Automation decisions
When not to automate a business process
Six conditions that make automation premature, unsafe or needlessly expensive, and what to change before revisiting the build.
Marc PaulOperational designer and automation builderThe work does not repeat yet
A new offer changes after every client. The owner is still learning which information matters and which promise is realistic. Encoding that process can slow the learning and make each correction a technical change.
Use a light checklist and collect examples. Revisit automation when a stable core appears across several cases.
Nobody owns the result
Automation cannot solve a disagreement about who approves, who receives the work or what complete means. It can hide the disagreement until a case fails.
Name the business owner first. That person does not have to perform every step. They must be able to settle the rule and accept the result.
The inputs are unreliable and there is no source
If every case begins with a different spreadsheet, an old email thread and a value copied from memory, the first job is information ownership. A workflow cannot choose the true value safely without a rule.
Set one approved source for each important fact. Decide how missing or conflicting values are handled. Then assess the build.
The process is mostly consequential judgement
Some work is valuable because it changes with context. A sensitive client negotiation, original diagnosis or high-risk approval may need a responsible person throughout.
There may still be administrative work around it. Prepare the information, schedule the review and record the decision. Do not treat the central judgement as a nuisance to remove.
A mistake would be invisible or hard to reverse
An early version needs a safe test environment, observable completion and a manual recovery route. If a wrong action could send money, expose confidential data or change a customer commitment without detection, more control is needed before automation.
Bring in legal, security, financial or specialist technical review where the consequence requires it. Process automation is not a substitute for those disciplines.
The cost rests on a perfect saving
A business case that only works if every current hour disappears is too optimistic. People still handle exceptions, learn the new route and make the decisions the system cannot own.
Use a conservative case. Include provider costs and ongoing care. If the investment still makes sense with partial adoption and a realistic ramp, it has room to survive normal work.
Waiting should produce evidence
‘Not yet’ is useful when it comes with a next observation. Collect ten recent cases. Track the missing information for a month. Name which exceptions genuinely needed the owner.
When you return to the decision, the process will be easier to bound and the quote will carry less uncertainty.