Rights and contracts
Procurement workflow data supplier packet: what to include
By SourceX Editorial · Reviewed by Noah Loul ·
9 sections
Short answer
A supplier packet for procurement decision histories should contain a scope statement, a record inventory, a small linked sample of complete purchase cases, a rights completeness checklist for every source and counterparty, and a note on exclusions. The sample must show choices, not just orders: the alternatives, the specification and supplier approval in force, the authority behind the buy and the delivery or quality outcome.
License terms
Procurement workflow data
Key takeaways
- A purchase order table records what was bought; a decision history adds the rejected alternatives, constraints, approvals and outcomes that explain the choice.
- Each purchase case needs one identifier that joins requisition, quotations, PO, receipts, bill and quality records.
- Rights completeness keeps access, privacy, ownership and contractual limits separate for every record source and counterparty.
- Sample cases should be chosen by a stated rule and scored on a utility checklist, with results reported as shares of the sample.
- Constraints such as specification locks, customer direction and supplier qualification turn an expensive-looking buy into a correct one, so label them.
What should a supplier packet for procurement decision histories contain?
A procurement workflow data supplier packet should let a prospective licensee judge two things before seeing the full dataset: whether the company has the rights to offer each record type, and whether a small sample really shows purchasing decisions rather than transactions. It contains a scope statement, a record inventory, a linked sample of a few complete purchase cases, a rights completeness checklist for every source and counterparty, and a short note on what was excluded and why.
The distinction matters because procurement data is easy to export and hard to license. A purchase order table is a ledger of what was bought. A decision history shows why that supplier, at that time, under that specification and approval, and what happened when the parts arrived. Buyers training procurement or supply chain agents need the second, and the second is also where supplier pricing, customer constraints and personal details sit.
This guide sets out the packet, the rights completeness and sample utility checklist, the constraints that make the obvious choice wrong, and an illustrative packet with one exception that removes a whole supplier relationship from scope.
Why a purchase order table is not a decision history
A purchasing decision is the choice among options under constraints, and most of those options never become a purchase order. Odoo's documentation describes calls for tenders as alternative requests for quotation sent to several vendors, with product lines compared before deciding which products to buy from which vendor. The losing quotations are the alternatives a model needs to see; a PO-only export drops them.
Standing agreements hide the decision further. Odoo describes blanket orders as long-term purchase agreements with a vendor to deliver products on a recurring basis at predetermined pricing. Each release against a blanket order looks like a routine PO, while the real sourcing decision happened once, when the agreement was negotiated.
Outcomes live in other documents. With 3-way matching enabled, Odoo's documentation says vendor bills are only paid once some or all of the ordered products have been received, so the receipt and the bill are part of the case, not separate tables. Process mining practice treats this as object-centric data: the OCEL 2.0 specification models events linked to several objects, such as orders, items and deliveries, and records object-to-object relationships, which is the shape a decision history needs.
Which records link a purchase choice to its constraints?
Each purchase case in the packet should carry identifiers that join the request, the alternatives, the authority, the order and the outcome. The table maps the evidence to where it usually sits and what goes wrong when it is missing.
| Evidence | Typical source | What it shows | If it is missing |
|---|---|---|---|
| Purchase requisition and need date | ERP requisitions, planning or MRP output | Why the purchase started and how urgent it was | Lead time pressure cannot be separated from preference |
| Specification and revision | PLM, engineering change records, drawing register | Which requirement the supplier had to meet | A cheaper part may look like a better choice when it failed the current revision |
| Alternatives considered | RFQs, quotations, bid tabulations, sourcing events | The options that were rejected and on what basis | Only winners remain, so the choice teaches nothing |
| Supplier approval status | Approved vendor list, supplier qualification records | Whether the supplier was eligible on that date | An unqualified supplier looks like an available option |
| Customer or contract constraints | Customer contracts, flow-down clauses, directed-buy instructions | Choices made because a customer required them | A directed buy looks like a poor commercial decision |
| Approval and authority | Approval workflow logs, delegation of authority matrix | Who could commit spend and who did | Policy exceptions cannot be told apart from normal buys |
| PO, change orders and receipts | ERP purchasing and receiving | What was ordered, changed and delivered, and when | Partial and late deliveries vanish from the record |
| Invoice match and quality outcome | Accounts payable match status, inspection, nonconformance reports | Whether the choice worked in practice | No outcome label for the decision |
The rights completeness checklist for each record source
Rights completeness means every record type in the packet has a known owner, a known set of contractual limits and a known privacy treatment. Access is not enough: being able to export a quotation from your own ERP does not mean the quoting supplier agreed to its reuse. Work through each source and counterparty, and do not mark the packet ready while any line is unknown.
Keep four questions separate for every row: can the company access it, does privacy law permit the intended use, who owns the content, and does any contract restrict sharing or licensing it. De-identifying a record answers part of the privacy question and none of the others.
| Item to confirm | Where to look | Ready when | Not ready when |
|---|---|---|---|
| Ownership of internal records | Records policy, system admin rights, prior acquisitions | The company created the records or acquired them with rights | Records came with an acquired business whose deal terms are unread |
| Supplier confidentiality terms | Supply agreements, NDAs, quotation and portal terms | Each supplier's terms are read and the treatment for its documents is recorded | Quotes and price lists are included before the terms are checked |
| Supplier-authored documents | Datasheets, drawings, certificates attached to RFQs | Third-party content is excluded or its license permits reuse | Attachments are bundled with the case by default |
| Customer constraints and flow-downs | Customer contracts, directed-buy letters, program terms | Customer-specific data is excluded, aggregated or cleared | Customer names or program details appear in sourcing notes |
| Personal data | Approver names, supplier contact details, buyer emails | Each personal field is mapped and its treatment chosen with privacy review | Contact names and emails ride along in free-text notes |
| Public sector or regulated buys | Government contract clauses, export control flags | Restricted records are identified and kept out | Controlled technical data sits inside specification attachments |
| Retention and legal holds | Retention schedule, hold notices | No record under a hold is offered | A dispute file is part of the sample |
The sample utility checklist: does the sample show decisions?
A sample is useful when a reviewer can follow a few cases end to end and see the choice, the constraint and the outcome. Pick cases by a stated rule, for example every tenth sourcing event in a quarter, rather than the cleanest ones, and note the rule in the packet, as the Datasheets for Datasets questions ask for the sampling strategy behind any sample.
Score each sampled case against the list below. Report the share of cases that pass each item; a case that fails several items is a sign the full export will too.
Checklist · 9 items
- The case has one stable identifier that joins requisition, RFQs, PO, receipts, bill and any quality record.
- At least one rejected alternative is present, with the reason or the comparison that excluded it.
- The specification revision in force on the order date is attached or referenced.
- Supplier approval status on that date can be read, not just today's status.
- Approvals show the approver role and date, and any exception to policy is marked.
- Change orders, partial receipts and late deliveries appear in time order.
- An outcome is recorded: matched and paid, disputed, returned or raised as a nonconformance.
- Free text is readable and in one language, with personal and customer details already flagged.
- Every field in the sample is defined in a data dictionary the buyer receives with the packet.
0 of 9 ticked. Tick each item as you go.
Which constraint makes the cheapest or fastest choice wrong?
The most instructive cases are those where the obvious option was not allowed. A packet that labels these constraints teaches a model why procurement is not simply price minimization, and it protects the company from a buyer misreading a compliant decision as a costly one.
Public procurement shows how formal this can be. The Federal Acquisition Regulation lists records normally kept in a contract file, including justifications and approvals, determinations and findings, and source selection documentation, and requires that bid, proposal and source selection information in those files be protected from unauthorized disclosure. Private companies keep looser versions of the same trail, and it is the part of the record that explains the choice.
- Specification lock: only parts qualified to the current revision could be bought, whatever the price.
- Approved vendor list: the cheaper supplier had not passed qualification or was on hold.
- Customer direction: a customer contract named the source, so purchasing had no choice.
- Lead time: the cheaper quote could not arrive before the need date in the requisition.
- Minimum order or tooling: the low unit price applied only above a volume the plan did not need.
- Sole or single source justification: a documented reason why competition was not possible.
- Quality history: an open corrective action against a supplier blocked new orders.
Illustrative packet: a contract manufacturer's component sourcing
The manufacturer here is hypothetical, and its volumes and sample results are illustrative figures chosen to show the checklist at work. A contract electronics manufacturer prepares five years of component sourcing history from its ERP, PLM and accounts payable systems. The record inventory lists about 30,000 sourcing events, 210,000 purchase order lines and 2,400 nonconformance reports, each joined by a sourcing event ID and part number.
The sample is every tenth sourcing event from one quarter, 40 cases in all. On the utility checklist, 36 cases join end to end, 31 include at least one rejected quotation, and 28 have a readable specification revision on the order date. The team reports these as shares of the sample and notes that most failures come from one plant that ran RFQs by email before a portal rollout.
On rights, the team reads the confidentiality terms of its top suppliers by spend and records a treatment for each: remove quoted prices, keep the ranking of quotes, and exclude attached datasheets. Approver names are replaced with role codes after privacy review.
The exception: one customer program used directed-buy instructions under a customer agreement that restricts disclosure of program information. Every sourcing event tied to that program is removed from the packet, and the packet states the exclusion and its approximate share of events. The checklist did its job by catching a constraint that made a whole relationship unlicensable rather than a single field.
Where the packet stops and counsel takes over
The packet records what the company knows about its rights; it does not decide them. Supplier quotations, price lists and sourcing strategies can be confidential business information, and under U.S. federal law information is a trade secret when its owner takes reasonable measures to keep it secret and it gains independent economic value from not being generally known. Whether a given supplier's data is protected, and by whom, is a question for counsel on the actual agreements.
Personal data needs the same discipline. The EU General Data Protection Regulation defines personal data as any information relating to an identified or identifiable natural person, which covers supplier contact names, buyer email addresses and approvers in workflow logs. A privacy reviewer should decide the treatment for each such field before a sample leaves the company.
Stop and narrow the scope when a contract is missing, a counterparty's terms cannot be found, or records under a legal hold are mixed in. A smaller packet with every line answered is more useful to a licensee than a larger one with open questions.
Taking a procurement packet into a licensing review
A packet like this is close to what a licensing review asks for. Procurement histories that a company decides to license can go through SourceX, which carries out the qualification, rights review, contracting, preparation and delivery steps. Before any of that, a short assessment of the company, with no upload, shows whether it qualifies.
No sourcing record is delivered until the company has approved the scope and terms and signed, and an ERP history too large to transfer comfortably can stay in the company's own storage or move on encrypted drives, since SourceX does not host it.
How should a supplier choice be graded when the cheapest quote lost?
Against the options that were feasible on the decision date, using only what the buyer knew then. The best supplier in hindsight is not a fair benchmark: it may have quoted after the award, failed a constraint the buyer could see, or looked good only because of a disruption nobody predicted.
U.S. federal acquisition rules show why a case must name its selection rule before it is graded. FAR 15.101-2 describes lowest price technically acceptable selection, where award goes to the lowest evaluated price among proposals that meet the stated acceptability standards. FAR 15.101-1 describes a tradeoff process that allows accepting other than the lowest priced proposal when the perceived benefits merit the additional cost and the rationale is documented, and FAR 15.308 requires the decision record to include the rationale for business judgments and tradeoffs, while not requiring the tradeoffs to be quantified. A private buyer's policy usually sits between the two. Record which rule applied, because a choice that is sound under a tradeoff policy fails under a lowest-price-acceptable one.
The Supplier Decision Tradeoff Matrix holds one row per option. Every figure below is illustrative and invented; it describes no real buyer or supplier.
| Option | Known on the decision date | Hard constraints, with evidence | Tradeoff against the chosen option | Observed later (never used to grade) |
|---|---|---|---|---|
| Supplier A | Lowest unit price; 9-week lead time | Fails: the requisition's need date was 6 weeks out | Infeasible, so not a comparison point | Not ordered |
| Supplier B (chosen) | Unit price 8 percent above A; 5-week lead time | Passes approved vendor list, current drawing revision and need date | Baseline | Delivered in 6 weeks; one lot raised a nonconformance |
| Supplier C | Unit price 3 percent above A; 5-week lead time | Fails: qualification audit still open on the approved vendor list | Infeasible on that date, even though it cleared later | Qualified two months after the award |
| Supplier D | Unit price 2 percent below B; 6-week lead time | Passes every hard constraint | Feasible alternative: cheaper, one week slower, inside the need date | Not ordered |
| Supplier E | Quote arrived after the award | Not assessable on the decision date | Excluded: a hindsight-only option | Kept for context only |
Which verdicts should the grading rubric allow, and where do outcomes go?
Four verdicts, signed by a procurement reviewer before anyone reads the outcome block: sound, defensible, unsound and not gradable, defined in the table below against the matrix above. The fourth keeps the dataset honest. A case missing its quotes, need date or constraint status on the decision date is reported as not gradable, never dropped and never guessed.
Outcomes go in fields marked as observed after the decision, out of the model's input and out of the verdict. Supplier B's nonconformance does not turn a defensible choice unsound, and an on-time delivery would not have rescued an unsound one; outcomes are labels for a different task, such as predicting late delivery. Anthropic's engineering post of January 9, 2026 treats a task as well built when two domain experts reach a matching pass or fail verdict without consulting each other, and a supplier-choice case should meet that test before it is scored. Have two reviewers grade each case without seeing each other's verdict, report counts per verdict with the disagreement count beside them, and send splits to adjudication instead of averaging them.
Record these fields so the tradeoff survives in the dataset without implying that the choice caused the outcome:
- Decision date and need date, from the requisition and award records.
- Every option with its quote timestamp, with late quotes flagged as hindsight-only.
- Each hard constraint with its status and the date of its evidence.
- The selection rule in force: lowest price acceptable, tradeoff or the company's own policy.
- The rationale as recorded, and whether it was quantified.
- The approver role and approval date.
- The outcome block, flagged as observed after the decision and kept out of model input.
| Verdict | When it applies | In the illustrative matrix |
|---|---|---|
| Sound | The chosen option passed every hard constraint and was best on the criteria the policy named | Would apply had Supplier B also been the cheapest feasible option |
| Defensible | It passed every hard constraint, a feasible alternative was better on price, and the recorded rationale names a criterion the policy allows | B over D, if the file records the one-week margin B gave against the need date |
| Unsound | It failed a hard constraint, or a feasible alternative was at least as good on every named criterion and no rationale was recorded | B over D, if no rationale was recorded and the policy names only price and need date |
| Not gradable | Quotes, need date or constraint status on the decision date cannot be reconstructed from records | Any case where Supplier C's audit status on the award date is unknown |
Who contributed each record in a cross-company purchase case, and what must be excluded?
Usually four or five organizations, and the licensing company can offer only what it authored or holds the right to share. List every artifact in one purchase case by contributor, record the confidentiality basis, choose a treatment, and log each exclusion with its reason so a licensee knows what is missing and why. The case continues the illustrative matrix; the treatments are examples for counsel to confirm, not a reading of any real agreement.
| Artifact | Contributed by | Confidentiality basis to check | Treatment (illustrative) | Exclusion log entry |
|---|---|---|---|---|
| RFQ, bid tabulation and award memo | Licensing company | Internal records policy | Keep; replace buyer names with role codes | None |
| Quotations from Suppliers B and D | Each quoting supplier | Quotation terms and any NDA signed for the RFQ | Keep lead time, rank and price relative to the chosen quote; drop the documents and unit prices | Quotation documents withheld under supplier terms |
| Customer drawing and revision history | End customer | Customer agreement and program terms | Keep the revision ID and the pass or fail result only | Drawing withheld under the customer agreement |
| Qualification audit on Supplier C | Licensing company, with supplier-provided attachments | Audit terms; supplier-authored content | Keep audit status and dates; drop findings and attachments | Audit findings withheld |
| Carrier delivery confirmation | Freight carrier | Carrier portal terms | Keep the delivery date only | Carrier documents withheld |
| Shared portal thread with supplier and customer | All parties | Mixed authorship and personal data | Exclude the thread | Entire thread excluded: mixed authorship |
Can several companies' purchase records go into one dataset?
Yes, if each contributor licenses only its own cases and the pooled pricing passes a competition review as well as a contract review. Keep a contributor ID on every record so one company's withdrawal or exclusion applies without touching the others, and keep a separate exclusion log per contributor.
Pricing is the sensitive part. In February 2023 the Justice Department's Antitrust Division withdrew three health care enforcement policy statements, saying they had become overly permissive on subjects such as information sharing. A dataset that would let one buyer or supplier read rivals' current prices is a question for competition counsel, so default to relative prices and data old enough not to reveal current terms until counsel has reviewed the design. Supplier contacts and approvers in these records are personal data and follow the privacy review already described above.
Sources for this section
- FAR 15.101-1 (GSA's FAR source, text through FAC 2025-06) describes a tradeoff process that allows accepting other than the lowest priced proposal when the perceived benefits merit the additional cost, with the rationale documented. Source
- FAR 15.101-2 describes lowest price technically acceptable selection: award on the lowest evaluated price among proposals meeting the acceptability standards. Source
- FAR 15.308 requires the source selection decision to be documented with the rationale for business judgments and tradeoffs, and says the documentation need not quantify the tradeoffs. Source
- On February 3, 2023 the Justice Department's Antitrust Division withdrew three health care enforcement policy statements, saying they were overly permissive on subjects such as information sharing. Source
- Anthropic's engineering post (January 9, 2026) says a good task is one where two domain experts would independently reach the same pass or fail verdict. Source
Frequently asked questions
Is a supplier packet the same as a data sample?
No. The sample is one part of it. The packet also carries the scope, the record inventory, the rights checklist and the exclusions, so a prospective licensee can judge both usefulness and rights before any wider access is discussed.
Can we include supplier quotations if we remove the prices?
Sometimes, but removing prices does not settle the question. Supplier terms may restrict disclosure of the quotation itself, its technical content or the fact of the relationship. Read each supplier's terms, record the treatment, and have counsel review the specific agreements.
How many cases should the sample include?
Enough to show the main purchase types and plants, chosen by a stated rule rather than picked for cleanliness. Dozens of complete cases usually reveal joining and documentation gaps; the buyer may then ask for a larger evaluation sample under its own terms.
Do approver names have to be removed?
Approver identity is personal data, so its treatment is a privacy decision for the specific use. Replacing names with role codes often keeps the authority signal a model needs while reducing exposure, but a privacy reviewer should confirm the approach.
What if our sourcing decisions were made in email?
Email approvals and quote comparisons can be recovered and linked to the case if they carry a PO, RFQ or part number. Treat the mailbox as another source in the rights checklist, because email also contains third-party and personal information.
Sources
- Odoo documents calls for tenders as alternative requests for quotation to several vendors whose product lines are compared before choosing. Source
- Odoo describes blanket orders as long-term purchase agreements with a vendor for recurring delivery at predetermined pricing. Source
- With 3-way matching in Odoo, vendor bills are only paid once some or all of the products on the purchase order have been received. Source
- The OCEL 2.0 specification models events related to several objects and supports object-to-object relationships and changing object attributes. Source
- FAR 4.803 lists justifications and approvals, determinations and findings, and source selection documentation among normal contract file contents, and requires protecting bid, proposal and source selection information. Source
- Under 18 U.S.C. 1839(3), information is a trade secret when its owner takes reasonable measures to keep it secret and it derives independent economic value from not being generally known. Source
- GDPR Article 4(1) defines personal data as any information relating to an identified or identifiable natural person. Source
- Datasheets for Datasets asks what sampling strategy was used when a dataset is a sample from a larger set. Source
See if your company qualifies
A short company assessment. No data uploads are needed.