Our intake process is completely manual and slowing down procurement, what do we automate first?

Automate the checking step, not the form and not the approvals. Checking a requisition against your policies, contracts and approved suppliers is the same work on every request, it is mandatory so the queue never clears, and doing it the moment a request arrives removes the corrections that cause round trips. Pactum's Alignment Agent does this before money is committed.

Our requests come in by email and spreadsheets, not through a system. Does this still work?

Yes, and that's the usual starting point. The agent becomes the front door, so people describe what they need in plain language in Microsoft Teams or a chat window instead of writing someone an email. What comes out the other end is a complete, coded requisition in Coupa or SAP Ariba, so your P2P system stays the system of record. See all integrations.

Do our policies need cleaning up before we start?

No, and don't wait until they are. Most teams find their rules spread across policy documents, category guidance, contract terms and a few people's heads. That's the normal starting point. The work is collecting them into one rulebook, not rewriting them first, and the gaps surface as the agents run, which is usually the first time anyone's seen them in one place.

What happens to requests already in the queue when we switch this on?

Nothing changes for them. Agents start in observation mode, checking every incoming request while your reviewers still decide, so work already in the queue carries on as it is. You'll see the agent's call next to your reviewer's on real requests before anything is handed over.

Who decides what the agents handle on their own?

You do. You set the thresholds for what agents handle alone, and you decide how fast each rule and category moves. Anything outside the rulebook goes to a buyer with the reason attached, strategic sourcing stays with people, and the rulebook stays yours. Agents apply your policies, they don't rewrite them.

How long does this take to stand up?

Most teams spend a week or two putting their existing rules into one rulebook, then run three to six weeks in observation mode with agents checking every request while reviewers still decide. Cycle time usually starts falling during observation, because corrections surface at the start instead of a week later.

Your Procurement Intake Is Manual. What Should You Automate First?

Most of the elapsed time in a manual intake process is queueing, not work. See where the days actually go, why the checking step comes first, and why handoffs are the wrong target.

Where the time actually goes

Before deciding what to automate, it’s worth knowing which step is eating the days. Pull one requisition that took three weeks and walk it backwards. You’ll find the same thing most teams find.

Your colleague fills in the form
Minutes, if they happen to know the coding. Days, if they guess and somebody has to go back to them.
It sits in a queue
This is usually your single biggest block of elapsed time, and none of it is work. It's a request waiting for a reviewer to be free.
A reviewer checks it against policy
Category rules, approved suppliers, contract coverage, thresholds. Real work, and it's the same work every time.
It goes back for a correction
Wrong code, missing quote, supplier who isn't set up. Every round trip puts it at the back of the queue again.
It's handed to the next team
Procurement to category, category to sourcing, sourcing to finance. Each handoff is another queue.
Someone decides how to buy it
Catalog, contract, quote, or competition. Often skipped when the queue is long, and that's where your savings went.

Almost none of your elapsed time is work. It's requests waiting for a person to become available.

That queue is also why a busy quarter hurts so much more than a quiet one. If your pain shows up in bursts rather than every week, read how agents absorb a volume peak alongside this.

What to automate first

Automate the checking step. Not the form, not the approvals, and not the handoffs.

Why that order? Four reasons.

It's the step that repeats identically.
Checking a requisition against your policies, your contracts, and your approved suppliers is the same work on every request. That makes it the step where an agent matches a good reviewer most closely.
It's mandatory, so it never clears.
Unlike negotiation, nobody gets to skip the check. So the queue in front of it doesn't go away on its own, no matter how hard your team pushes.
It kills the round trips.
Most of your corrections exist because somebody found a problem late. Check everything the moment it arrives, fix what can be fixed, and the request stops bouncing.
It unlocks the step after it.
Once requests arrive clean, how to buy can be answered automatically too. And that's where the money is.

At SUEZ, automating the review step cut requisition cycle time by 90%, took away 80% of the manual review work, and gave more than 10% of buyer time back to strategic sourcing.

90%: shorter requisition cycle time
80%: less manual review work
10%+: buyer time returned to sourcing

Why handoffs are the wrong thing to fix first

Handoffs look like the obvious target. The pain is visible, everyone complains about them, and the fix feels like something you can design your way out of.

So you map them. You streamline them. And six months later? The work is stalling somewhere else.

A handoff exists because the next person holds knowledge or authority the last one didn’t have, so taking the handoff away without moving the knowledge doesn’t remove the wait. It just moves it somewhere less visible.

Check every request against one rulebook, at the point it arrives, and the handoffs go on their own. Whatever each team was waiting to supply is already attached.

What happens to the data entry

Data entry is the part colleagues hate and the part your team spends the most time correcting. It goes away when the request stops being a form.

With Pactum, someone describes what they need in plain language, in Microsoft Teams or in your P2P system. Pactum’s Intake Agent works out the category, the commodity and GL codes, the supplier, and the contract from your own data. Then the Alignment Agent checks the finished requisition before any money is committed.

Pactum has now completed more than one million requisition checks.

What stays manual, on purpose

Automating the check doesn’t mean automating the judgment.

Your thresholds. You decide what agents handle alone and what always reaches a person, by value, category, or supplier.
Your exceptions. Anything outside the rulebook goes to a buyer with the reason attached.
The strategic work. Complex sourcing, supplier relationships, and the deals that need you in the room.
The rulebook itself. The agents apply your policies. They don’t rewrite them.

A realistic sequence

Weeks one to two. Get your existing rules into one rulebook. They don’t need tidying first. The Alignment Agent will show you where they contradict each other, which is usually worth the exercise on its own.
Weeks three to six. Run in observation mode. The agents check every request against your rulebook while your reviewers still decide. Then you compare.
After that. Hand over the requests where the agent matched your reviewers, keep the rest with your people, and widen the thresholds as the match rate holds.

Most teams see cycle time start falling during observation mode, before they’ve handed anything over. The corrections are just surfacing at the start now instead of a week later.

The step after that one is getting people to use the new route at all. If you have been through a system change before and watched adoption stall, why intake adoption stalls after a system change covers what actually moves it. For the full picture of what each agent does across intake, start with the guide to intake agents in procurement.