Skip to content

AI data market

B2B sales histories AI data investment: judgment data, not a lead list

By SourceX Editorial · Reviewed by Noah Loul ·

9 sections

Short answer

B2B sales histories are an investable AI data source only when stage changes, approvals and won or lost outcomes survive with every contact identity removed. Default CRM schemas make that harder than it sounds: in our count of Odoo's lead model, 20 of 54 stored fields identify or locate a person or company, while 5 record the outcome. Back suppliers whose records describe decisions, not people.

Record timeline

B2B sales histories AI data

This page covers B2B sales histories AI data investment: judgment data, not a lead list.Illustrative

Key takeaways

  • Commercial judgment data is the record of what a sales team decided and how the deal ended; the contact directory around it is a different, regulated product.
  • In Odoo's default lead model, 20 of 54 stored fields identify or locate a person or company, 5 record the outcome, and 13 of the 22 fields whose changes are logged are identity fields.
  • SuiteCRM's opportunity table stores no person identity at all, which shows that the cost of removing identity depends on the source schema.
  • Why a deal was lost is the scarcest label: optional in Odoo, absent from SuiteCRM's opportunity record.
  • A CRM benchmark built on generated Salesforce orgs found leading agents near 58% on single-turn tasks and 35% in multi-turn ones, which signals both demand and a synthetic substitute.

What is commercial judgment, and can sales records teach it?

Commercial judgment is the set of choices a sales team makes under uncertainty: qualify or drop a lead, push or hold a close date, approve a discount, escalate to an executive, walk away. Sales histories can teach it when they keep each choice, its timing and the deal's eventual outcome, because that turns a pipeline into labeled decisions a model can be trained or graded on.

The same records also hold something an AI developer does not need for that purpose: who the buyers were and how to reach them. A dataset built around names, emails and phone numbers is a prospecting list, sold to people who want to contact those buyers. A dataset built around stage changes, approvals and outcomes is judgment data. The two can come from one CRM, but they answer different buyer tasks and sit under different rules.

The B2B sales histories AI data investment case therefore turns on one test: how much of the decision record survives once every identity is stripped out. This article runs that test on two open-source CRM schemas, sets out where the legal line runs, rates five segments of sales history and lists what could make the thesis wrong.

Who wants to grade commercial judgment, and with what data?

Agent developers do, and the most detailed public benchmark had to generate its own CRM data to do it. CRMArena-Pro, posted to arXiv in May 2025, defines 19 expert-validated tasks across sales, service and configure-price-quote work in business-to-business and business-to-consumer settings. Leading agents reached about 58% success on single-turn tasks and about 35% when the interaction ran over several turns, and the authors report almost no built-in awareness of confidentiality.

The environment behind those numbers is synthetic. The benchmark populates 25 interconnected Salesforce objects with records generated using gpt-4o, about 29,101 in its business-to-business org and 54,569 in the business-to-consumer one. That is a demand signal and a warning at once: developers want to grade commercial judgment, multi-turn work is where agents fail, and a benchmark team could still build its test org from generated records rather than a licensed pipeline.

For a data business, the opening lies in what generated orgs cannot supply: real reasons, real counterparty behavior and outcomes that took months to arrive. Volume is the part a generator makes cheaply.

How much of a CRM deal record is about people?

In Odoo's open-source CRM, more of the deal record is about people than about the outcome. We counted the fields on Odoo's lead and opportunity model, crm.lead, on the 20.0 branch (commit ee70793, October 27, 2026). The model declares 67 fields, 54 of them stored. By our grouping, 20 stored fields identify or locate a person or company, including contact name, company name, job position, email, phone, website and street address, while 5 record the outcome: stage, won or lost status, lost reason, closed date and the active flag that Odoo clears when a deal is marked lost.

The change log leans the same way. Odoo records each change to a field marked for tracking in the record's message history, and 13 of the 22 tracked fields on crm.lead are identity fields; the other 9 are salesperson, sales team, active flag, type, stage, expected revenue, recurring revenue, won or lost status and lost reason. Two details raise preparation cost further. The lost reason is optional, with three default choices: too expensive, no people or skills, and not enough stock. And when a deal has a linked contact but no title, Odoo names it after that contact, so identity leaks into a free-text field.

SuiteCRM, which descends from SugarCRM Community Edition, shows the alternative. Its Opportunity definition (hotfix branch, commit be52770) declares 32 fields, 11 stored, and none holds a person's name, email, phone or address; people are joined through a separate table that records each contact's role on the deal. Six fields are audited, including sales stage, probability and close date, and the ten sales stages end in Closed Won or Closed Lost. There is no field for why a deal was lost.

The table below sets the two designs side by side:

How much of a CRM deal record is about people?
What the schema storesOdoo 20.0, crm.leadSuiteCRM, OpportunityWhy an investor should care
Fields on the deal model67 declared, 54 stored32 declared, 11 storedHow much must be reviewed field by field before licensing
Stored fields that identify or locate a person or company200; contacts joined through a role tableIdentity sits on the deal row in one design and beside it in the other
Stored fields that record the outcome5: stage, won or lost status, lost reason, closed date, active flag2: sales stage and expected or actual close dateWon or lost is cheap to label; the rest is not
Fields whose changes are logged22 tracked, 13 of them identity6 audited, 1 of them the customer account linkA change history can carry more identity than decisions
Structured reason for a lossOptional field with 3 default choicesNone on the opportunity recordThe label judgment models need most is the least captured
A treemap of the 54 stored fields on Odoo's lead model: identity 20, deal context 19, timing 6, outcome 5, free text 2 and system 2, beside notes that 13 of 22 change-tracked fields are identity and that SuiteCRM's opportunity row stores none of 11.
What a default CRM deal record stores. Source: Odoo 20.0 and SuiteCRM source code, SourceX count, October 2026; field groups are editorial.

Where the line runs between judgment data and prospecting

The line runs through the contact fields: a sales history that carries named buyers holds personal data under the UK GDPR and California's CCPA, and selling it can start to look like data brokering. Strip those fields and what remains is commercial context and outcomes, which is what a judgment dataset needs anyway.

Business contact details get no blanket exemption. The UK Information Commissioner's Office says the UK GDPR still applies to business-to-business marketing whenever personal data is processed, for example the name of the individual who represents a business. In California, the CCPA's exemption for business-to-business contact data lapsed at the end of 2022, and the California Privacy Protection Agency's FAQ now lists contacts at business customers and vendors among the people the law covers.

Records sold as a contact list also run into data broker rules. Under California's Delete Act, beginning August 1, 2026, registered data brokers must check the state's deletion platform at least once every 45 days and process the deletion requests it holds. Whether a given dataset falls under that regime is a question for counsel, but a supplier whose value depends on identities carries that exposure, and one whose value survives de-identification largely does not.

Five segments of sales history, rated for an investor

Stage histories and forecast changes rate best, because the outcome arrives on its own and identity can be removed without losing the decision; loss reasons rate worst on capture despite being the most valuable label. CRMs already keep the raw material: Salesforce's documentation describes an OpportunityHistory object holding an opportunity's stage history and an OpportunityFieldHistory object holding changes to its field values.

Each segment is rated on how firmly the outcome is recorded, how exposed the records are to identity, whether a public or synthetic substitute exists, and the gap that would most change the rating. The ratings are editorial, built from the sources in this article as of October 27, 2026, and rate segments, not companies.

Five segments of sales history, rated for an investor
Segment and buyer taskOutcome labelIdentity exposureSubstituteEvidence gap to close
Stage histories: judge whether a deal is real and what moves itStrong for won or lost; weak for whyHigh in lead-style schemas, low where contacts sit in a separate tableMedium: generated CRM orgs already back public benchmarksShare of closed deals with a complete stage history and a structured loss reason
Forecast and close-date changes: call the quarter honestlyStrong in hindsight: forecast against actual closeLow once account names are replacedLow: none foundWeekly field history of amount and close date, not only the final values
Quote and discount approvals: approve, counter or escalateMedium: the approval is logged; the later margin often is notLow to moderate: mostly products and pricesMedium: generated quote workflows existApproval reason linked to whether the deal closed and renewed
Loss reasons and win-loss reviews: learn why deals failWeak: optional or absent in default schemasHigh: free text names buyers and competitorsLow: none foundShare of losses with a reason captured at the time, not months later
Renewal and expansion decisions: keep, grow or let go an accountStrong: renewal or churn is a contract eventModerate: account names and usage contactsLow: none foundLink from account activity to the renewal outcome and its date

Does a larger CRM mean more licensable judgment data?

Not by itself. Sector size and record volume overstate usable supply here for three reasons: much of a CRM is leads that never became deals, much of each deal record is identity, and the most valuable label is often missing.

Volume is also the part synthetic generation handles best. CRMArena-Pro's generated business-to-business org holds about 29,101 records, enough to test workflows, and it was generated rather than licensed. What a generator cannot produce is the reason a real buyer went quiet in week six, or a forecast that slipped three times before a deal closed. A supplier with a modest CRM but structured loss reasons and complete stage histories may hold more licensable judgment data than one with millions of contact rows.

Schema design changes the preparation bill too. In a lead-style design such as Odoo's, identity has to be removed column by column, and then again from titles and notes; in a split design such as SuiteCRM's, much of it can be left behind by not exporting the contact tables. Neither removes the need to review free text.

What could make the sales-history thesis wrong

Five developments could weaken the case for licensed sales histories. Each is observable.

  • Synthetic orgs prove good enough. If benchmarks built on generated CRM data, such as CRMArena-Pro's, keep predicting real-world agent performance, the premium for real pipelines shrinks.
  • Reasons stay unstructured. If loss reasons remain optional fields or free text, preparation cost per labeled deal may exceed what buyers will pay.
  • Privacy rules reach derived datasets. Regulators treating de-identified sales histories as broker data would add deletion and registration duties that judgment datasets were designed to avoid.
  • Pricing data proves too sensitive. Discount and quote histories reveal how a company prices, and suppliers or their customers may refuse to license them at any price.
  • Agents learn judgment elsewhere. If agents acquire commercial judgment from general reasoning and their users' own CRMs at run time, demand for licensed histories narrows to evaluation.

Questions for a sales-history supplier, SourceX included

Ask suppliers to show what survives de-identification, measured with your definitions, before discussing volume.

Apply the list to SourceX as well. SourceX prepares business records that a company decides to license as AI datasets and handles the full path itself: it sources and qualifies the supplier, reviews rights, contracts, prepares and delivers the data, and pays the company. The supplying company signs off on scope and terms, and signs, before any data reaches SourceX. Its supply and deal figures are not public; SourceX provides company and commercial metrics to investors on request rather than in articles. The company is raising under Rule 506(c) from accredited investors only, and its investor materials can be requested at sourcex.si/investors.

  • After identity removal, what share of closed deals still carry a full stage history, an amount history and an outcome?
  • What share of lost deals have a structured loss reason recorded within a week of the loss?
  • Which fields were dropped, which were transformed, and how were titles, notes and email bodies reviewed for names?
  • How many distinct sales teams, products and years does the history span, and when did the schema last change?
  • Do customer contracts restrict use of deal or pricing data, and how were those accounts handled?
  • Has any buyer tested the data against a written task, such as forecasting close dates or approving discounts?

Methods note: the schema count and the segment ratings

For Odoo, the script parsed addons/crm/models/crm_lead.py on branch 20.0 (commit ee70793) with Python's ast module and counted every fields.Type(...) assignment in class CrmLead, without counting mixin fields the class does not redeclare. A field counts as stored when it has neither compute nor related, except One2many fields, or when it sets store=True; a field counts as tracked when it sets tracking. Each field was assigned by hand to one group: identity (who a person or company is or how to reach or locate them), outcome, deal context, timing, free text or system.

For SuiteCRM, the script read the fields array of the Opportunity dictionary in modules/Opportunities/vardefs.php on the hotfix branch (commit be52770). A field counts as stored when it is not marked non-db, and as audited when it is marked audited; sales stages come from the English language file. Template fields shared by all modules, such as record IDs, dates, user IDs and a description, were not counted. It runs on the Python standard library alone and is kept with the review materials for this piece.

The segment ratings use the schema count, Salesforce's object reference, CRMArena-Pro and the privacy sources cited above. Not measured: real CRM contents, the share of companies whose deals carry loss reasons, buyer demand by segment, preparation cost or prices. Default schemas differ from customized deployments. No SourceX supply, pipeline or deal figure is used.

Frequently asked questions

Can CRM data be licensed for AI training without contact details?

Often it can, if identity fields are dropped and titles, notes and emails are reviewed for names, and if customer contracts allow the use. How much work that takes depends on the schema: some CRMs store contact details on the deal record itself, others keep them in separate tables. Counsel should review the rights and privacy questions for the specific records.

What separates sales history data from a B2B lead list?

A lead list sells who to contact; a sales history built for judgment sells what was decided and how the deal ended. The first depends on identities and is regulated as personal data and, in some cases, as data brokering. The second should survive with every identity removed.

Are business contact details personal data?

Generally yes when they identify a person. The UK ICO says the UK GDPR applies to business-to-business marketing when you hold the name of the individual representing a business, and California's CCPA has covered business contacts since its business-to-business exemption lapsed at the end of 2022.

Can synthetic CRM data replace real sales histories?

For testing workflows, partly: CRMArena-Pro built its benchmark on generated Salesforce orgs. Generated data cannot supply real loss reasons, real buyer behavior or outcomes that took months to arrive, which is where licensed histories could still add value.

Are SourceX's sales-history supply figures public?

They are not. SourceX keeps supply, pipeline and deal figures out of its articles and shares company and commercial metrics with investors who ask. Its offering is open only to verified accredited investors, and materials can be requested at sourcex.si/investors.

Sources

  • Odoo's crm.lead model on branch 20.0 declares its fields in addons/crm/models/crm_lead.py, including contact, address, stage, won or lost status and an optional lost reason, and names an untitled deal after its linked contact. Source
  • SuiteCRM's Opportunity definition stores deal fields without person contact details, joins contacts through a role table, audits sales stage, probability and close date, and ends its sales stages in Closed Won or Closed Lost. Source
  • CRMArena-Pro defines 19 expert-validated sales, service and CPQ tasks on generated Salesforce orgs; leading agents reach about 58% single-turn and about 35% multi-turn success. Source
  • The ICO says the UK GDPR still applies to B2B marketing when personal data is processed, for example the name of an individual who represents a business. Source
  • The California Privacy Protection Agency says the CCPA's business-to-business exemption lapsed at the end of 2022 and the law covers contacts at business customers and vendors. Source
  • Under the Delete Act, beginning August 1, 2026, data brokers must access the deletion mechanism at least once every 45 days and process deletion requests. Source
  • Salesforce's OpportunityHistory object represents the stage history of an opportunity. Source
  • Salesforce's OpportunityFieldHistory object represents the history of changes to the values in an opportunity's fields. Source

Request SourceX investor materials

SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.

See if your company qualifies