rate on new projects) = $880/month.
Cost: $50/month (tools + API).
Real ROI: 880 / 50 = 1,760%.
#The big picture
| Business | Savings/mo | Cost/mo | ROI (month 2+) |
|---|---|---|---|
| Coffee shop | $250 | $5 | 4,900% |
| Online store | $650 | $180 | 261% |
| Freelancer | $880 | $50 | 1,760% |
Three completely different businesses. ROI > 100% everywhere. Pays off in 1–2 months everywhere.
#When AI does NOT pay off
It happens. Red flags: the task is rare (a handful of times a month), every case is unique and creative, or the cost of a mistake is so high that a human must check everything anyway. If that's you, automation won't earn its keep yet.
#Calculate your own ROI right now
Grab a sheet of paper. Write down: how many hours a week a routine task eats, your hourly rate, and the monthly cost of an agent. If the freed value comes out above ~$200/month — an AI agent will pay for itself, guaranteed.
💼 Want me to calculate the ROI for your case?
Message @iandagent — send me your numbers and I'll tell you if an agent is worth it.