Automating purchase orders is usually attempted as one project and works better as four steps, each of which delivers something on its own. Done in this order, each step makes the next easier and none of them depends on the whole thing being finished. Done in the reverse order, which is common, the early wins never arrive.
Step one and two: numbering and approval routing
Numbers issued automatically from one source, never reused. Then approval routed by value and cost centre to a named person, before the order is sent, with cover during absence. These two are quick, they remove the failures that make everything downstream impossible, and they deliver benefit within days rather than months.
Step three: receipting where deliveries happen
Whoever takes the delivery records it against the order line, at the time, on whatever device is available. This is the step that unlocks matching, and it is the one most often left until last because it involves people outside finance. Doing it third rather than last is what makes the whole exercise pay.
Step four: reorder rules, where they apply
Automatic orders against agreed prices for predictable, repeat purchasing. This is the only genuinely automatic ordering, it depends on the price list being current, and it applies to a narrower slice of most organisations' spend than the phrase suggests. Doing it last means it runs on data that is already reliable.
Questions people ask about how to automate purchase orders
What is the most commonly skipped step?
Receipting, because it involves people outside finance. It is also the step everything downstream depends on.
Can we automate ordering entirely?
For repeat buying against agreed prices, yes. For anything requiring a commercial judgement, the paperwork automates and the decision does not.
How long should this take?
The first two steps in weeks, receipting over a quarter because it is a habit change, and reorder rules whenever the data justifies it.