Systems and records
Sales procurement dependency data: what blocks a deal that is ready
By SourceX Editorial · Reviewed by Noah Loul ·
9 sections
Short answer
Represent procurement dependencies as their own dated records linked to the opportunity, not as a stage or a line in the next-step field. Each dependency names its type, the customer and seller owners, the artifact that closes it, its state and the evidence of completion. The correct next action then follows from which dependencies are still open, and sales procurement dependency data can be graded on it.
Checklist
Sales procurement dependency
Key takeaways
- Store each procurement dependency as its own dated record with owners, required artifact, state history and completion evidence.
- Stage and next-step fields hold one value each, so they hide which approval is open and lose the order things cleared.
- Vendor registration, security review, a DPA and the purchase order run on separate clocks with separate owners.
- Evaluate an agent on the next action it picks and the record changes it makes, rebuilt as of the task date.
- Grade against signed documents and later outcomes, which often contradict the CRM's summary.
How do you represent procurement dependencies blocking an otherwise ready deal?
Give every dependency its own record with a type, an owner on each side, the artifact the customer needs, a state with dated transitions and a link to the evidence that closed it. Link those records to the opportunity and leave the sales stage where the buying decision actually is. A deal can be commercially agreed and still blocked by vendor registration, a security review and a data processing agreement, and those are three separate clocks.
With dependencies stored this way, the next action is computed rather than typed: it is the step that unblocks the earliest open dependency on the critical path, assigned to whoever owns it. That makes the record useful twice, once for the revenue team running the deal and again as evaluation data for an AI agent that must pick the right next move.
The rest of this page sets out the record, the evaluation cases built from it, and the contradictions between the CRM and the signed documents that the evaluation has to catch.
Why the stage and next-step fields cannot carry this
Standard CRM fields hold one stage, one expected close date and one free-text next step. Salesforce's Data Cloud model describes Next Step as a description of the next task needed to close an opportunity, a single text value. Odoo's open-source CRM defines its Expected Closing field as an estimate of the date on which the opportunity will be won, and records the date of the last stage change. None of these can say that three independent approvals are in flight with different owners.
History is the other gap. Salesforce stores tracked field changes in history objects such as OpportunityFieldHistory, and its documentation says field history is retained for up to 18 months, or 24 months through the API, unless Field Audit Trail is in place. A team that relies on overwritten next-step text loses the sequence of dependencies after that window, which is exactly the sequence an evaluation needs.
The practical result is a pipeline full of deals marked as negotiation or commit with next-step text such as waiting on procurement. That text is true and useless: it hides which approval is open, who owns it and what would close it.
Which approvals usually hold a deal that is otherwise agreed?
Business-to-business deals that are agreed but not signed often wait on some mix of the dependencies below. The requirement behind each is set by the customer's policies and, in some cases, by law; the record should store the requirement as the customer stated it, with the date.
- IRS requester instructions describe Form W-9 as the way a person required to file information returns obtains a payee's correct name and taxpayer identification number.
- FAR 4.1102 requires offerors and quoters to be registered in SAM when an offer or quotation is submitted, with listed exceptions.
- The Cloud Security Alliance describes its CAIQ as a set of yes-or-no questions mapped to Cloud Controls Matrix controls, used as a self-assessment rather than a certification.
- GDPR Article 28 requires processing by a processor to be governed by a written contract or other legal act that sets out the subject matter, duration, nature and purpose of processing, the types of personal data and the categories of data subjects.
| Dependency | What the customer needs | Artifact that closes it | Evidence of completion |
|---|---|---|---|
| Vendor registration | The seller set up as a payee, with tax and banking details verified | Completed supplier form; for U.S. payees often a Form W-9 | Supplier ID issued by the customer's procurement or payables system |
| Federal registration (government buyers) | Registration in the System for Award Management before offering | Active SAM registration | Registration record matching the legal entity on the offer |
| Security review | Answers about the seller's controls, often on a standard questionnaire | Completed questionnaire, such as a CAIQ, plus requested documents | Written approval, or approval with listed conditions and dates |
| Data processing agreement | A written contract governing processing of personal data | Signed DPA with the required content | Executed DPA version and signature date |
| Purchase order | Internal budget approval and an order issued to the seller | PO referencing the agreed quote or contract | PO number, amount basis and issue date |
The dependency record: fields and rules
This is the first half of the dependency and next action evaluation specification. Every field is filled from a dated system record or document, never from memory, so that the state of the deal can be rebuilt as of any past date.
Most of the evidence already exists outside the CRM. Procurement portals send dated notifications when a form is submitted or a supplier ID is issued, security teams return approval letters or findings by email, and signed agreements sit in an e-signature or contract system with their completion dates. The work is linking each of those to a dependency record when it happens, rather than summarizing them later in a next-step line.
Keep the state list short and fixed. Seven states are enough for the cases here, and a fixed list lets two reviewers label the same history the same way. Free-text states such as almost done or chasing are what make old pipelines impossible to evaluate.
| Field | What it holds | Rule | Typical source |
|---|---|---|---|
| Dependency ID and type | Unique ID and one type from a fixed list | One record per dependency, never one per deal | CRM custom object or linked task |
| Opportunity and legal entities | Opportunity ID, the customer entity that buys and the seller entity that contracts | Entities must match the documents, not the account name | CRM account and contract records |
| Owners | Named customer-side owner and seller-side owner | Both required; unknown is a state, not a blank | Email threads, procurement portal |
| Required artifact and version | The form, questionnaire or contract version requested | Store the version the customer asked for | Customer request, portal download |
| State history | Requested, submitted, in review, conditionally approved, approved, rejected, withdrawn, each dated | Append only; no overwriting | CRM history, portal notifications |
| Completion evidence | Pointer to the approval, supplier ID, signed document or PO | No approved state without evidence | Document store, email, ERP |
| Conditions and deadlines | Remediation items or terms attached to an approval | Conditions keep the dependency open until met or waived | Approval letter, security findings |
| Blocks | Which later step it blocks: signature, PO, go-live or invoice | Required, so the critical path is explicit | Customer's stated process |
Evaluation cases: picking the right next action
The second half of the specification turns dependency records into tasks. Each task gives an agent the opportunity, its dependency records as of a date and the latest messages, and asks for the next action. The grader checks the action against the answer key and checks that nothing in the records was changed without evidence.
Answer keys come from two reviewers working from the dependency records and the customer's stated process, with disagreements settled by the revenue operations owner. A case where reviewers cannot agree is dropped or rewritten; an ambiguous key teaches the agent nothing and inflates or deflates its score at random. Include cases where the right action is to wait, so an agent cannot score well by always sending another reminder.
| Starting state | Correct next action | Grader check | Tempting wrong action |
|---|---|---|---|
| Security review approved, DPA unsigned, PO not issued | Chase the DPA signature with the named customer owner | Action targets the DPA owner; stage unchanged | Move the deal to closed won |
| Vendor form submitted, no supplier ID after the customer's stated timeline | Ask the procurement owner for status, citing the submission date | Message cites the dated submission | Resubmit the form unprompted |
| Security approval with conditions due before go-live | Schedule the remediation items and track them as open | Conditions recorded as open with dates | Mark security as complete |
| Customer requested a newer questionnaire version mid-review | Answer the new version; keep the old answers as history | Artifact version updated; history kept | Send the old answers again |
| DPA signed by an entity that differs from the buyer on the order form | Flag the mismatch to legal before signature | Escalation recorded; no signature request sent | Proceed because a DPA exists |
| All dependencies approved, PO issued | Request signature or confirm order per the agreed process | Next action is signature or booking | Wait for further procurement steps |
| Government buyer, seller entity not active in SAM | Fix the registration before submitting the offer | Offer not submitted; registration task created | Submit the offer and register later |
Where the documents and the CRM disagree about the deal
Grade against the documents, not the CRM's summary. The CRM may say closed won while the purchase order has not been issued, or say security approved when the approval letter lists conditions that were never met. Each contradiction is a test the agent must notice.
Common contradictions are worth building into the answer key on purpose: an executed DPA whose version differs from the one in the CRM attachment, a supplier ID issued to a different legal entity than the one on the contract, a signed order whose start date falls before the security approval, and a forecast category that moved to commit on a date when two dependencies were still open. The later outcome settles the label: whether the deal was signed, invoiced and paid, or slipped and why.
Keep the CRM's own claims in the task as they were at the time, because the agent will meet the same misleading fields in production. The point of the case is whether the agent trusts a stage label over a dated document, and the only way to test that is to leave the label wrong.
Worked example: one deal, three open dependencies (illustrative)
Every figure in this example is an illustrative duration or count. A software vendor and a mid-sized customer agree on scope and price. The champion confirms the budget. The CRM moves the deal to commit with the next step waiting on procurement.
The dependency records tell a more useful story. Vendor registration was requested on day 1 and submitted on day 3, with no supplier ID yet. The security questionnaire was requested on day 2; on day 9 the customer replaced it with a newer version, which reset the review. The DPA draft was sent on day 4 and returned with redlines on day 12. On day 14, the right next action is not a reminder to the champion; it is answering the new questionnaire, because the customer's process puts security approval before both PO and signature.
The exception: on day 16 the customer's legal team signs the DPA using a subsidiary's name, while the order form names the parent. An agent that reads only the DPA state sees approved and moves on. The correct action is to stop and route the mismatch to both legal teams, because the entity that pays, the entity that signs and the entity that processes data have to line up.
Scoring rules and the limits of the specification
- Rebuild each task as of its date: only records and messages that existed then are visible, so later approvals cannot leak into the prompt.
- Score the next action and the record changes together; a correct message that also marks an open dependency approved fails.
- Repeat each task several times and report how often the agent gets every run right, not just the average.
- The specification evaluates sequencing and evidence handling; it does not predict whether a deal will close or forecast revenue.
- Legal sufficiency of a DPA or registration is a counsel question; tasks that hinge on it carry a reviewer label, not an automated answer.
Deal histories with dependency records as licensable data
Opportunity histories that keep dated dependency records, approvals and outcomes describe how business purchases actually move, which is hard to reconstruct from stage fields alone. Revenue teams with that kind of history can put selected deals forward for licensing, and SourceX then carries the rights review, contracting, data preparation, delivery and payment.
Customer names, deal terms and contract text in these records are confidential and need preparation before they leave the company. The company signs off on scope and terms before any record leaves, and qualification is checked up front in a brief assessment that needs nothing uploaded.
Frequently asked questions
Should procurement dependencies be a sales stage?
Usually not. A stage describes where the buying decision is, while dependencies run in parallel with their own owners and dates. Putting them in the stage field hides which approval is open and erases the order in which they cleared.
What evidence shows a procurement dependency is really closed?
A dated record from the customer's side: a supplier ID, an approval letter, an executed agreement or an issued purchase order. A rep's note that procurement is done is not evidence. Conditional approvals stay open until their conditions are met or waived.
How long does CRM field history keep next-step changes?
It depends on the system and configuration. Salesforce documents retention of tracked field history for up to 18 months, or 24 months through the API, unless Field Audit Trail is enabled. Export the history or keep dependencies as separate append-only records if the sequence matters.
Does every deal need a data processing agreement?
No. Under GDPR, a written contract is required when one party processes personal data on behalf of another, and other laws and customer policies set their own rules. Whether a deal needs one is a question for counsel and the customer's privacy team.
Can an AI agent decide the next action from CRM data alone?
Only if the dependency records are in the CRM with dates, owners and evidence. With stage and free-text next step alone, the agent cannot tell which approval blocks the deal. The evaluation cases on this page test exactly that difference.
Sources
- Odoo's CRM defines an Expected Closing date field as an estimate of the date on which the opportunity will be won, and records the date of the last stage update. Source
- Salesforce's Data Cloud opportunity model describes Next Step as a description of the next task needed to close an opportunity. Source
- Salesforce stores tracked field history in history objects and retains it for up to 18 months, or 24 months via the API, unless Field Audit Trail is enabled. Source
- The IRS requester instructions say Form W-9 is used by persons required to file information returns to obtain the payee's correct name and TIN. Source
- FAR 4.1102 requires offerors and quoters to be registered in SAM when an offer or quotation is submitted, with listed exceptions. Source
- The Cloud Security Alliance describes the CAIQ as yes-or-no questions mapped to Cloud Controls Matrix controls, used for self-assessment and not itself a certification. Source
- GDPR Article 28 requires processing by a processor to be governed by a written contract or other legal act setting out the subject matter, duration, nature and purpose of processing, the types of personal data and categories of data subjects. Source
See if your company qualifies
A short company assessment. No data uploads are needed.