Skip to content
LucidMetric
All posts
  • roi
  • automation costs
  • planning

What Automation Costs to Run: The Ongoing Bill Most ROI Estimates Leave Out

Most automation ROI math stops at the build cost. Here's how to add the running costs, maintenance, and exceptions so your payback estimate holds up in the real world.

Kamal Farooqi5 min read

The ROI Calculation Most People Do

Here's the usual math. An automation will save 25 hours a month. Your time is worth $30 an hour, so that's $750 a month. The build costs $4,000. Payback in just over five months. Looks great on a slide.

The problem is that this math treats the automation as a one-time purchase, like a piece of equipment that runs forever for free. It isn't. Automations have a monthly running cost, and they rarely remove 100% of the work. Leave both out and your payback estimate is optimistic by months.

This isn't an argument against automating. Most projects still pay back well. It's an argument for doing the math properly so you can pick the right projects and not get surprised.

The Running Costs That Get Left Out

There are five that show up again and again.

1. Platform fees. Tools like Make, Zapier, and n8n Cloud charge by volume, either per task or per operation. A workflow that handles 200 records a month is cheap. The same workflow at 2,000 records, with six steps each, can push you into a higher plan. Growth is good news for your business, but it changes the bill.

2. AI usage fees. If the automation reads documents, classifies emails, or drafts replies, you're paying per request to an AI provider. The amounts are usually small per item and noticeable in aggregate. Long documents and retries cost more than people expect.

3. Upgrades in connected tools. Some software only gives you API or webhook access on a higher plan. Finding that out after the build means a surprise $50 to $100 a month.

4. Maintenance. Apps change their fields, expire their login tokens, and update their APIs. Plan for an hour or two a month on average. Some months it's zero and one month it's a full afternoon.

5. Exception handling. No automation handles every case. The odd invoice format, the customer who fills in the form wrong, the duplicate record. Someone has to review those. That time is real, and it comes out of the hours you thought you were saving.

A Worked Example

Take a composite example: a 12-person services firm automating how incoming supplier invoices are read, coded, and entered into accounting. The build is $4,000.

The naive estimate says 25 hours a month saved. In practice, about one invoice in five has something odd about it and still needs a person, so the realistic saving is 20 hours.

Naive estimateRealistic estimate
Hours saved per month2520
Value of time saved ($30/hr)$750$600
Platform fees$0$30
AI usage fees$0$35
Maintenance and monitoring (about 2 hrs)$0$60
Review of flagged exceptionsincluded above$30
Net monthly benefit$750$445
Payback on $4,0005.3 months9 months

Year one, after paying back the build, still nets about $1,340. In year two, with no build cost, the project returns roughly $5,300. That's a good project. But it's a nine-month payback, not five, and if you'd budgeted for five you'd have spent months thinking it was underperforming.

How to Do This Math Before You Build

A short checklist you can run in 20 minutes:

  • Count the real volume. How many items a month, today and in a year? Multiply by the number of steps to estimate platform usage.
  • Discount the hours. Take your estimated hours saved and knock off 15 to 25% for exceptions. If the process is very standardized, use the low end. If it's messy, use the high end.
  • Ask what else needs upgrading. Check whether every tool in the chain gives you API access on your current plan.
  • Set a maintenance line. A reasonable default is 5 to 10% of the build cost per year, or a small monthly retainer, whichever you prefer. Just make it a line item.
  • Look at 12 months, not just payback. A project that pays back in nine months and then returns $5,000 a year is very different from one that barely breaks even.

When the Running Costs Change the Decision

Most of the time, running costs just stretch the payback a little. Occasionally they flip the decision.

They matter most when the time saved is small and the volume is high. Saving two minutes on each of 10,000 records can look attractive until per-task pricing eats the savings. In that case a custom script or a self-hosted tool may be cheaper to run, even though it costs more to build. That's the same trade-off as low-code versus custom, just viewed through running costs.

They also matter when the process is still changing. If you rewrite the workflow every quarter, maintenance will be a constant cost. Fix the process first, then automate it.

On the other side, don't forget what doing nothing costs. Manual work has its own hidden running costs: errors, delays, and the time of your best people spent on low-value tasks. A fair comparison puts both columns side by side.

Keep the Bill Visible After Launch

Once the automation is live, check three numbers monthly for the first quarter: how many runs it handled, what the platform and AI bills were, and how many items landed in the exception queue. If the bill grows faster than the volume, something is looping or retrying. If the exception queue grows, the inputs have changed and the logic needs a tune-up. Ten minutes a month catches both early.

The Bottom Line

Automation is still one of the better returns available to a small business. But the honest number is net of running costs and exceptions, and it's always a little less exciting than the napkin math. That's fine. A realistic nine-month payback you can trust beats a five-month promise that quietly slips.

If you have a process in mind and want a realistic estimate, including running costs, not just the build quote, book a free call. We'll go through the numbers together and tell you plainly whether it's worth doing.

Want this done in your business?

Tell us the task that eats the most hours. We'll show you exactly how to automate it.