Automate the approval, not the judgment

Business process automation uses software to carry out repetitive operational tasks with minimal manual intervention. In practice that means moving data between systems, routing requests to the right person, sending notifications and reminders, and running scheduled jobs such as reports and syncs.

A good candidate is a step with clear rules and predictable inputs. If you can write the rule down in a sentence, software can follow it.

• Routing: a purchase request over a limit goes to a manager; under it, straight to purchasing.
• Reminders: an approval waiting more than a day nudges the approver, then their backup.
• Data movement: a confirmed order creates the invoice and updates stock without anyone retyping it.
• Notifications: the customer is told when their order ships; the team is told when a job is blocked.
• Scheduled work: the weekly report builds itself and lands in the right inbox.

Exceptions, edge cases and anything that needs judgment about a customer, a risk or a relationship. Should we extend credit to this customer? Is this defect acceptable? Is this request reasonable? Automation should make these decisions easier to reach, by bringing the right information to the right person at the right time, not make them.

This is also why automation rarely removes people from a process. It removes the chasing, copying and reminding, so the people who make decisions have time to make them well.

Follow one task from start to finish and write down every manual step. Then rank the steps by two questions: how often does it happen, and how much does it cost when it goes wrong? Start with the step that scores high on both and has clear rules.

Some of the most useful automation isn't a separate tool at all; it is a rule built into the system people already work in. In Budve SmartFlow, for example, a production stage stays blocked until the previous stage is complete. Nobody has to remember to check, and nobody can skip ahead by accident. The judgment, whether a part passes inspection, stays with quality; the rule that stages happen in order is enforced by the software.

Here is how the rule of thumb applies to a common process, step by step.

• Someone submits a purchase request through a form, with the amount, supplier and reason as required fields. Automated: the form refuses incomplete requests.
• The request is routed by amount: small purchases go straight to purchasing, larger ones to a manager. Automated: routing by a written rule.
• The approver reviews the reason and the supplier. Human: this is the judgment.
• If nobody acts within a set time, the approver gets a reminder, then a named backup. Automated.
• Once approved, the purchase order is created in the accounting system from the request data. Automated: no retyping.
• If the order fails to sync, someone is alerted with the details. Automated, with a person to fix it.

• Automating a broken process. If the manual process is unclear, automation makes the confusion faster. Fix the steps first.
• Hiding the rules. If nobody can see why a request went where it did, people stop trusting the system. Log the rule that fired.
• No owner. Every automation needs a named person who is alerted when it fails and who can change it.
• Automating the exception. Rare cases are usually cheaper to handle by hand than to automate.

Automation that fails silently is worse than none, because people stop checking. Every automation should be observable and recoverable.

• Logs of what ran, when, and with what data.
• Retries for temporary failures, such as a service being briefly unavailable.
• Alerts to a named person when something fails for real.
• A clear way for a person to step in and finish the task by hand.

Our automation service starts with exactly this workflow review. For small teams such as local businesses, the first automations are usually order confirmations and follow-ups; for clinics, appointment reminders.