From Policy Checker to Procurement Brain

Procurement policy builds up over decades. There is the written policy, with its SOPs and category cards. There is the unwritten tribal knowledge of how things are really done. And there is the policy that is written down somewhere but that nobody remembers exactly. New rules go on top of old ones until they contradict each other. In government, whole teams of hundreds of people work on nothing else.

Every request has to pass through all of it. At a fintech growing from Series B to Series C, every new request went through an Excel form: about 20 fields for the requester, then about 75 more cells for a procurement manager to check before it moved on.

Most requesters try to do the right thing, but most requests still arrive with something wrong: the wrong supplier entity, a missing contract, a rule nobody told them about. Each one goes back to the requester, and the back-and-forth stretches a simple purchase over two or three days or more.

Many businesses set aside dedicated people in operational procurement for this work. In reality, their job is checking every request against a vast set of buying rules and category strategies, and making sure the whole organization applies them. A digital worker is far better suited to it: it checks every request against every rule in under a minute, the same way every time.

The checking also pulls skilled people down to the lowest-value work. Pactum has met highly paid category managers who lose hours of every day to it, time meant for strategy and negotiations, and procurement becomes the gatekeeper in everyone’s eyes.

Daniel started in defense procurement, moving between the Ministry of Defence and private sector defense. The work came with nine months of security clearance and the promise of buying tanks and visiting suppliers. His first two years were mostly checking requisitions: check, check, update the policy, phone the Ministry of Defence, then wait three months for an answer. The policy instructions lived in huge books behind every desk, and to buy anything you found the category and looked it up. As his roles grew, the checking became one part of the job.

Dan worked in procurement at Rolls-Royce, where the operations center team checked requisitions all day. Every buy linked to US Department of Defense work needed its own sourcing pack, with minutes and an audit trail, and the packs filled filing cabinets in the office.

Daniel later led contract management and compliance at that fintech, where the checks ran in the spreadsheet in FIG. 01. People he worked with still do this work today: they set up policies, make sure people buy properly and run the checks, and fit the supplier relationships in when there is a spare minute.

Dan knew from his own job, and from five years with Coupa customers, that requisition checking looked the same almost everywhere. At Pactum he took it to a two-day hackathon. “I wouldn’t say I was the brain behind it,” he says. He brought the problem, and some very clever people solved it.

He had pictured automating the steps he used to take by hand. The team went much further and built one place that holds the context of every policy at once. It has since grown into the Alignment Agent.

The agent checks every request against every policy in under a minute, with no value threshold and no sampling. The requester gets the reason while the request is still in their head. Many pull it back, fix it and resubmit, and the request reaches a PO without a buyer touching it.

It works alongside intake. No CPO wants a locked door into procurement, so there is always a way in, from a free text request to an email asking for a new order. Intake guides the request, and the Alignment Agent checks whatever arrives.

Procurement knowledge is spread across people and tools: category managers, heads of sourcing, CPOs and CFOs, Slack threads, Teams chats and the Dave in the office who knows every commodity code. Copilots in those tools help, but they do not hold the rules of how procurement operates. As Daniel puts it, “your brain doesn’t live there.”

The brain holds it in one place: category strategies, policies, contracts, the attachments behind every request and the decisions reviewers make. Change a category strategy once and the guidance is in front of the next requester, and every request is checked against it when it is submitted. Over time Dan sees it joining up wider decisions too, such as a category strategy that moves to one global contract while the company strategy favors local contracts.

Strategy turns into policy the same way. The team fed a genuine category strategy to the agent that builds agents, and it came back with the 12 policies needed to carry it out. When the strategy changes, the brain reviews every policy and shows what needs to change.

The brain also learns. Reviewers mark the agent’s calls right or wrong, and the rules it keeps getting wrong show procurement where its own policy is unclear.

Autonomy comes a step at a time. Observer mode comes first: every rule runs against decisions a customer’s own reviewers already made, and at one customer the agent agreed with them 99% of the time before it acted alone. From there, rules are switched on one by one as the numbers earn it.

That is what lets digital workers operate on their own. Every digital worker that buys for you, whether it guides a request, checks it or negotiates with a supplier, follows the same rulebook, kept current by the people who know procurement best.