Before you spend a single pound on AI automation, you need to answer one question: will this actually pay for itself? Too many businesses jump into automation projects based on enthusiasm or competitive pressure, then struggle to justify the cost six months later. The good news is that calculating ROI on an AI automation project isn't complicated — it just requires you to be honest about three things: what the automation costs, what it replaces, and how quickly you'll see the difference. Get those numbers right before you commit, and you'll either confirm you've found a genuine winner or save yourself from an expensive mistake.
Start With the Cost of Doing Nothing
The first number you need isn't the price of the software — it's the cost of the problem you're trying to solve. This is where most people skip a step, and it leads to vague justifications like "it'll save us time" without ever quantifying how much that time is worth.
Start by identifying the exact task you want to automate. Then calculate what it currently costs you in three ways:
Labour time. How many hours per week does your team spend on this task? Multiply that by their hourly cost (salary plus employer NI, pension, and overhead — typically 1.3–1.5x the base salary). A marketing coordinator spending four hours a week manually updating a CRM after client calls costs you roughly £80–£120 in labour per week, or £4,000–£6,000 per year, before you account for what else they could be doing instead.
Error costs. Manual processes create mistakes. A missed follow-up email might cost a law firm a £5,000 client. A data entry error in an invoicing workflow can delay payment by weeks. These are real costs, even if they don't show up on a single line of your accounts.
Opportunity cost. What would your team do with that reclaimed time? If a fee earner spends three hours a week on administrative hand-offs between your project management tool and your billing system, that's three hours not spent on billable work. At £150 an hour, that's £450 a week — £23,000 a year — in lost billing capacity.
Add these three figures together, and you have your baseline: the annual cost of doing nothing.
Build a Realistic Cost Model for the Automation
Now you can look at what the automation actually costs. A common mistake is only counting the software licence. A proper cost model includes four components:
Setup and integration costs. Whether you're working with an agency or using a no-code platform, there's an upfront cost to build and configure the automation. For a straightforward single-workflow project — say, connecting your intake form to your CRM and triggering a follow-up email sequence — expect to pay anywhere from £500 to £3,000 depending on complexity.
Software and platform fees. Most AI automation tools run on monthly subscriptions. Depending on the tools involved (Zapier, Make, a dedicated AI agent layer), budget between £50 and £300 per month for a typical SME workflow.
Maintenance and iteration. Automations aren't entirely set-and-forget. Expect to spend a few hours per quarter reviewing performance and making adjustments, or factor in a small retainer with your agency. Budget roughly £500–£1,000 per year for maintenance.
Change management. There's always a short settling-in period where your team adapts. This is hard to quantify, but building in two to four weeks of reduced efficiency during rollout is realistic.
A typical single-workflow AI automation project might therefore cost £2,000–£4,000 to set up and £1,500–£3,600 per year to run. Total first-year cost: roughly £3,500–£7,600.
Do the ROI Calculation (With a Real Example)
Once you have both numbers, the calculation is straightforward:
ROI = (Annual benefit − Annual cost) ÷ Annual cost × 100
Let's make this concrete. Take a 12-person consultancy that manually moves project updates between Slack, their project management tool (Asana), and their client-facing status reports every week. Two project managers each spend about three hours a week on this — copying notes, reformatting updates, sending emails. That's six hours a week, at an average cost of £45 per hour all-in.
- Annual labour cost of the task: 6 hours × £45 × 52 weeks = £14,040
- Error and delay costs (missed updates, client chasing): estimated £3,000/year
- Total annual cost of doing nothing: £17,040
They implement an AI agent that monitors Slack for project update messages, extracts key information, pushes it to Asana, and generates a formatted weekly client report automatically.
- Setup cost: £2,500
- Annual running cost: £1,800 (tools + light maintenance)
- Total first-year cost: £4,300
First-year ROI: (£17,040 − £4,300) ÷ £4,300 × 100 = 297%
Payback period: approximately 3 months.
From year two, the running cost drops to £1,800 against an annual saving of £17,040 — nearly a 10:1 return. That's not a marginal efficiency gain. That's a fundamental shift in how much capacity those two project managers have to do higher-value work.
The Three Factors That Skew Your Numbers (and How to Adjust)
Not every automation project will yield a 300% ROI. Three factors commonly distort projections — in both directions.
Volume matters more than you think. Automations deliver the strongest returns on high-frequency, repetitive tasks. If the process you're automating only happens ten times a month, the maths often won't stack up unless the per-instance cost or error risk is very high. Before committing, count how many times the task actually happens. If it's fewer than 50 times a month, scrutinise your numbers carefully.
Partial automation is still valuable. Not every workflow can be fully automated. An AI agent might handle 80% of a task and still require a human to review or approve. That's fine — factor in the remaining human time honestly, and you'll still likely see a strong return. A dental clinic that automates appointment reminders and rescheduling requests might still need a receptionist to handle complex queries. But if automation handles 70% of inbound messages, the labour saving is still significant.
Adoption determines whether savings materialise. An automation your team ignores or routes around delivers zero ROI. Before calculating savings, ask honestly: will the people affected actually use this? If there's likely resistance, factor in time for training and a longer transition period before you see full benefit.
Conclusion
The ROI framework here isn't complicated, but it does require discipline. Identify the true cost of the status quo — labour, errors, and opportunity cost combined. Build an honest cost model for the automation, including setup, tools, and maintenance. Run the numbers before you commit, not after. If the payback period is under 12 months and the second-year return is strong, you have a clear case to proceed. If the numbers are marginal, either the problem isn't the right one to automate first, or the solution is more complex than it needs to be. Either way, you'll have made the decision with your eyes open — which is exactly where you want to be.