AI automation
Process automation ROI calculator
Put one manual process in and read off the month it pays for itself. The whole cost is in it, the benefit is discounted by how much of the work really goes the new way, and the hours it gives back are shown apart from the cash.
Softblues is a London-based AI practice and Registered Anthropic Partner Network member. The calculator is free and open. Nothing you type leaves your browser, and it will sometimes tell you not to build.
The calculator
Payback, expected case: Month 10. Conservative: month 18. Upside: month 8.
Payback, expected case
Month 10
Cumulative cash benefit covers cumulative cash cost.
- Conservative
- Month 18
- Expected
- Month 10
- Upside
- Month 8
01The process today
How many times this process runs in a month
Include review and exceptions
Include the new review step
Share of volume actually done the new way
Salary plus on costs, per worked hour
02How the hours become money
Somebody has to commit to one
Contribution, never revenue
03What it costs, all in
One time
One time. Usually underestimated
Only seats you would not otherwise buy
Credits, tokens, runtime
Economic, not cash. Kept out of the payback
Payback, expected case
Month 10
Cumulative cash benefit covers cumulative cash cost.
- Conservative
- Month 18
- Expected
- Month 10
- Upside
- Month 8
Cost against benefit, 24 months
Show the chart as a table
| Month | Cost | Conservative | Expected | Upside |
|---|---|---|---|---|
| 1 | £4,500 | £0 | £0 | £0 |
| 2 | £9,000 | £0 | £0 | £0 |
| 3 | £13,500 | £0 | £0 | £0 |
| 4 | £14,231 | £750 | £1,562 | £2,812 |
| 5 | £14,962 | £1,500 | £3,125 | £5,625 |
| 6 | £15,693 | £2,250 | £4,687 | £8,437 |
| 7 | £16,423 | £4,125 | £8,593 | £15,468 |
| 8 | £17,154 | £6,000 | £12,499 | £22,499 |
| 9 | £17,885 | £7,875 | £16,405 | £29,530 |
| 10 | £18,616 | £9,750 | £20,311 | £36,561 |
| 11 | £19,347 | £11,625 | £24,217 | £43,592 |
| 12 | £20,077 | £13,500 | £28,123 | £50,623 |
| 13 | £20,808 | £15,375 | £32,029 | £57,654 |
| 14 | £21,539 | £17,250 | £35,935 | £64,685 |
| 15 | £22,270 | £19,125 | £39,841 | £71,716 |
| 16 | £23,001 | £21,000 | £43,747 | £78,747 |
| 17 | £23,732 | £22,875 | £47,653 | £85,778 |
| 18 | £24,462 | £24,750 | £51,559 | £92,809 |
| 19 | £25,193 | £26,625 | £55,465 | £99,840 |
| 20 | £25,924 | £28,500 | £59,371 | £106,871 |
| 21 | £26,655 | £30,375 | £63,277 | £113,902 |
| 22 | £27,386 | £32,250 | £67,183 | £120,933 |
| 23 | £28,117 | £34,125 | £71,089 | £127,964 |
| 24 | £28,847 | £36,000 | £74,995 | £134,995 |
Capacity releasednot cash
- Minutes saved per run
- 43
- Runs done the new way each month
- 204
- Hours released per month
- 146
- Value at loaded cost, per year
- £45,614excluded from the payback above
The investment
- One time cash
- £13,500
- Recurring cash per year
- £8,770
- Year one cash
- £22,270
- Year two onward
- £8,770
- Year one including internal time
- £25,910economic view
The figures loaded are the manufacturing example from Building the business case for AI adoption. They are illustrative assumptions, not a Softblues result and not any client's result.
A payback month that appears only in the upside column is not a payback. Build the case on the conservative one.
Nothing you type here leaves your browser.
How do you work out the return on automating a process?
Measure the process before anything changes. Take runs per month, hands-on minutes per run including checking and exceptions, the rework rate and the current spend, over four weeks of real work rather than the best run of it. Model the after state on the same sample and include the new review step, because an output somebody has to check is not free. Then discount the benefit by the share of runs that will genuinely go the new way. The hours that come back are capacity, not cash, and they become cash by four routes only: the team handles more volume that already exists to be had, a planned hire is cancelled or deferred in writing, an outsourced service or licence or overtime line is switched off, or a costly event becomes less likely. Price the whole twelve or twenty-four months, so integration and data preparation, consumption, licences, review, training and support are all in it rather than the build alone. Set cumulative cash benefit against cumulative cash cost month by month, on conservative, expected and upside figures. The month they cross is the payback month. If they never cross, that is a finding, and it is far cheaper to have it now.
Why this calculator no longer multiplies hours by an hourly rate
It used to. Runs times minutes times a loaded hourly cost gave a number, and the number was called an annual saving. It is the arithmetic most automation calculators still do, and it is wrong in two places at once. It assumes every run goes the new way, and it treats returned hours as though they had already turned into money.
Both assumptions inflate the same figure, and the second one is the reason a finance director asks where the saving actually shows up and nobody can answer. This version discounts the benefit by adoption, prices the whole cost stack rather than the build, and reports released hours on a line of their own, clearly outside the payback. The number it gives you is smaller. It is also one you can defend in a budget round.
Process automation ROI calculator: at a glance
- What it does
- Returns the payback month on automating one process, from your own baseline, at an adoption rate you choose.
- Inputs
- Runs per month, minutes per run before and after, adoption, loaded hourly cost, the route, the monthly value on three scenarios, and the whole cost stack.
- What it will not do
- Call released hours a saving. Hours are capacity. They are shown on their own and kept out of the payback.
- Payback
- The month cumulative cash benefit covers cumulative cash cost, across a 24 month horizon, on three scenarios.
- Accuracy
- As good as the baseline you put in. Measure the process for four weeks first. We confirm the figure in a paid Process Discovery.
- Cost
- Free. No sign-up, and nothing you type leaves your browser.
- Next step
- Book a Process Discovery to map the process and get a fixed-price build with a payback line.
- Based in
- London-based, working with mid-market firms.
Get the guide and the spreadsheet
The method in full, with two worked examples, the ten rules that keep the arithmetic honest and a one page scorecard to review monthly. The spreadsheet carries the same maths as the calculator on this page, so you can keep a copy with your own numbers in it.
Common questions about automation payback
How do you calculate automation ROI?
Measure the process as it runs today, before anything changes: runs per month, hands-on minutes per run including checking and exceptions, the rework rate and the current spend. Model the after state on the same sample. Discount the benefit by the share of runs that will genuinely go the new way, price the whole twelve or twenty-four months rather than the build alone, and set cumulative benefit against cumulative cost month by month. The month they cross is the payback month.
Why are the hours a process gives back not a saving?
Because nothing has left the bank. Hours returned are capacity, and capacity becomes money by four routes only: the same people handle more volume and that volume exists to be had, a planned hire is cancelled or deferred in writing, an outsourced service or licence or overtime line is switched off, or a costly event becomes less likely. Until somebody commits to one of the four, an automation has produced a productivity story rather than a saving. The calculator shows the released hours and keeps them out of the payback for that reason.
What is automation payback?
The month at which everything the automation has returned in cash covers everything it has cost in cash. It is not the build cost divided by a monthly number, because the benefit does not start on day one. A build that runs for three months with nothing coming back, reaches partial effect for another three and full effect from month seven behaves very differently from one that starts paying immediately, even on the same annual figures.
What belongs in the cost of automating a process?
The build is the part everyone remembers. Underneath it sit integration and data preparation, which is the line most often left at zero and most often wrong, plus consumption for tokens and runtime, any incremental licences, the person who checks the output, training time, and monitoring and support. Internal time is a real economic cost and belongs in the case, marked as economic rather than cash so nobody presents it to the board as money in the bank.
What adoption rate should I assume for one process?
The share of runs that will actually go the new way, measured in a pilot rather than hoped for. A single well-scoped process inside one team commonly runs high, because the team has no other way to do the work, which is exactly why process automation often builds a better case than a firm-wide seat rollout. Anything at or above 95% is flagged, because it is a strong claim and it should be one somebody measured.
Does it include error and rework savings?
Not automatically. Lower rework is usually real and it usually has a route, most often avoid a loss. Work out what the reduction is worth per month and put it into the value figures yourself, on the conservative scenario first. Pair it with a quality measure from the same sample. A faster process that is worse is not a saving.
When is automating a process not worth it?
When the process is broken, undocumented or about to be replaced, fix that first and spend nothing. When the team is already fully booked on work with no queue behind it, the released hours go nowhere unless a hire is cancelled. When the rubric cannot be written down in two weeks, the process is not ready for an agent. Set the route to "No route named yet" and the calculator will say the same.
What happens after I get a figure?
Book a Process Discovery. We map the process end to end, find the steps worth automating, and come back with a fixed-price plan carrying a payback line. Production in 90 days, money back if the proof of concept fails. If the honest answer is that a process change would serve you better than a build, that is what you will get.
Turn the figure into a fixed-price plan
Book a Process Discovery and we will map the process, then quote a fixed-price build with a payback line. Where a process change would serve you better than a build, that is what you will hear.