An automated purchase order system removes the clerical parts of ordering and leaves the commercial ones alone, which is the correct division. Being explicit about which is which prevents both disappointment and a more dangerous error, which is configuring the system to make decisions it has no context for and then trusting the result.
What runs by rule
Numbering, routing for approval by value and cost centre, issuing to the supplier, recording that it was sent, reminding an approver who has not responded, receipting against lines and flagging orders that are overdue. All of these are deterministic, high volume and repetitive, which is exactly the profile automation handles well.
What stays human
Choosing what to buy and from whom, agreeing a price, deciding whether a short delivery should be chased or accepted, and judging whether a substitution is acceptable. These are commercial judgements with context no rule holds. The system's job is to present the information and record the outcome, which is genuinely useful and not the same as deciding.
The narrow middle: reorder rules
Automatic orders against agreed prices for predictable repeat purchasing. This works well and depends entirely on the price agreement being current. An automatic order against a stale price list commits you at the wrong price, faster than a person would have. Automation multiplies existing discipline in both directions.
Questions people ask about automated purchase order system
How much of our ordering could be automatic?
Look at how much is repeat buying against agreed prices. For most organisations it is a minority, which is worth knowing before buying on that promise.
Does automation reduce errors?
It removes transcription and numbering errors. It does not catch ordering the wrong thing, which stays a human check at the request stage.
What should we automate first?
Numbering and approval routing. Both are quick, both remove failures that make everything downstream impossible.