The short version
- The core sum: hours a week on the task, times people, times the share of the task removed, times the loaded cost of an hour.
- Use the loaded hourly cost (wages plus employer NI and pension, spread over hours actually worked), not the headline wage.
- Count every cost: build, running costs, AI usage, staff time and maintenance.
- Keep soft benefits like speed and morale out of the main sum. If it only pays with them added, it probably doesn't pay.
- Payback period is the easiest number to explain to a board. Under six months is strong for a small business.
If you've been pitched AI automation, you've probably been shown a big savings figure and had a nagging feeling it was a bit too neat. You're right to be sceptical. Most of those numbers count every minute saved as cash in the bank and quietly skip half the costs.
Yes, AI automation is often worth it, but only for the right tasks, and you can work out which ones with a simple sum. Multiply the hours a week spent on the task by the number of people doing it, by the share of the task the automation removes, by the loaded cost of an hour. Compare that with the full cost of building and running it. If it pays back within a few months on hard numbers alone, it's worth doing.
This guide shows the sum, two worked examples with every step visible, and the costs people tend to forget. If you'd like to plug in your own figures as you read, our savings calculator does the arithmetic for you.
The formula
Weekly saving = hours a week on the task × number of people × share of the task removed × loaded hourly cost. Everything else is detail.
Each part needs a little care.
- Hours a week on the task. Measure it, don't guess. Ask people to note the time for two normal weeks. Estimates are almost always wrong, sometimes in both directions.
- Number of people. Everyone who does the task, including the manager who checks it.
- Share of the task removed. Automation rarely removes all of a task. Someone still reviews the output, handles exceptions and deals with the odd case that doesn't fit. For a well suited task, 60% to 90% is realistic. Be conservative.
- Loaded hourly cost. The real cost of an hour to the business, covered in the next section.
Multiply the weekly saving by 52 for the yearly figure, and divide that by 12 for a monthly one.
Working out the loaded cost of an hour
The loaded hourly cost is the full yearly cost of employing someone divided by the hours they actually work. It's higher than their hourly wage because of employer National Insurance, pension contributions and paid holiday.
Here's a sensible way to estimate it for one role.
- Start with gross salary. Say £30,000 a year.
- Add employer National Insurance. For the 2026 to 2027 tax year, employers pay 15% on earnings above £5,000 a year, according to HMRC's rates and thresholds for employers. On £30,000 that's 15% of £25,000, which is £3,750.
- Add employer pension. Under auto enrolment, employers must pay at least 3% of qualifying earnings, as set out on GOV.UK. On this salary that's roughly £700. If you pay more, use your real figure.
- Total yearly cost. About £34,450.
- Divide by hours actually worked. A 37.5 hour week, less 5.6 weeks of statutory holiday, is about 46.4 working weeks, or roughly 1,740 hours a year.
£34,450 ÷ 1,740 is about £19.80 an hour. So someone on £30,000 costs you roughly £20 for every hour they actually work. Their headline hourly rate (£30,000 ÷ 52 ÷ 37.5) is about £15.40, so using that would understate the saving by nearly a quarter.
Keep it simple. You don't need payroll precision. If you don't want to work it out per person, multiplying the hourly wage by about 1.25 gets most small businesses into the right area. Rates and thresholds change each April, so check the current year.
We've deliberately left out office space, equipment and management time. Some people add them, which pushes the figure higher. Leaving them out keeps your sums cautious, which is what you want when you're making a case.
Two worked examples
The first example is a team task where an internal tool removes most of the work. The second is a single person and a small automation, which is closer to where most businesses start.
Example 1: a team of six and an internal tool
Six people in an operations team each spend 20 hours a week processing orders: reading emails and PDFs, keying details into the system, checking stock and sending confirmations. A custom internal tool built around Claude reads the orders, fills in the system and drafts confirmations, with staff reviewing exceptions. Testing shows it removes 85% of the work. Their loaded cost is £20 an hour.
- Hours on the task: 6 people × 20 hours = 120 hours a week
- Hours removed: 120 × 85% = 102 hours a week
- Weekly saving: 102 × £20 = £2,040
- Yearly saving: £2,040 × 52 = £106,080, about £106,000
- Monthly saving: £106,080 ÷ 12 = £8,840
Now the costs. The tool costs £7,500 to build, plus £950 a month to run, covering hosting, AI usage and support.
- Net monthly saving: £8,840 minus £950 = £7,890
- Running cost per week: £950 × 12 ÷ 52 = about £219
- Net weekly saving: £2,040 minus £219 = about £1,821
- Payback: £7,500 ÷ £1,821 = about 4.1 weeks
On paper it pays for itself in just over four weeks. Allow a slow first week or two while the team gets used to it and call it roughly five weeks. Over the first year, total cost is £7,500 + (£950 × 12) = £18,900 against a saving of £106,080, a net gain of £87,180.
Example 2: one person and a single automation
An office manager spends 5 hours a week chasing unpaid invoices: checking the accounts package, writing reminder emails and updating a spreadsheet. A single automation pulls overdue invoices each morning, drafts a polite reminder matched to how late each one is, and updates the tracker. She approves the emails in a few clicks. It removes about 70% of the work. Her loaded cost is £25 an hour. The automation costs £1,250 to build, plus about £20 a month in software and AI usage.
- Hours removed: 5 × 70% = 3.5 hours a week
- Weekly saving: 3.5 × £25 = £87.50
- Yearly saving: £87.50 × 52 = £4,550
- Running cost per week: £20 × 12 ÷ 52 = about £4.60
- Net weekly saving: £87.50 minus £4.60 = about £82.90
- Payback: £1,250 ÷ £82.90 = about 15 weeks, or three and a half months
First year: £4,550 saved against £1,250 + £240 = £1,490 of cost, a net gain of £3,060. Smaller, but still a good return, and invoices that get chased on time tend to get paid sooner, which helps cash flow too. You can run your own version of either example in the savings calculator.
Which tasks give the best ROI
The best returns come from tasks that are frequent, repetitive, done by more than one person and easy to check. Multiplying across people is what turns a nice saving into a big one.
- Data entry from emails and documents. Orders, invoices, timesheets and enquiry forms keyed into another system.
- Report building. Weekly or monthly reports pulled together from several spreadsheets or systems.
- Chasing and follow up. Quotes, unpaid invoices, missing paperwork, overdue tasks.
- First drafts of standard documents. Proposals, quotes, onboarding packs and client letters built from templates.
- Inbox sorting. Shared inboxes where a person spends the first hour of the day deciding what goes where. Our guide to AI agents covers this one in detail.
The weaker candidates are tasks done rarely, tasks where every case is different, and tasks where checking the output takes nearly as long as doing it.
The hidden costs to include
A fair ROI sum includes every cost, not just the build price. These are the ones that get missed.
- Setup. The build itself, plus any time mapping the process and gathering examples before work starts.
- Running costs. Software subscriptions, hosting and any automation platform fees.
- AI usage. Most AI services charge by the amount of text processed. It's usually small, but it scales with volume, so estimate it on your real numbers rather than a demo.
- Staff time. Testing, learning the new process and the first few weeks of checking everything closely. Count those hours at the loaded cost too.
- Maintenance. Suppliers change their invoice layouts, you change your price list, a system gets updated. Budget a few hours a month, or a support plan.
Saved time isn't always saved money. Freeing 100 hours a week only becomes cash if those hours go somewhere useful: more sales calls, more jobs handled without hiring, overtime you no longer pay, or a role you don't need to recruit for. Decide where the time goes before you count it as a saving.
Soft benefits, and why to keep them separate
Automation often brings real benefits that are hard to price: faster replies to customers, fewer keying errors, and staff who no longer dread Monday's report. Count them, but keep them out of the main sum.
The reason is simple. Soft benefits are easy to inflate, and once they're in the total nobody can check them. If an automation only looks worthwhile after you add "improved morale" at £10,000 a year, it probably isn't worthwhile. List the soft benefits underneath the hard numbers as a bonus. A decision that stands up on hard savings alone is one you won't regret.
The one exception is errors with a clear cost. If mistyped orders cost you £300 a month in returns and reshipping, that's a hard saving and belongs in the sum.
Payback period vs ROI
Payback period tells you how fast you get your money back. ROI tells you how much you make overall. For most small businesses, payback is the more useful number.
ROI is net gain divided by total cost over a period, usually a year. In example 1, that's £87,180 ÷ £18,900, about 4.6 times the money spent, or 461%. In example 2 it's £3,060 ÷ £1,490, about 205%. Those percentages look impressive, which is partly why they get used in sales pitches.
Payback period is easier to reason about: how many weeks until the automation has covered its build cost. It also protects you against uncertainty. Processes change, software changes, businesses change. An automation that pays back in five weeks is low risk even if your estimates are a bit off. One that pays back in two years is a bet that nothing changes for two years.
A rough rule we use: under six months is strong, six to twelve months is fine if the process is stable, and over a year needs a very good reason.
When not to automate
Don't automate a process that's broken, rare, about to change or mostly judgement. Fix it, leave it, or wait.
- The process is a mess. Automating a muddled process gives you a faster muddle. Tidy it up first, and you may find the tidy version is quick enough by hand.
- It happens a few times a year. Even a cheap automation struggles to pay back on a task that takes ten hours a year.
- It's about to change. If you're switching accounts package next quarter, wait.
- It's mostly judgement or relationship. Difficult client conversations, pricing big bids, sensitive HR matters. AI can help prepare, but the work is the person.
- Mistakes would be costly and hard to spot. If checking the output properly takes as long as doing the task, the saving disappears.
If you're not sure which of your tasks pass these tests, our £750 automation audit looks at how your team actually spends its week, ranks the opportunities by payback and gives you the numbers before you spend anything on building. Or start with the savings calculator and see what your own figures say.
Questions people ask
Is AI automation worth it for a small business?
Often, yes, but only for the right tasks. It pays best on work that's frequent, repetitive and done by several people, such as data entry, report building or chasing quotes. Work out the hours saved and the loaded cost of those hours before you commit, and compare that with the full cost of building and running it.
How do you calculate ROI on automation?
Multiply weekly hours spent on the task by the number of people, the share of the task the automation removes and the loaded hourly cost. That gives the weekly saving. Subtract running costs, then compare the yearly net saving with the total cost to get ROI, or divide the build cost by the weekly net saving to get the payback period in weeks.
What is a good payback period for automation?
For a small business, under six months is strong and under twelve months is usually fine. Longer than that and you're exposed to the process changing before the automation has paid for itself.
What is a loaded hourly cost?
It's the true cost of an hour of someone's time to the business. Add salary, employer National Insurance and employer pension contributions for the year, then divide by the hours the person actually works after holidays. It's usually noticeably higher than their hourly wage.
What are the hidden costs of AI automation?
Setup and build, monthly software subscriptions, AI usage charges, the staff time to test and learn the new process, and ongoing maintenance when systems or prices change. Leave any of these out and the ROI will look better than it really is.