Most small businesses have somewhere between five and fifty candidate processes for automation. Only a handful are worth doing. The difficulty is that from the inside they all look similar — annoying, repetitive, and obviously fixable.
Here are six questions that separate them. Run any process through these before spending anything.
1. How many hours a month does it genuinely consume?
Not “it feels like ages”. An actual number.
Owners are consistently wrong about this in both directions. Tasks that feel enormous are often forty minutes done twice a week. Tasks nobody mentions turn out to be eating a day and a half a month because they are spread thin across four people.
If you cannot answer this within a reasonable margin, that is your first task, and it costs nothing. Have whoever does the work log it for two weeks. The results reorder people’s priorities more often than not.
Below roughly two hours a month, stop here. Almost nothing at that volume justifies the setup cost, the subscription and the maintenance.
2. Are the rules writable down?
Sit with the person who does the task and ask them to explain every step, including what they do when something is unusual.
If it can be written as a sequence of decisions someone else could follow — if this, then that; if the amount is over X, do Y — it is automatable. If the answer involves “you just get a feel for it” or “it depends on the client”, you have found judgement, and judgement does not automate cleanly.
Many processes are a mix: eighty percent mechanical with a judgement call in the middle. Those are excellent candidates, provided you automate the eighty percent and route the rest to a person rather than trying to make software guess.
3. What happens when it goes wrong?
Ask specifically: if this produces a wrong result and nobody notices for a month, what is the damage?
Drafting internal email: negligible. Filing documents in the wrong folder: annoying, recoverable. Miscoding transactions in the ledger: expensive, and possibly a restatement. Sending wrong information to a client: reputational.
This does not disqualify anything. It determines how much review and testing the project needs, which determines the cost. High-consequence processes can absolutely be automated — they just need gates, and the gates are frequently more expensive than the automation.
4. Is the process stable?
Has it changed materially in the last six months? Is it about to, because you are switching software, changing suppliers, or restructuring the team?
Automating a process that is in flux means building something twice. If you know a change is coming, wait for it. This is one of the most common reasons we tell people to do nothing yet, and it is nearly always the right call.
5. Does it depend on one person’s head?
If the answer is yes, that raises the value of automating it beyond the hours saved.
A process only one person can perform is a genuine business risk — the risk shows up the week they resign, or go on holiday during month end. Automating it forces the process to be documented, which has value even if the automation itself never runs.
Conversely, if it is a process everyone can do and everyone shares, the key-person argument does not apply and the decision comes back to pure arithmetic.
6. Would fixing the process be cheaper than automating it?
The question people skip.
A surprising number of painful processes exist because of a decision nobody has revisited. A report produced monthly that nobody reads. A double-entry step introduced when two systems could not talk, three years before the integration existed. An approval requirement added after one incident in 2022.
Before automating anything, ask why the process exists at all and whether it still needs to. Deleting a step is free, instant, and has no maintenance cost. It is by a wide margin the best possible outcome, and it is the one an automation vendor has no incentive to suggest.
Putting it together
A process worth automating is usually: more than a few hours a month, rule-based enough to write down, stable, and either low-consequence or high-consequence-with-a-clear-gate. Bonus points if only one person can do it.
A process not worth automating is usually: low volume, judgement-heavy, about to change, or — best of all — one that should simply be deleted.
If you run your top five candidates through these six questions honestly, you will typically find one clear winner, two maybes and two you should stop thinking about. That is a genuinely useful afternoon’s work and it does not require hiring anyone.
If you would rather work through it with someone who does this regularly, that is exactly what the free 15-minute call is for. If you want your time drains ranked and matched to specific tools, the $999 AI Tools Assessment does that.