AI automation is easy to start and easy to overspend on. Before you build anything, from a simple workflow in Zapier to a custom AI agent, it's worth spending 30 minutes estimating whether it will actually pay for itself.

Here's a simple method that works for most small businesses.

The formula

Monthly benefit − monthly running cost = monthly gain Upfront cost ÷ monthly gain = payback period in months

The hard part is estimating each number honestly.

Step 1: Measure the current process

Before estimating savings, measure what happens now. For one or two weeks, record:

QuestionExample
How often does the task happen?60 enquiries a month
How long does each take?12 minutes
Who does it, and what does their time cost?Office manager, £18/hour fully loaded
How often does it go wrong?5% of enquiries get no reply within a day
What does a mistake cost?A lost customer worth £400 profit

"Fully loaded" cost means salary plus employer taxes, pension and overheads, often 20–30% above the hourly wage.

Step 2: Estimate the benefits

Time saved

60 tasks × 12 minutes = 12 hours a month. If automation removes 75% of that, you save 9 hours. At £18/hour, that's £162 a month.

But be honest: is that time used for something valuable? If the office manager uses it for sales follow-ups that bring in revenue, count it. If it simply disappears, count it at a lower value.

Errors avoided

If 5% of 60 enquiries (3 a month) currently get missed, and faster automated replies win back one of those a month at £400 profit, that's £400 a month. This is often worth more than the time saved.

Revenue gained

Faster response, 24/7 availability and consistent follow-up can increase conversion. Use conservative estimates: half of what you hope.

Total monthly benefit in this example

£162 (time) + £400 (one extra customer) = £562 a month.

Step 3: Estimate the costs

CostExample
Upfront setup (your time or a developer)£1,500
Software subscriptions£40/month
AI usage fees£20/month
Monitoring and fixes (1 hour/month)£18/month
Training staffIncluded in setup
Monthly running cost£78

AI usage fees depend on volume and the model you use. Estimate on your busiest month, not your quietest.

Step 4: Work out payback

Monthly gain = £562 − £78 = £484 Payback = £1,500 ÷ £484 = about 3 months

That's a strong case. A payback over 12 months deserves much more scrutiny.

Hidden costs to include

  • Mistakes the automation makes. AI can be confidently wrong. Budget for a review step and the occasional fix. See AI hallucinations.
  • Exceptions. Most processes have edge cases the automation can't handle. Someone still deals with those.
  • Changes in your tools. When an app changes its API or a form changes, automations break. Allow maintenance time.
  • Data protection work. Checking providers, updating your privacy policy. Read using AI tools without breaking GDPR.
  • Lock-in. Moving off a platform later can mean rebuilding.

Quick sense checks

SignWhat it suggests
The task happens rarelyAutomation probably isn't worth it
The process isn't written downDocument and simplify it first
Every case is differentHard to automate reliably
Mistakes would be expensive or publicKeep a human approval step
An existing tool already does itUse that before building custom

Where the best returns usually are

From what we see with clients, the highest-return automations are usually:

  1. Lead response and follow-up, because speed wins customers.
  2. Invoice chasing, because cash arrives sooner.
  3. Data entry between systems (forms to CRM to accounts).
  4. Meeting notes and follow-ups.
  5. First-line customer support for repetitive questions.

See 12 AI agent use cases small businesses actually run and workflow automations that save 10 hours a week for more examples.

Build, buy or use what you have?

OptionUpfrontRunningBest when
Feature in software you already pay forLowLowIt covers 80% of what you need
No-code automation (Zapier, Make, n8n)Low–mediumMediumConnecting several tools
Custom AI agentHigherVariesHigh volume, specific process, clear gains

Our comparison of Zapier vs Make vs n8n helps with the middle option.

After launch: measure the real result

Run the numbers again after two months, using real data. Did time actually fall? Did response speed and conversion improve? Did errors drop? Keep what works, fix or switch off what doesn't.

The short version

Measure the current process first, estimate time saved (only where it's used well), errors avoided and revenue gained conservatively, then subtract setup and running costs, including the hidden ones. A payback under six months is strong; over twelve needs a hard look. Recheck with real numbers after launch.

Want help working out which automations are worth it for your business? Our AI automation team starts every project with exactly this calculation, or try the free AI Project Blueprint for scope and cost.