You Manage Only What Your System Can Export
A diligence team working a procurement assessment asked a target company for line-item purchase data.
The export is the finding
A diligence team working a procurement assessment asked a target company for line-item purchase data. The data came back. It came back as tens of thousands of pages of PDF documents, which then had to be converted into a usable spreadsheet before anyone could sort a single category.
Two sentences from that session are worth keeping together. The company's leadership had described its purchasing system as a strong one, used on every job, giving buyers feedback in live time. The same people said the only way to get reporting out of it was to run it through their own conversion tooling.
Both statements can be true at once, and reading them as a contradiction misses the useful part. A system can be good at the transaction and have no path to the analysis. It can authorize, release, receive and pay with very few touches, and still have no export that puts one row per receipt line in front of a buyer. Those are separate capabilities, and most buying organizations only ever test the first one.
The practical consequence landed inside the week. The report that went out on the committed date was built from older purchase order data with no item detail, so it took the form of broad category trees with savings assumptions attached. That is a legitimate instrument. It is also a different instrument from an item-level finding, and the data format, not the analyst, decided which one the date would get.
Grain is a capability, and so is the key
Treat the purchase record as a measuring instrument and two properties decide what it can measure.
The first is grain. Supplier-level and purchase-order-level data support one family of questions: who we pay, how much, on what terms, how concentrated. Those questions lead to price levers, consolidation and terms. Item-level data supports a different family: what exactly we buy, in what unit of measure, at what unit price, across which plants, and how many distinct part numbers carry the same specification. Those are the questions that lead to specification change, and specification is where most packaging cost actually sits. A category managed off purchase order data is managed at the price of the thing you already buy. Nobody decides that. The export decides it.
The second is key stability, and it fails later, after the money is supposed to appear. In a separate engagement, a savings benchmark could not be built because the part numbers on the final transition file did not reconcile to the original award data. Items that had previously rolled up into one number now split across several. Nothing was lost and nothing was hidden. The join key changed, and the before and after could no longer be set side by side at all. A number you cannot reconcile is not a disputed number; it is an absent one, and the dispute that follows is about who absorbs the gap.
Put the two together and you get the rule I apply to any purchase data request. The item number is the unit of leverage and the unit of proof. Where it is absent, you can negotiate price but you cannot change a specification. Where it is unstable, you can change a specification but you cannot prove what it earned.
Run the export test before you commit to the date
This is checkable in an afternoon, which is the reason to do it before a date is set rather than during the work.
Pick one packaging category. Ask for twelve months, one row per receipt line, with item number, description, quantity, unit of measure, unit price, plant, supplier and purchase order reference. Then watch three things: how long it takes, who has to be involved, and whether unit of measure arrives populated and consistent. Unit of measure is the quiet one. Mixed or blank units turn unit price into a number that cannot be compared between two suppliers of the same item, and that defect survives every later cleanup because nobody can tell which rows were wrong.
Then run the identity check on the same file. Count distinct item numbers, then count how many of them carry the same specification under different numbers. Pull the items whose numbers changed in the last two years and see whether the predecessor number is recorded anywhere a buyer can reach. If a specification change is coming, write the crosswalk from old number to new number before the change goes in, not after, because the file that would let you build it later is the file the change replaces.
If the export cannot be produced, you still have a decision rather than a delay. You can accept a category-grain estimate and say on the artifact that it is one, which keeps the date and keeps the claim honest. You can hold the date and build the item-level file first, which is the right call whenever the output is a savings commitment someone will be paid against. What you cannot do is accept a category-grain estimate and then let it be read as an item-level finding, because that gap closes at invoice time, in a conversation about credit, with no reconcilable baseline on the table.
The ownership is specific. One named buyer per category owns the file, its grain and its key, in the same way they own the supplier relationship. Reporting capability is a procurement deliverable, not an information technology request.
What a well-run buying record looks like
Any packaging category can be exported for twelve months at one row per receipt line, with item number, description, quantity, unit of measure, unit price, plant, supplier and purchase order reference, by the buyer who owns it, without a special request and without a conversion step. Unit of measure is populated on every row and consistent within an item. Every item number changed in the last two years carries its predecessor number. The count of distinct item numbers sharing a single specification is known and reported, because that count is a live duplication list. Each savings claim names the item numbers it covers and the volume period it is measured over, and each one still reconciles after the specification changes.
The next test
Ask for the file on one category this week and keep the clock. The elapsed time, and the state the unit of measure column arrives in, tell you which lever your organization currently has and which one you have been assuming.