Skip to content

Systems and records

Procurement dataset missing decisions: recover why each order was placed

By SourceX Editorial · Updated

10 sections

Short answer

A procurement dataset is missing decisions when it holds orders but not the reasons behind them. Recover each reason by tracing the order line back to the records around it: the requisition and specification revision, the quotes compared and rejected, supplier eligibility on the order date, the approval trail, any change versions and the receipt match. Label every reason recorded, linked, inferred or unknown rather than inventing one.

License terms

Procurement dataset missing

This page covers procurement dataset missing decisions: recover why each order was placed.Illustrative

Key takeaways

  • An ERP purchase order records the outcome of several decisions but rarely the reasons, which live in quotes, approvals, eligibility lists and email.
  • Split each order line into need, sourcing method, supplier choice, price, specification and change decisions before searching for evidence.
  • Label every recovered reason recorded, linked, inferred, retrospective, policy-driven, unknown or excluded, and never backfill reasons from outcomes.
  • Orders from reorder rules, agreements and emergencies need the rule, agreement or justification recovered instead of a supplier choice rationale.
  • Measure coverage per decision type with a blind second-reviewer sample, not per order.

Why does an ERP purchase order record the order but not the choice?

Because the choice is made before the order exists. A purchase order line stores what was bought, from whom, at what price, in what quantity and by when. The comparison of quotes, the check against the approved supplier list, the reading of the drawing revision and the phone call about lead time all happen upstream, and most ERP systems keep them, if at all, in separate records that the order export never touches.

Odoo shows the pattern clearly. Its documentation describes creating alternative requests for quotation for other vendors, comparing them line by line on a Compare Order Lines page, picking lines with a Choose button, and then cancelling or keeping the alternatives when the winning order is confirmed. The comparison survives as a set of linked, often cancelled, quotations. It does not survive as a reason field on the order.

Microsoft Dynamics 365 Supply Chain Management goes further but still leaves the reason optional. Bids can carry a reason code explaining acceptance or rejection only if someone has set up codes on the Vendor reasons page, and scoring criteria appear only when the RFQ case uses a solicitation type with a scoring method. Plenty of companies never switch either on.

Federal contracting is the useful contrast. The Federal Acquisition Regulation requires a source selection decision to be documented with the rationale for business judgments and tradeoffs, and lists source selection documentation, justifications and the list of sources solicited among normal contract file contents. Even there, the rationale lives in the contract file, not in the order record. A procurement dataset built from order tables alone is missing decisions by construction, and the fix is to go and get the file around the order.

Which decisions sit behind a single purchase order line?

At least six, and each one has a different home. Treating the order as one decision is the most common reason an enrichment project recovers the wrong evidence: a team finds the approval trail, labels the line explained, and never learns why that supplier was asked in the first place.

Split each order line into the decisions below before looking for evidence. The table shows where the outcome of each decision appears on the order and where the reason usually sits.

Which decisions sit behind a single purchase order line?
DecisionQuestion it answersOutcome visible on the orderWhere the reason usually lives
Need and quantityWhy buy this, this much, now?Item, quantity, requested dateRequisition, reorder rule, sales order or project demand
Sourcing methodCompete it, use an agreement or go to one source?Link to agreement or RFQ, often blankRFQ case, blanket agreement, single-source memo
Supplier choiceWhy this supplier over the others asked?Vendor fieldQuote comparison, bid reason codes, approved supplier status, buyer notes
Price acceptanceWhy was this price accepted as reasonable?Unit priceCompeting quotes, prior prices, price list, negotiation notes
SpecificationWhich revision and requirements were ordered?Item description, sometimes a revisionDrawing or spec revision on the requisition, engineering change records
Change after confirmationWhy did the order change once placed?Current values only, unless versionedOrder versions or journals, supplier change requests, email

The decision evidence enrichment inventory

This inventory is the working tool. Run it per order line, or per decision when lines share one sourcing event. Each row names an evidence item, where it is usually found, the key that ties it back to the order line, what it can and cannot prove, and the status label to record. Fill the status column for every row, including the rows where nothing was found; an explicit unknown is data, a blank is not.

At a glance

The decision evidence enrichment inventory

The rows are ordered roughly by how cheaply the evidence can be recovered, from structured system records to documents and correspondence. The last row is an outcome, not a reason, and is kept for scoring the decision later. It is usually the easiest to join: with Odoo's 3-way matching switched on, for example, a vendor bill is only paid once some or all of the ordered products have been received, so receipts and bills already point back to the order line.

The decision evidence enrichment inventory
Evidence itemTypical sourceJoin key to the order lineWhat it proves about the decision
Requisition or demand triggerRequisition, reorder rule, sales or project orderRequisition or demand line ID copied to the PO lineWho asked, what need was stated, whether demand was automatic
Specification revisionRequisition line, drawing register, PLM or document controlItem plus revision and the date the requisition was raisedWhich requirement the buyer was sourcing against at the time
Alternatives consideredLinked alternative RFQs, RFQ case replies, quote emailsRFQ case or alternative group IDWho else was asked and what each offered, including rejected bids
Stated selection reasonBid reason codes, scoring panel, buyer notes fieldRFQ reply or bid IDThe reason the buyer recorded, if any was recorded
Supplier eligibility at the timeApproved supplier list, vendor onboarding or hold statusVendor ID plus effective date rangeWhether excluded suppliers were ineligible rather than overlooked
Approval trailWorkflow history, approval statuses, delegation recordsPO number and versionWho authorized the commitment and under which limit
Single-source or exception memoJustification document, emergency purchase formPO, requisition or RFQ reference in the memoWhy competition was skipped and who accepted that
Customer or contract constraintSales order, customer-directed source clause, blanket agreementSales order line or agreement IDRequirements the buyer could not trade away
Change historyConfirmed order versions or journals, supplier change requestsPO number plus version or journal numberWhat changed after placement, when and at whose request
Delivery and match outcomeReceipts, three-way match, quality notifications, returnsPO line ID on receipt and bill linesWhether the choice delivered: on time, accepted, billed as ordered

What status should each recovered reason carry?

Use a short, fixed set of labels so reviewers and buyers can filter. The labels separate what a record says from what an analyst concluded, which matters more than the volume of evidence recovered.

  • Recorded: the source system or document states the reason in so many words, such as a bid reason code or a written justification.
  • Linked: the evidence exists and is joined by key, but the reason is implicit, such as three quotes where the chosen one was second-lowest and the lowest supplier was on hold.
  • Inferred: an analyst reconstructed the reason from circumstantial evidence, such as an expedite email two days before the order. Keep the inference text and the reviewer's name.
  • Retrospective: a buyer explained the decision afterwards in an interview. Useful, but kept apart because memory is shaped by how the order turned out.
  • Policy-driven: no person chose; a reorder rule, agreement release or catalog default produced the order. Recover the rule version instead of a reason.
  • Unknown: searched for and not found. Record where you looked so the next pass does not repeat the search.
  • Excluded: evidence exists but cannot be used, for example because it sits under a supplier NDA or contains personal data you will not process.

How do you recover a rationale that the system never stored?

Work outward from the order in a fixed order: structured records first, linked documents second, correspondence third, people last. Each step costs more and yields weaker evidence, so stop as soon as the decision has a recorded or linked reason.

Structured records include the alternative quotations Odoo links to an order, the RFQ replies and acceptance and rejection journals that Dynamics 365 generates when bids are accepted or rejected, and approval statuses. Dynamics 365 also creates a journal holding an exact copy of each confirmed order, and further journals when a changed order is reconfirmed, which gives you dated versions to compare instead of only the current values.

Documents come next: single-source justifications, quote PDFs, supplier qualification files and drawing revisions. Federal practice is a good model for what a minimal record of competition looks like. For small purchases, the FAR says records of oral quotes will usually just show the suppliers contacted and the prices and terms each one quoted. A spreadsheet with that much is enough to rebuild the comparison set.

Correspondence, usually buyer email, often holds the real constraint: a supplier who cannot meet the date, an engineer insisting on a revision. Treat it as evidence for a specific decision, extract only the passage that supports the label and keep the message ID, rather than attaching whole threads. Interviews come last, are labeled retrospective and should be run against the record as it stood on the order date, not against the delivery outcome.

One rule holds throughout: only evidence that existed at or before the decision can explain it. A late delivery or a quality reject is an outcome to record in its own column, never a reason to backfill.

When was the lowest quote not the right order?

When a constraint the order table does not show ruled it out. These are the decisions a model most needs to see, because a model trained on order rows alone will learn that buyers pick the cheapest available supplier and be wrong exactly where it matters. Check each enriched decision for the constraints below before labeling the choice.

Steps

When was the lowest quote not the right order

Minimum order terms are a good example of a constraint that is only visible in the quote. The FAR anticipates a supplier whose minimum order price or quantity exceeds the stated requirement and tells the contracting officer to ask the requiring activity to confirm or change the requirement, then document the final action. A commercial buyer facing the same quote leaves a trace only if someone writes it down.

  • Specification mismatch: the cheapest supplier quoted an earlier drawing revision or a substitute material.
  • Eligibility: the supplier was not approved for that part family, was on hold, or had not finished onboarding on the order date.
  • Lead time against need date: the faster supplier was the only one that met a customer or line-down date.
  • Minimum order quantity or price: the quote forced buying far more than required.
  • Customer or contract direction: the customer specified the source, or a blanket agreement committed volume to one supplier.
  • Authority: the preferred option exceeded the buyer's approval limit and the approved alternative was chosen instead.
  • Total cost terms: freight, tooling or payment terms changed which quote was actually cheaper, without any unit price showing it.

Worked example (illustrative): three quotes for a machined bracket

This example is invented to show the inventory in use; the counts and durations are illustrative. A contract manufacturer's order export shows one purchase order line for a machined bracket, placed with Supplier B. On price alone, Supplier B was the second-lowest of three quotes, so the order looks like a mistake or a favor.

Enrichment recovers the following. The requisition carries drawing revision D and a need date 21 days out, driven by a customer sales order. Two alternative RFQs are linked to the order, both cancelled: Supplier A, the lowest, and Supplier C, the highest. The approved supplier list shows Supplier A approved for revision C only on the order date; its approval for revision D came 40 days later. Supplier C quoted a 35-day lead time against Supplier B's 14 days. The approval trail shows the buyer's manager approved the order within the buyer's limit. There is no bid reason code because the company never configured them.

The labeled result is three decisions with three different statuses. Supplier choice is linked: A was ineligible for the revision ordered and C could not meet the date. Specification is recorded on the requisition. Price acceptance is unknown, because nothing shows whether Supplier B was asked to match. The delivery outcome, received on day 15 with a one-day slip and accepted at inspection, sits in its own column and does not change any label.

Without the inventory, this line would have taught a model the wrong lesson. With it, the line becomes a clean example of an eligibility constraint overriding price, with an honest gap where the negotiation evidence is missing.

Which orders have no choice to recover, and what should replace the reason?

Orders produced by rules, agreements or emergencies have a different decision structure, and forcing a supplier choice rationale onto them produces fiction. Separate them before measuring coverage.

Replenishment orders are the largest group. In Odoo, reordering rules trigger a request for quotation whenever forecast stock falls below the minimum, with the vendor, price and quantity already filled in. The decision worth recording is the rule: its minimum and maximum, the vendor sequence and who last changed them. Label these lines policy-driven and attach the rule version in force on the order date.

Releases against a blanket agreement inherit the supplier choice made when the agreement was signed, so recover that sourcing event once and link every release to it. Emergency and single-source buys need the justification and the person who accepted it; under the FAR, a justification for other than full and open competition must set out the facts and rationale, the market research done and the efforts to solicit other sources, which is a useful checklist even for private companies writing their own memos.

How do you know the enriched dataset is good enough to use?

Measure coverage per decision type, not per order, and agree the acceptance bar before you count. A dataset where every line has an approval trail but almost none has alternatives looks complete on a per-order count and is useless for supplier selection.

Report four numbers for each decision type: the share of lines with a recorded or linked reason, the share inferred, the share unknown, and the share policy-driven. Then have a second reviewer relabel a random sample of enriched decisions blind to the first labels and report how often they agree. Low agreement on inferred labels is the usual signal that inference rules need tightening or that those lines should be downgraded to unknown.

Finally, check the timing rule mechanically: every evidence item's timestamp must be at or before the decision it explains, apart from outcome columns. Violations usually come from approved supplier lists or specification registers that were overwritten rather than versioned, and they are a reason to drop the eligibility label for the affected period rather than guess.

What to settle before enriched procurement records are licensed

Enrichment pulls in material the order table never contained: competitors' quotes, supplier qualification files, buyer email and sometimes customer drawings. Each source needs its own answer to three separate questions: whether you hold the records, whether a supplier NDA, customer contract or quote terms restrict their use, and whether they contain personal data that has to be removed. De-identifying a quote does not by itself create permission to license it, and pricing confidentiality usually needs the most attention.

A manufacturer or distributor that decides to license enriched procurement records can work through SourceX, which coordinates sourcing, rights review, contracting, preparation, delivery and payment. Volume expectations vary, but a common bar is two or more years of steady history and tens of thousands of orders in one system, with room for rare work at smaller sizes. The process opens with a short assessment and no data upload, and delivery follows only an approved scope and a signed agreement.

Frequently asked questions

Can purchase order data alone train a model to choose suppliers?

Not reliably. An order table shows which supplier won but not who else was asked, which suppliers were ineligible or which constraint ruled out the cheaper option. A model trained on it tends to learn that price wins, which fails on exactly the orders where an eligibility, lead time or specification constraint decided the outcome.

Should we interview buyers to fill the gaps?

Only after structured records, documents and correspondence have been exhausted, and only against the record as it stood on the order date. Label interview answers as retrospective and keep them separate from recorded reasons, because explanations given after the delivery outcome is known tend to drift toward it.

What if our approved supplier list was overwritten instead of versioned?

Then you cannot prove a supplier's eligibility on an older order date. Rebuild what you can from onboarding dates, hold notices and audit records, and mark eligibility as unknown for periods you cannot reconstruct instead of using today's list.

Do rejected quotes belong in the dataset?

Yes, if the rights allow it. Rejected quotes are what make a supplier choice a decision rather than a record of a purchase. They often carry the strictest confidentiality terms, so check the quote terms and any supplier agreements before including them, and consider keeping relative rank instead of the price.

Which orders should be enriched first?

Start with competitively sourced orders where alternatives are already linked in the ERP, then single-source and exception buys with written justifications. Leave replenishment and agreement releases for a separate pass that recovers rule and agreement versions rather than reasons.

Sources

  • Odoo lets buyers create alternative RFQs, compare them on a Compare Order Lines page, choose lines, and cancel or keep the alternatives when the order is confirmed. Source
  • With Odoo's 3-way matching enabled, vendor bills are only paid once some or all of the ordered products have been received. Source
  • Odoo reordering rules trigger an RFQ each time forecast stock falls below the minimum, with the vendor, price and reorder quantity filled in. Source
  • In Dynamics 365 Supply Chain Management, bids can carry reason codes for acceptance or rejection if codes are set up on the Vendor reasons page, scoring depends on the solicitation type, and acceptance and rejection journals are generated. Source
  • Dynamics 365 stores an exact copy of each confirmed purchase order in a journal and adds journals when a changed order is reconfirmed, giving a history of confirmed versions. Source
  • FAR 15.308 requires the source selection decision to be documented, including the rationale for business judgments and tradeoffs. Source
  • FAR 4.803 lists records normally in contract files, including justifications and approvals, the list of sources solicited and source selection documentation. Source
  • FAR 13.106-3 says records of oral quotes usually show the suppliers contacted and the prices and terms quoted, and requires documenting the action taken when a minimum order exceeds the requirement. Source
  • FAR 6.303-2 requires a justification for other than full and open competition to contain facts and rationale, a description of market research and the efforts made to solicit sources. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies