The PO-first culture problem: why buyers work around the process, and how agents fix it

A PO-first policy requires an approved purchase order before any purchase is committed – before the phone order, the email go-ahead, or the signed engagement letter. The PO created at that moment records what was agreed, at what price, under which contract, and with whose approval, while the company can still change its mind. Buyers work around the policy whenever following it is slower than ordering directly, and that speed gap is what actually sets purchase order compliance.

This entry sits in a wiki built for companies where AI agents carry the operational work – drafting requisitions, matching invoices, reading contracts – and people keep the judgment calls. Most glossaries treat PO compliance as a policy to enforce; this page treats it as a race the compliant path keeps losing.

Companies adopt PO-first for the record chain it creates

The purchase order does three jobs no other document can. It applies negotiated pricing at the moment of ordering, so contracted discounts actually reach the invoice. It gives category managers forward visibility: a requisition can still be consolidated or steered, while an invoice is already spent. And it anchors downstream matching – three-way match, the line-by-line comparison of an invoice against the purchase order and the goods receipt (the warehouse record of what arrived), needs a PO to compare against. The purchase order is the hinge record of the source-to-pay lifecycle, and a purchase that skips it breaks the chain in both directions.

The workaround is usually a rational decision

Take a plant engineer at 6 a.m. with a packaging line down. The failed part is a $6,700 servo drive, the regional distributor has one on the shelf, and each hour of idle line costs a multiple of the part. The compliant path runs through a requisition in Ariba, a cost-center approval, a buyer's queue, and requisition-to-PO conversion – two days on a good week. The engineer calls the distributor at 6:15, the drive ships by noon, and the call ends with "PO number to follow". Measured against the job the engineer is paid to do, which is uptime, that was the right call. The process asked for a trade of two days of production against a document, and the trade fails on its own arithmetic.

The second scenario is quieter. A marketing manager holds a $15,000-a-month agency retainer under a master services agreement. The retainer fits no catalog category: the punchout catalog covers laptops, MRO supplies (maintenance, repair, and operations items), and office furniture, and there is no tile for ongoing creative services. So each month the agency sends an invoice with no PO behind it, AP codes it by hand, and after a year nobody involved thinks of it as a workaround. It is simply how the retainer gets paid.

The costs land downstream, on people who never saw the order

Both purchases become maverick spend – spend committed outside the negotiated channel – and the cost arrives in stages. First at the invoice: each one lands as a non-PO invoice with nothing to match against, so someone in AP reconstructs what was ordered, who approved it, and which cost center carries it. Second at the price: contract pricing attaches at the PO, so the negotiated 12% distributor discount never touches the servo drive and the company pays list – roughly $800 on one emergency part. Third at the category level: the spend reaches the category manager months later as a general ledger line, too late to consolidate, forecast, or count toward a volume rebate. One engineer's phone call is trivial. The same call across forty plants, every week, is a category strategy quietly failing.

Mandates and catalogs punish the symptom

The standard responses raise the price of the workaround and leave the slow path slow. A no-PO-no-pay rule, under which AP rejects any invoice arriving without a purchase order number, mostly produces the retroactive PO – cut after the invoice shows up, dated to look as if it preceded the order. That document carries the full administrative cost of a real PO and none of the control value, and a false assertion besides: the file now claims a control that never ran. Enforcement teaches concealment. Compliance training works for a quarter and fades; it argues against incentives instead of changing them. And catalogs, genuinely useful for the head of the demand curve – laptops, safety gloves, toner – never reach the tail where the agency retainers and specialty gaskets live. This is one of the places procurement transformation programs stall: the policy stays officially in force while the compliance number goes flat.

PO-first compliance is a latency problem

People comply with processes that are faster than their workaround. The engineer's workaround took a ten-minute phone call; the compliant path took two days. That ratio made the decision. The proof runs the other way too: where the catalog is the fastest way to get a laptop, purchase order compliance sits near 100% without enforcement. Low compliance clusters in urgent, small-dollar, and uncatalogable purchases because those are the purchases where the process is slowest relative to a phone call. Every low-compliance category is a measurement of how slow the compliant path is, and the program goal shifts from policing behavior to closing latency.

Agents shorten the compliant path until it wins

Agentic procurement automation attacks the latency at three points.

1. Intake that keeps pace with the phone call. An agent reads a plain-language request – "packaging line 4 servo failed, Motion Industries has a Lenze i550 in stock, about $6,700, need it today" – and drafts the requisition itself: supplier matched to the master record, GL coding and cost center filled in, the right approver identified and pinged on their phone. The engineer sends one message instead of dialing the distributor and has an approved PO before the distributor opens.
2. Honest paper for the order that already happened. Some purchases will still precede their paperwork; a down line waits for nothing. The agent papers those the same day, with the true order date, flagged as post-hoc. The record stays truthful, and the flag turns concealment into signal: post-hoc POs map exactly where the process is still too slow.
3. The tail handled as structured non-PO. The agency retainer never needs a pretend catalog entry. The agent validates each monthly invoice against the master services agreement – rate, term, scope, remaining budget – so the purchase gets contract-backed control without forcing it into a PO shape it does not fit.

Requisition intake shows up early in most answers to which procurement processes to automate first because its payoff is measured in adoption. When the compliant path wins the race, people take it voluntarily.

Measure the latency gap and coverage will explain itself

Three numbers show whether this is working. PO coverage rate by category – the share of spend committed on a purchase order before the invoice arrived – because the aggregate hides everything. Workaround latency against process latency – the phone call versus the requisition – since that ratio predicts compliance better than any policy document. And the trend after the process stops being the slow path: coverage should climb as each latency gap closes, and any category that fails to move points at a path that is still too slow. A compliance chart read this way stops being a scorecard of buyer behavior and becomes a diagnostic of process speed.

Fragment builds AI agents that work this way across the procure-to-pay process – drafting requisitions from plain-language intake, papering post-hoc purchases honestly, validating tail spend against contracts – inside a company's existing SAP or Ariba environment, with nothing ripped out or replaced. If your compliance report clusters the way this page describes, see how Fragment's workflows run or request a demo.