Skip to content

AI uses for records

Who could say yes? Labeling a sales commitment evidence dataset

By SourceX Editorial · Reviewed by Noah Loul ·

8 sections

Short answer

Label every yes in a sales commitment evidence dataset by two facts the records can prove: the form the commitment took and whether the person giving it could bind the buyer. A champion's verbal yes, a budget owner's e-mail and a purchase order are three different labels. Seller systems rarely store the second fact: in ERPNext's sales and contract records, no field captures a customer signer's authority.

Agent run

Sales commitment evidence

This page covers who could say yes? Labeling a sales commitment evidence dataset.Illustrative

Key takeaways

  • A commitment label needs two coordinates, the form of the evidence and the authority of the person behind it; either one alone mislabels deals.
  • Paper is not the test: IFRS 15 accepts approval in writing, orally or by customary practice, provided the parties are committed to perform.
  • Authority is the stricter test: U.S. federal rules call an agreement made without authority an unauthorized commitment, which only a separate ratification can repair.
  • In ERPNext, 6 of 279 sales and contract fields hold customer commitment evidence, none is required, and none records whether the signer could approve the purchase.
  • Label each yes as of the day it happened; a later purchase order or payment is a separate event, not proof the earlier yes was binding.

What separates a verbal yes from procurement approval?

Two facts: the form the evidence takes and the authority of the person who gave it. A verbal yes is a statement, usually captured only in the seller's notes, from someone whose power to commit the buyer is unknown. Procurement approval is the buyer's own process completing: a purchase order issued, or an order form or contract executed by someone the buyer has authorized to sign.

Form matters less than many sales teams assume. Under IFRS 15, a contract exists for revenue purposes when the parties have approved it "in writing, orally or in accordance with other customary business practices" and are committed to perform their obligations (paragraph 9(a), as adopted in Commission Regulation (EU) 2016/1905). For sales of goods above a set price, the Uniform Commercial Code's statute of frauds generally asks for a writing signed by the party resisting enforcement or that party's agent or broker, and between merchants a written confirmation can count if the recipient does not object in writing within 10 days (UCC 2-201).

Authority is the stricter line, and public procurement law states it plainly. The U.S. Federal Acquisition Regulation calls an agreement an unauthorized commitment when it is not binding solely because the government representative who made it lacked the authority to enter into it (FAR 1.602-3). A sales commitment evidence dataset needs the same distinction. A model trained on a field that says the customer agreed learns to forecast, and act on, statements that may bind no one.

Where a seller's records keep the yes, and where they keep nobody's authority

Mostly in documents the seller creates, with the buyer's authority left out. We read the schemas and status code of four record types in ERPNext, the open-source ERP, at commit 821ca1d (October 27, 2026): Opportunity, Quotation, Sales Order and Contract. Together they hold 279 data fields.

Six of the 279 record the customer's side of the commitment: the customer's purchase order number and its date on the Sales Order, and on the Contract a Signed check box, a free-text Signee name, a signing time and an IP address. The schema requires none of them, and the code demands a signee and time only once someone ticks Signed. No field in the four records says whether the customer's signer could approve the purchase. The nearest is a free-text Job Title on the Opportunity, and a title is not authority. The Contract record is lopsided too: on submission it stores the company's signer as the logged-in user account, while the customer's signer is whatever name someone types.

Status moves without any of it. ERPNext sets an Opportunity to Converted when a submitted Quotation linked to it reaches Ordered or Partially Ordered, and a Quotation reaches Ordered when submitted Sales Order lines cover its items, all documents the seller's users create.

Commercial systems vary. Salesforce's standard Order object, for instance, has CustomerAuthorizedById and CustomerAuthorizedDate fields beside PoNumber and PoDate, a slot for which customer contact authorized the order and when. A slot holds a name and a date; it does not show that the contact could commit the buyer. That proof usually sits outside the CRM, in the purchase order, a signature record or an approval e-mail.

Where a seller's records keep the yes, and where they keep nobody's authority
ERPNext record (commit 821ca1d)Data fieldsCustomer commitment evidenceRequired by schemaField for signer authority
Opportunity52NoneNot applicableNone; free-text Job Title only
Quotation89NoneNot applicableNone
Sales Order116Customer's PO number and PO dateNoNone
Contract22Signed, Signee, Signed On, IP AddressNo; signee and time required once Signed is tickedNone

Seven commitment labels, from interest to performance

Assign the highest label the evidence supported on the day it was recorded, and keep authority in its own field so a reviewer can see why a label stopped short. The labels are ordered, but real deals skip steps and move backward.

Three modifiers sit beside a label instead of replacing it: withdrawn, when the buyer retracts a yes; superseded, when a later commitment replaces it; and ratified, when someone with authority later approves a commitment the first speaker could not make. Ratification never upgrades the original label. It is a new event with its own date and its own evidence.

Seven commitment labels, from interest to performance
LabelObservable evidenceAuthority testCommon false positive
C0 InterestPositive language; no commitment asked for or givenNoneA demo request logged as intent to buy
C1 Verbal yesA spoken commitment captured only in seller notes or a call summarySpeaker's role noted; authority unverifiedA next-step field reading they're in after a friendly call
C2 Written yesE-mail or chat from the buyer stating a decision to proceedSender's role noted; authority unverifiedA user with no budget forwarding internal enthusiasm
C3 Budget approvalWritten approval from whoever controls the budget for this purchaseBudget ownership shown in the buyer's own words or documentsA senior title taken as budget ownership
C4 Procurement under wayA buyer-side artifact: requisition, vendor onboarding, security or legal reviewThe buyer's process, not a person, is the evidenceA completed vendor form read as approval
C5 Binding commitmentPurchase order issued, or order form or contract executed by the buyerIssuer or signer authorized under the buyer's own rulesA signature from someone outside the buyer's signing authority
C6 PerformedPayment received or delivery accepted under that commitmentNot needed once performance is recordedA payment on an older order matched to the new one
A decision tree that labels a buyer's yes: no stated decision gives C0 Interest, a decision only in seller notes gives C1 Verbal yes, a buyer record without a purchase order or signed order gives C2 to C4, and a purchase order or signed order becomes C5 Binding only if the signer's authority is shown.
Labeling one commitment by evidence and authority. A framework, not measured data.

The evidence map: what to keep, where it lives, what to drop

Each label rests on records from both sides of the deal, and most buyer-side records reach the seller only as attachments or e-mail. Keep the evidence and the role of the person behind it; drop the identity unless a task genuinely needs it.

Replacing names with role codes (operations lead, finance controller, procurement, signatory) keeps everything the labels need. The model has to learn which roles commit which purchases, not who the people were, and a dataset that never carries contact identities is easier to license and to share with reviewers.

The evidence map: what to keep, where it lives, what to drop
Evidence itemUsually found inSupportsKeepRemove or replace
Call notes and meeting summariesCRM activities, call recorder summariesC0 to C1Date, speaker role, the commitment sentenceNames, phone numbers, unrelated talk
Buyer e-mail and chatSynced mailboxes, shared channelsC2 to C3Date, sender role, decision wording, any stated limitNames, addresses, signature blocks
Procurement artifactsSupplier portals, onboarding and security questionnairesC4Artifact type, dates, buyer reference as a tokenPortal logins, bank details
Purchase orders and executed order formsE-mail attachments, e-signature platforms, ERP order recordsC5Issue or signing date, signer role, document version, scopeSigner name, signature image, IP address
Payment and acceptance recordsAccounting and billing systemsC6Date, link to the order, full or partialAccount numbers, invoice addresses
Authority evidenceDelegation letters, buyer policy excerpts, PO terms naming who may orderAuthority field for C3 and C5Role, limit and the dates the authority ranNames of individual approvers

How to test whether the person who said yes could bind the buyer

Treat authority as a dated fact with its own evidence, and accept only evidence the buyer produced. A title typed into the CRM, a confident tone on a call or a rep's belief that someone runs the budget are claims about authority, not proof of it. Apply five tests and record which ones passed.

At a glance

How to test whether the person who said yes could bind the buyer

Checklist · 5 items

  1. Source: the buyer created the evidence of authority, such as a purchase order, a signature under the buyer's signing block, a delegation letter or a policy excerpt the buyer shared; the seller did not type it.
  2. Scope: the authority covers this kind and size of purchase. Someone who approves software renewals may have no authority over a new services contract.
  3. Date: the authority was in force on the day of the commitment, because delegations start, end and change.
  4. Process: the buyer's own required steps were complete. Public procurement makes these steps visible; under Directive 2007/66/EC, an EU contracting authority may not conclude a contract until at least 10 calendar days after sending its award decision by fax or electronic means, so even a formal decision to award is not yet a contract.
  5. Ratification: when someone without authority committed and an authorized person approved it later, record two events. FAR 1.602-3 lists seven conditions before an official may ratify an unauthorized commitment, among them that the ratifying official has authority to contract and that funds were available when the commitment was made.

0 of 5 ticked. Tick each item as you go.

Illustrative: one renewal, three yeses and one purchase order

Illustrative case, not a real company or deal: a software vendor renewing a long-standing customer and offering seats for a second team.

Had the CRM marked the whole expansion won on day 6, it would have counted seats nobody bound the buyer to buy. Labeled per commitment and per scope, as in the timeline that follows, the same history shows what a model should learn: the operations lead's yes predicted the renewal, not the expansion, which procurement left out.

  • Day 0: on a call, the customer's operations lead says they are renewing and adding the new team. Label C1 for both scopes; authority unverified.
  • Day 6: the IT director e-mails that the purchase is approved on their side and asks for paperwork. Label C2; the e-mail mentions no budget and the director's limit is unknown.
  • Day 15: the finance controller confirms in writing that the renewal is in budget and the expansion is not yet. Renewal C3, expansion still C2.
  • Day 21: the buyer's supplier portal opens a vendor update and a security review. Renewal C4.
  • Day 38: procurement issues a purchase order for the renewal only. Renewal C5 with the PO as authority evidence; expansion stays at C2 and is later marked withdrawn.
  • Day 52: the renewal invoice is paid. Renewal C6.

Does a deal that closed prove the early yes was binding?

No, and the dataset should not be built as if it did. A deal can close for reasons that have little to do with an early yes: a rival drops out, budget appears at the end of a quarter, a new executive arrives. A deal can also close despite a weak step along the way, such as a commitment by someone who lacked authority that was later put right.

Federal procurement treats that repair as a separate act. Under FAR 1.602-3, ratification is the approval of an unauthorized commitment by an official who has the authority, under listed conditions, and nothing in the rule treats the later approval as proof that the first agreement was valid. Label each yes as of the day it was made, store the outcome in its own field, and never back-fill a C5 because a C6 arrived.

The strongest objection is practical: authority evidence is often missing, so strict labels leave many deals between C1 and C3. That is the information the label set exists to keep. A model trained on history that separates a verbal yes from a binding commitment can learn how often each turns into performance; one trained on blended labels learns the sales team's optimism.

Writing a data request for sales commitment evidence

Ask for commitment events, not deal snapshots. A useful request names the labels wanted, the evidence each must carry, the authority field and its tests, the identity fields to remove, the period, and an acceptance test: for a random sample of commitment events, a reviewer can see the evidence behind each label and its authority status without learning who anyone was.

Ask for the negatives too: yeses that were withdrawn, purchase orders smaller than the verbal scope, and commitments ratified after the fact. Without them a model never sees the gap this kind of dataset exists to teach. And keep scope per commitment, because one deal can hold a binding renewal and an expansion nobody approved.

SourceX works from a buyer's specification instead of an inventory. Through its request form a buyer sets out what the records must cover, the volume, language and format, the timing and the license terms, and SourceX then sources matching records from businesses willing to license them. A commitment-evidence request should add the label set and the acceptance test, so candidate records can be checked against both.

Methods for the ERPNext figures: the script erpnext_commitment_fields.py reads the doctype JSON and Python files of a sparse checkout at commit 821ca1d. A data field is any schema entry that is not a layout element (section, column or tab break, HTML, fold, heading or button). The six evidence fields are a fixed list named in the script; authority fields are those whose name or label contains the strings authori, designation, signatory, delegat or job title, or the word role; only Job Title matched. Not measured: how any company configures or uses ERPNext, custom fields, or workflows added on top.

Frequently asked questions

Is a verbal agreement a binding commitment in B2B sales?

Sometimes, depending on the law that applies, what is being sold and how the parties usually deal. IFRS 15 accepts oral approval for accounting when the parties are committed to perform, while the UCC generally wants a signed writing for goods above a set price. A dataset does not decide enforceability; it records the evidence and the authority behind each yes, and counsel decides the legal question.

How is a purchase order different from a signed contract as evidence?

A purchase order is the buyer's own instruction, issued through its procurement process, so it carries the buyer's authority by design. A signed contract proves commitment only if the signer could bind the buyer. Both can reach label C5; record which one you have, its date and the scope it covers.

How can a seller tell whether a buyer contact has signing authority?

Only from evidence the buyer produced: a purchase order, a delegation letter, a policy excerpt or a signature under the buyer's signing block. A job title, an org chart or a confident statement on a call is not enough. When no such evidence exists, keep the authority field unverified rather than inferring it.

Should a sales dataset keep the names of people who approved deals?

Rarely. The labels need each person's role, the date and the evidence, not the name. Replace names with role codes, drop phone numbers, signature images and IP addresses, and tokenize buyer reference numbers; that keeps the commitment pattern while removing contact identities the task does not use.

Can a model learn which verbal commitments turn into orders?

It can be trained and tested on it if the history keeps commitment events separate from outcomes, with authority recorded at the time. Outcome-blind labels plus a separate performance field let you measure how often each label became a purchase order or payment. Blended won or lost labels cannot show that.

Sources

  • IFRS 15 paragraph 9(a), as adopted in the EU, requires that the parties have approved the contract in writing, orally or by customary business practice and are committed to perform. Source
  • UCC 2-201 generally requires a writing signed by the party resisting enforcement or its agent or broker for sales of goods above a set price, and lets an unanswered merchant confirmation count after 10 days. Source
  • FAR 1.602-3 defines an unauthorized commitment and ratification, and lists seven conditions for ratifying one. Source
  • Directive 2007/66/EC bars concluding a public contract until a standstill period of at least 10 calendar days after an award decision sent by fax or electronically. Source
  • ERPNext's Sales Order schema has optional Customer's Purchase Order and Customer's Purchase Order Date fields. Source
  • ERPNext's Contract records a Signed flag, a free-text Signee, Signed On and IP Address, and sets the company signer to the logged-in user at submission. Source
  • ERPNext sets an Opportunity to Converted when a linked submitted Quotation is Ordered or Partially Ordered. Source
  • ERPNext marks a Quotation Ordered when submitted Sales Order items cover its items. Source
  • Salesforce's Order field reference lists CustomerAuthorizedById, CustomerAuthorizedDate, PoNumber and PoDate. Source

Tell us what data you need

Share scope, volume, format, timing and licensing needs. Datasets are sourced on request.

See if your company qualifies