Procurement Automation: What Should You Automate First?

Ask ten procurement teams what they automated first and most will say something different. Ask what it was worth and the answers get vague.

There is a better way to choose, and it is not about the technology. It is about the shape of the work.

#The test: three questions per task

Before automating anything, put it through three questions. A task that answers yes to all three is a strong candidate. A no to any one of them means wait.

1. Does it happen a lot?

Automation pays back on volume. Something done four times a year does not repay the effort of automating it, however annoying those four times are.

2. Are the rules written down and stable?

If you cannot write the rule as a sentence beginning "if", the task is not ready. "If the invoice matches the purchase order and receipt within tolerance, post it" is a rule. "Use judgement on the tricky ones" is not.

3. Is the underlying data good enough?

Automation applies your data faster. If supplier records are duplicated or spend is unclassified, you will get wrong answers at speed.

That third question is where most projects quietly fail. Gartner found that 74% of procurement leaders say their own data is not ready for AI. The same problem applies to plainer automation.

#The priority order

For a mid-sized company, this order reflects volume, how well defined the rules usually are, and how quickly you see the result.

1. Invoice matching and posting

This is almost always the right first move, because it is the highest-volume rules-based task in procurement and because you can prove the result.

The measure to watch is the touchless rate: the share of invoices that go from arrival to posting with no human involvement at all.

The cost difference is large. Ardent puts the average all-in cost of processing an invoice at $9.40. The best 20% pay $2.78. Everyone else pays $12.88. Cycle time follows the same pattern: 9.2 days on average, 3.1 days for the best, 17.4 for the rest.

Note what the touchless figures really say. Even the leaders touch half their invoices by hand. Anyone promising 90% touchless from a standing start is selling something.

2. Purchase order creation from approved requisitions

Once a request is approved, turning it into a purchase order is mechanical. Somebody rekeys what the system already knows.

The Hackett Group's leading procurement organisations process requisitions, purchase orders and PO change orders entirely electronically, and handle 27% more requisitions electronically than their peers. Their requisition-to-purchase-order cycle time is 58% shorter.

3. Catalogues and guided buying

This is the one most teams underrate. A catalogue does not speed up processing — it prevents the processing from being needed. The approved supplier and the agreed price are built in before anyone clicks buy.

Hackett's advisers reported that leading teams put 81% more indirect spend through electronic catalogues than their peers, and that technology let leaders reduce the cost per order by 76%.

It also fixes a behaviour problem without rules. People buy outside the process when the process is slow. Make the compliant route the fastest route and most of the problem disappears.

4. Approval routing

Before automating approvals, count them. Most slow cycle times come from too many approvers rather than slow ones. Automating a five-step approval chain nobody can justify gives you a fast chain nobody can justify.

Fix the thresholds first — who can approve what, at what value — then automate the routing. The first part costs nothing.

5. Contract renewal alerts

The cheapest item on this list. One list of every contract with its end date, notice period and auto-renewal status, and a reminder 90 days before each deadline.

This does not need a procurement system. A shared calendar works. What it needs is a named owner.

6, 7 and 8: onboarding, classification and reporting

Supplier onboarding forms collect structured data once, properly, instead of someone typing bank details from an email. Assisted spend classification speeds up categorising transactions, but needs a taxonomy your team agreed and human review of anything uncertain. Reporting comes last on purpose: an automated dashboard built on unclassified spend produces confident, wrong pictures faster than a human could.

#The deadline you do not control: e-invoicing

For many companies, part of this decision has already been made by tax authorities. Electronic invoicing is becoming a legal requirement across Europe on a published timetable.

Two things follow from this. First, if you trade with German or French businesses, being able to receive and send structured invoices stops being a choice. Second, a company that is already processing invoices electronically will find compliance straightforward, while one still handling PDFs by hand has a project on a fixed deadline.

#What not to automate

1. Anything where the rule is really a judgement. If two experienced people would handle a case differently, write the rule down first or leave it alone.
2. Supplier selection. Automation can prepare the comparison. It should not make the choice.
3. Exception handling, at least at first. Exceptions are where your business's actual complexity lives, and they are the worst place to start.
4. Anything sitting on bad data. Automating on duplicated supplier records multiplies the error rather than the productivity.
5. A process you cannot explain. If nobody can say why a step exists, remove it rather than automate it.

The consultancy Comprara describes the general failure well: technology bought before the process is fixed just embeds the problem at speed and scale.

#Be realistic about what automation returns

Panorama Consulting's 2026 ERP report, based on 170 organisations surveyed between January 2025 and January 2026 with a median annual revenue of $200.5 million, is a useful reality check on enterprise software projects generally.

Two findings stand out. The benefits most commonly realised to the extent expected were those related to productivity and efficiency — not the more ambitious strategic claims. And more than a quarter of organisations reported going over budget, with almost a quarter over schedule. The median project ran nine months.

Worth noting for anyone who has seen alarming figures quoted elsewhere: much higher failure rates circulate online attributed to this report series, but the 2026 edition's own text does not support them. Check the current report rather than the summary.

The practical lesson is to promise what automation reliably delivers — time and accuracy on repetitive work — rather than transformation.

#How to start, in four steps

1. Count the volumes. How many invoices, purchase orders and requisitions per month, and how many of each go wrong.
2. Time the work. Roughly how long each takes, and how much of the team's week that adds up to.
3. Score each task against the three questions at the top of this article.
4. Take the highest-volume task that passes all three, measure it for a month, then change it.

That baseline matters more than the tool you pick. Without a measurement taken beforehand, a successful automation still gets argued about afterwards, because nobody can say what changed.

One more thing worth checking before buying anything: how much of this your current systems already do. The Hackett Group found that 69% of organisations get their AI capability through features embedded in software they already own. The same is often true of basic automation — the feature is there and was never switched on.