Key takeaways
- Answer four questions before opening any calculator: how often the process runs, whether you can describe it, what a wrong run costs, and who owns it. A calculator only prices a process that has already passed those.
- Treat the four as gates in order, not a score out of ten. The first one you fail tells you what to do next, and it is usually to write the procedure down rather than to buy anything.
- Failing the blast-radius question does not disqualify a process. It identifies the one step that keeps a permanent human approval while the rest of the run stops needing you.
- A tool that hides its formula, or asks for an email before showing a result, is a lead magnet. Every tool here runs in your browser, prints its arithmetic and needs no account.
Four questions that decide whether a process is a candidate
Most badly automated processes failed one of these before anyone started, and nobody asked. Take them in order and stop at the first one you cannot answer confidently. That failure is the useful output: it tells you what to do next, and it is rarely to buy something.
| Question | A pass sounds like | A fail sounds like | What a fail means you should do |
|---|---|---|---|
| Frequency. How often does it run, and does the volume grow as the business does? | Weekly at least, and every new customer or order adds to it. | A few times a quarter, and the volume is flat. | Leave it manual and write a checklist. Google's site reliability engineers reserve automation for toil: repetitive work that produces no enduring value and grows with the business. Flat work is not it. |
| Describability. Could a new joiner run it from your written instructions, exceptions included? | You can name the trigger, the inputs, the two or three branches and the condition that means it is finished. | “It depends”, or you would have to sit next to them for a week. | Write the procedure first. That is the actual work, and no tool shortens it. Start from a template rather than a blank page. |
| Blast radius. What is the worst a wrong run can do before a person sees it? | A wrong field, noticed the same day, fixed by editing one record. | Money moves, a customer is told something you cannot retract, or a decision lands on a named individual. | Automate it anyway, with that one step permanently gated behind a human approval. Everything either side of the gate still stops needing you. |
| Ownership. Who will notice when it quietly stops being right? | One named person can say what a correct run looks like, and sees the output every week. | Split between three people, or the output lands in a folder nobody opens. | Name the owner first. NIST's AI Risk Management Framework puts this under GOVERN: policies defining who is responsible for what when people and software share a job. |
The order is deliberate. Frequency and describability decide whether there is anything worth automating; blast radius and ownership decide how it has to be supervised once there is. Only then is the remaining question a money question, which is what a calculator is for.
Be honest about the standard you are holding it to as well. Google's SRE book is blunt that 100% is the wrong reliability target for basically everything. The real question is not whether the automation will ever be wrong, but what rate of wrong runs you can catch and absorb.
The four questions applied to one process
A two-person ops team sets up every new customer by hand: create the billing account, open the shared workspace, send the welcome email with credentials. It happens around 25 times a month and climbs with every good sales quarter. Frequency: pass.
They can write it down: the trigger is a closed-won deal, the inputs are company name, plan and billing contact, and there are two branches (annual invoicing or card) plus one exception (a missing tax number puts the account on hold). Describability: pass.
A wrong workspace name is a five-minute fix, but a welcome email sent to the wrong address cannot be retracted, so that step keeps a permanent approval while the two system updates do not. Blast radius: pass, with one gate. One of the two owns the process and checks the queue each morning. Ownership: pass.
Only now is the calculator the right tool, and the minutes spent approving that email belong in the estimate. These figures are illustrative, not measured results.
What every tool on this page has to do
- Run in your browser. What you type stays on your machine. Nothing is sent to us to produce the result.
- Show its formula. Every line of arithmetic is printed on the page, so you can rebuild it in a spreadsheet and disagree with it.
- Work without an account. The result appears before any invitation to talk to us, not after.
- Label assumptions, and say where it stops. Defaults are starting points chosen to be argued with, never industry benchmarks dressed up as facts, and every tool carries its own limitations.
Those rules are not specific to the tools. Our methodology sets out how everything here is researched, sourced and corrected, and the tools are held to the same standard as the guides.
Where to start
The automation ROI calculator answers the fourth question with numbers rather than instinct: it works out the monthly cost at which automating a given process stops being worth doing, from your own volumes and handling times.
Every formula behind it is printed on the page, so you can check the arithmetic, disagree with a default, or rebuild the whole thing in a spreadsheet without asking us for anything.
In this section
Limitations and when not to use this
- This test sorts candidates; it does not design the automation. Deciding which steps become plain code, which need model judgment and which keep a person accountable is the next piece of work, and the guides cover it.
- The tools here estimate. None of them reads your systems or measures your current process, so every output is only as good as the figures you type in and the assumptions you accept.
- If a process runs monthly or less, or changes shape every few weeks, the honest answer is no tool at all. A written checklist costs an afternoon and will still be correct next quarter.
- Nothing here is financial, legal or employment advice, and no output is a quote or a prediction of what any product will achieve for your process.
Sources
- Eliminating Toil — Google, The Site Reliability Workbook Accessed 24 July 2026
- Embracing Risk — Google, Site Reliability Engineering Accessed 24 July 2026
- AI Risk Management Framework — NIST Accessed 24 July 2026
Price your first candidate process
Once a process clears all four questions, the ROI calculator works out the monthly cost at which automating it stops being worth doing.
Price your first candidate processUli Prantz
Builds and operates all-agents
Uli Prantz builds all-agents, the process-automation platform this site documents. He writes about the operational side of automating recurring business work: where deterministic code beats model judgment, where it does not, and where a human still has to approve.