AI ROI: How to Tell If a Tool Is Actually Paying Off
Here’s a conversation I have with business owners every month: they’ve been paying $200-500 in AI subscriptions for six months and have a vague feeling they’re “using AI now.” When I ask what it’s actually saving them, they pause. They can’t tell me. That pause is the problem.
Measuring AI ROI doesn’t require a spreadsheet PhD. It requires one honest question per tool: what would this have cost us to do the old way? If you can’t answer that, you don’t know whether the tool is earning its keep or quietly draining $2,400 a year from your operating budget.
The three numbers that actually matter
Every AI tool ROI calculation needs three inputs:
- Hours freed per week — ask the person using it to track this for two weeks. “Eyeballing it” consistently underestimates; a time log is worth five minutes of effort.
- Loaded hourly cost — salary plus super, software, and overhead. For most SMBs this lands between $35-60/hour for operations staff, more for senior roles.
- Monthly tool cost — what you’re actually paying, not the headline plan price.
Monthly value = hours freed per week × 4.3 × loaded hourly cost. If that number is less than 3× the tool’s monthly cost, you have a marginal tool that may not be worth the workflow disruption.
How to run the measurement
Don’t wait six months to evaluate. Set a 30-day trial for every new tool with a written hypothesis: “This tool will save X hours per week on Y task.” At the end of the month, check. Either it did or it didn’t.
The measurement itself is simple. For a content or admin tool, count the hours the task used to take versus how long it takes now. For a customer-support tool, pull your average response time and ticket volume from the dashboard — every decent support platform has these numbers. For a bookkeeping tool like Dext or AutoEntry, your bookkeeper can tell you in ten minutes how many hours of data entry disappeared.
Here’s a realistic comparison across common use cases for a 15-person business running each tool for 30 days:
These aren’t guarantees — they’re the midpoint of what we see when a tool is actually being used on the right tasks. If your numbers are well below this range, either the task volume doesn’t justify it or the tool isn’t set up correctly.
A real example: the proposal bottleneck
A 14-person marketing agency came to us paying $120/month for a writing assistant they called “underwhelming.” When we dug in, they’d been using it to write blog intros. That’s not where their time was going.
Their actual bottleneck: client proposals. Four hours each, eight a month, written by a senior account manager at a loaded cost of roughly $65/hour. That’s $2,080/month in proposal-writing time. We rebuilt their workflow around the same tool, fed it their proposal templates, past wins, and service descriptions, and cut the per-proposal time from four hours to just under one.
Net result: 24 hours per month freed, $1,560/month in recovered capacity. Against a $120/month tool, that’s a 13:1 return. The tool didn’t change — the task they aimed it at did.
Most underperforming AI tools aren't bad tools. They're aimed at the wrong task. Find the one that eats the most senior hours and redirect the tool there.
Red flags: when to cut a tool
Some tools genuinely aren’t worth keeping. These are the signs to look for after a proper 30-day trial:
- Nobody's using it — if your team routes around it out of habit, that habit won't change without structured onboarding. The savings are theoretical.
- Output requires more fixing than creating from scratch — some tools fit your workflow; others generate 80% garbage and 20% usable. The 80% still costs someone time to review.
- The ROI math is less than 2× — a tool saving $200/month that costs $150/month isn't an investment. It's administrative overhead with a slight upside.
- You're paying for features you never touch — tools that start at a basic tier and grew don't always fit a small team's needs. Audit what you actually use quarterly.
A before/after framework you can use today
Take any single AI tool you’re currently paying for and run this in 20 minutes:
| Measurement | Before AI | With AI |
|---|---|---|
| Hours spent on the task | X hours/week | Y hours/week |
| Monthly time cost | X × 4.3 × hourly rate | Y × 4.3 × hourly rate |
| Monthly tool cost | — | Subscription fee |
| Net monthly saving | — | Time cost difference minus tool cost |
| Payback verdict | — | Green if saving ≥ 3× tool cost |
If you can’t fill in the “before AI” column because you never measured it, that’s the first thing to fix. Baseline before you buy.
The honest takeaway
Most SMBs I speak to are not under-investing in AI tools. A surprising number are over-subscribed and under-measuring. Two well-aimed tools you actually track will outperform six subscriptions running on autopilot.
Pick your highest-volume, most repetitive task. Measure how long it takes now. Trial one tool for 30 days with a specific time target. Check the math at the end of the month. That loop — pick, measure, check — is the entire playbook.
If you want help identifying which tools in your stack are earning their keep and which to cut, book a free AI diagnostic at Metatree Lab. We’ll look at your actual workflows, not a generic checklist, and tell you where the real savings are.