Supplier Scorecards From Documents You Already Have

Why spreadsheet scorecards die

The request lands two weeks before the quarterly business review: scorecards for the top vendors, covering delivery performance, pricing, responsiveness, and issues. Everyone knows what happens next. Someone exports receipt history from the ERP, digs through email to find when acknowledgments actually came back, pulls the invoice register, and reconstructs the quarter's disputes from memory and old threads. The scorecard exists for one meeting, then goes stale until the next request revives it.

That cycle is structural, not a discipline problem. Three things kill the hand-fed scorecard every time:

• Manual feeding.A spreadsheet scorecard is only as current as its last feeding, and feeding it is nobody's first job. The numbers decay the moment attention moves elsewhere, and a stale scorecard is worse than none, because it still looks authoritative.
• Recency bias.Assembled in the two weeks before the review, it weights whatever is easiest to find. Last month's late shipment makes it in; the three from January do not. It is a memory-weighted sample presented as a history.
• One owner.The workbook lives with one analyst, and so does the methodology. When that person is out, the scorecard is out, and no two owners ever compute "on-time" the same way, so scores drift across quarters even when the vendor does not.

The fix is not a better template. Every number the scorecard needs is already carried by documents your vendors send in the ordinary course of business, whether or not anyone logs them. The work is a set of joins, and joins are what software is for.

The four metrics and the document pairs that carry them

Each metric below is a join between documents you already receive. None of it requires a supplier portal, a survey, or a data-entry pass, which matters because the suppliers you most need to score are usually the ones that will never adopt a portal.

1. On-time rate: PO promise date against receipt date

The PO carries the promised date, and the acknowledgment either confirms that date or moves it. The receipt or bill of lading carries the actual. Join the two per PO and you get the two numbers a review needs: the vendor's on-time rate, and its typical lateness when it is late. The second number is the one spreadsheets never carry: a vendor that misses by two days and a vendor that misses by two weeks can both show the same on-time rate, and only one of them threatens a production schedule. The same history also powers delivery risk on open POs, which is vendor-level by design: a vendor's historical on-time rate and typical lateness, applied to what it currently owes you. The full derivation is in the on-time delivery guide.

2. Response lag: PO sent against acknowledgment received

The gap between the PO going out and the acknowledgment coming back is the cleanest responsiveness measure there is, and it needs two layers. The first is the flag: POs that pass a follow-up window with no acknowledgment get surfaced, so silence never coasts. The second is the learned norm: each vendor's normal confirmation time is learned from its own history, so a vendor that normally confirms in two days and just took six is a signal, even though six days would be routine for a different vendor.

The acknowledgment carries more than a timestamp. The quantities and prices the vendor echoes back are cross-checked against the PO deterministically, so a quiet requote or a short-confirmed quantity buried inside a routine confirmation is caught when the document arrives, not at the receiving dock. Those catches feed the fourth metric below.

3. Price stability: invoice price against contract and history

Two checks produce this one. Contract compliance catches the invoiced price that sits above the contracted rate. Price creep catches the invoiced price drifting against the vendor's own history even where no contract pins it. Rolled up per vendor per quarter, that becomes the stability read a review wants: prices held, drifted, or stepped, with the specific invoice lines behind each answer. The same data assembles the negotiation page when it matters most: contracts get flagged at 120, 60, and 30 days before renewal, with briefs showing creep above the contracted rate alongside alternative-supplier counts.

4. Dispute frequency: the exception history

Nobody needs to maintain a dispute log, because every failed cross-check already left a record: duplicate invoices in their exact-number, same-amount, and fuzzy variants, including invoices already paid; split-invoicing patterns; invoiced prices above contract; acknowledged quantities that do not match the PO; and early-payment discount windows that expired unused, with discount capture rate tracked as its own number. Count those exceptions per vendor per quarter and dispute frequency falls out. Each finding also landed in the team's work queue when it happened, ready to assign, so the QBR number is a recount of work already handled, not a reconstruction. The fraud-facing side of the same history is covered in vendor intelligence for AP.

What a QBR-ready vendor view looks like

The end state is that "assembling the scorecard" stops being a verb. You open the vendor's page and the review material is already there: a risk score, enrichment on the business itself, including shell-company flags for the vendor nobody has ever visited, and the four performance metrics above, computed from the full document record instead of the fraction someone remembered to log. When a side question comes up mid-review, the answer is a single plain-English question to the AI chat over the document graph, not an action item for next week.

None of this replaces the ERP. Your ERP is a system of record: it stores the data you enter. The scorecard metrics live mostly in the documents around it, the acknowledgment, the bill of lading, the invoice attached to an email, which is exactly the paper an ERP never reads. The intelligence layer reads those documents, connects each one to the PO, contract, vendor, and item it references, and keeps score continuously. How that same vendor record changes sourcing decisions is covered in vendor intelligence for procurement.

Two vendors, side by side

On a hand-fed scorecard, both of these vendors plausibly rate "green," because each of Vendor B's individual misses got resolved and none got logged. The document-derived view shows the pattern instead: confirmation time stretching from three days to seven, prices drifting above contract on two items, and exceptions clustering in one quarter. That changes the QBR conversation from "you were late once," which any vendor can deflect, to "your confirmation time doubled this quarter, and here are the POs," which is a discussion about specific documents at line level.

The scorecard compounds

Every metric above is a baseline learned from the vendor's own history: its normal confirmation time, its typical lateness, its own price history. Those baselines sharpen as documents accumulate, so the second quarter's scorecard is sharper than the first, and comparisons across quarters finally hold up because the method never changes hands.

The feeding, meanwhile, is the forwarding your team already does. Procurement emails go to a Kynthar address; bodies and attachments, PDF and Excel alike, are read together; there are no templates and no per-vendor setup, and historical documents can be uploaded to backfill the record. A document becomes part of the vendor's score within minutes of arriving, because the clock starts when the email lands, not when a human gets around to opening it.