The Automation You Still Check Every Time
In about half the businesses we mapped, an automation got enough wrong that someone checked every result. Where that happens, and what the check costs.
28 September 2026
Software that fills in fields, files emails or codes transactions is bought to take typing off someone’s desk. In about half the businesses we mapped over the past nine months, an automation like this got enough wrong that a person checked every result it produced. The typing went away. The time mostly stayed.
We see this in the sessions where we map a role’s work with an owner before a new hire starts. The owner walks through each task as it runs today, and the automations show up with the check attached.
The automations that still get checked every time
The pattern turned up across very different tools:
- Job details from a work order. Job management software creates a new job from an incoming work order and fills in the contact details. It often takes them from the wrong part of the document, so every field is read against the work order before anyone uses the job.
- Transaction coding. Accounting software learns bank rules and codes transactions on its own. It misclassifies enough of them that each monthly reconciliation still needs a judgement pass, plus a written guide to the known exceptions.
- Publishing settings. A video editing tool publishes with the wrong default settings unless someone overrides them on every upload.
- Filing submitted forms. An automation files forms into the CRM and the shared drive. When the name on the form differs from the name on the record, a nickname against a legal name for example, the filing fails and the submission is moved by hand.
- Adding contacts from the inbox. An “add to CRM” button can attach a contact to a system or alias address instead of the customer’s own, so the address is checked every time.
Two more tools added a manual step to every entry. Timesheet software needed end times corrected by hand, and its entries could not be deleted once they had been paid. Accounting software lost entered data unless each tab was saved separately.
Why every result got checked
These automations got enough wrong that the person using each output looked at every result before relying on it.
Our reading of why the check had to cover everything: a wrong output here looks like a right one. A contact detail pulled from the wrong part of a document is still a valid name and phone number. Misclassified transactions still sit in real accounts. And a post published with the wrong settings is still published. When a mistake looks like a correct result, the only way to find it is to look at all of them.
Where a tool reliably marks the results it is unsure of, a person can work through just those. In every example above, finding the mistakes was left to a person.
What the checking costs
Checking an automated result usually means opening the source and reading it against the output. For a job created from a work order, that is the work order and the job, side by side, field by field. For a coded transaction, it is the bank line and the account it landed in.
That is close to the work the automation replaced. The keystrokes are gone. The reading and the judgement remain, and someone still has to do them.
Working out the check before the software goes in
The check can be worked out before the tool is running. Three things settle it:
- Who reads each output before it is used. If the answer is “whoever notices”, the check has no owner.
- How long one check takes, and how many results arrive in a week. Multiplied together, that is the time the automation keeps.
- How a wrong result would be spotted. If a wrong output looks the same as a right one, all of them get checked.
An automation can still be worth having with a check attached. The transaction coding in these sessions ran that way: the software coded everything, and each month a person made a judgement pass using a written guide to the known exceptions. That is a different job from the one the software was bought to remove, and it needs a person and time set aside for it.
Giving the check an owner
When we map a role before a new hire starts, the checks sit in the task list alongside everything else: which outputs get read, against what, and how often. That puts the work the automation still needs in front of the owner, with a named person responsible for it.
You can see how that setup works. If you are about to put in software like this, or already check one every day, book a Role Plan call and we can map where the checking sits.
Ready to build
your back office?
Book a call with Dusty. We work out the first role to hire, and afterwards you get your Role Plan in writing.
Related articles

Clean-Up Comes Before the Daily Routine
In about one in four roles we mapped this year, the first task was clearing a backlog before any daily routine could start. What that backlog looked like.

A Five-Minute Lead Target and the Intake Behind It
A trades business set a five-minute call-back target while leads arrived only by email, with no alert. Why intake comes before the response target.

The Signal That Lets an Owner Step Back
An owner stepped back from a remote role and had no regular view of what was done. The setup behind it, and the signal agreed for the new role.