# Enterprise decision trails as training data | SourceX

Canonical: https://refer.sourcex.si/guides/enterprise-decision-trail-training-data
Last updated: 2026-11-04

A business decision is explained by a chain of records, not one document: the request that started it, the analysis and its revisions, the approval with the approver's authority at the time, any reversal and its reason, the revised decision, and the review that measured what happened. Those records usually sit in four or five systems. A trail is complete when one identifier or reliable cross-reference links all of them, and when version and approval history were kept rather than overwritten.

## Which records explain a business decision, its approval, its reversal and its final outcome?

Six record types, linked (seven records in the example below, because the analysis has two versions): request, analysis versions, approval, reversal, revised approval and outcome review. Each answers a different question, and losing any one breaks the story.

| Record | Question it answers | Typical system | What makes it usable |
|---|---|---|---|
| Request | Who asked for what, and why then? | Ticketing or project tool | Creation date, requester, problem statement |
| Analysis versions | What options were weighed, and how did the case change? | Document library | Version history kept, not overwritten |
| Approval | Who said yes, under what authority? | Approval workflow or email | Approver, timestamp, authority limit in force |
| Reversal | What changed, and who withdrew the approval? | Ticket, email, meeting minutes | Reason recorded, link to the original approval |
| Revised approval | What was approved the second time? | Approval workflow | New version referenced, same decision ID |
| Outcome review | Did it work, measured against the forecast? | Review document, ERP or BI report | Planned versus actual figures, review date |

Public text rarely shows the middle of this chain. Company announcements report a final decision; they almost never show the rejected option, the approval that was withdrawn or the review that found the forecast was wrong. That middle is what makes a trail useful for training and evaluating AI systems that must recommend, approve or check business decisions. See [what AI training data is](/proof/what-is-ai-training-data) for the broader picture, and the [decision records data type](/data-types/decision-records) for the categories of records involved.

## One illustrative decision, fully linked

Illustrative. A fictional 240-person regional distributor decides to close one of its two warehouses. Every record below carries the identifier DEC-117 or links to a record that does.

| Step | Date | Record | System | Authority shown |
|---|---|---|---|---|
| 1. Request | 2023-02-06 | Ticket: consolidate Site B into Site A | Jira | Operations director as requester |
| 2. Analysis v1 | 2023-02-20 | Memo with three options and a savings forecast | SharePoint library | Finance analyst as author |
| 3. Analysis v2 | 2023-03-03 | Revised memo after freight quotes changed the forecast | SharePoint library | Same author, version 2.0 |
| 4. Approval | 2023-03-10 | Approval of option B, citing memo v2 | Approvals app | CFO, within the authority limit in the policy version then in force |
| 5. Reversal | 2023-05-22 | Approval withdrawn after a lease exit clause was found; memo v3 | Email thread and ticket comment | CEO, as the only person able to override the CFO |
| 6. Revised approval | 2023-06-14 | Partial consolidation approved, citing memo v4 | Approvals app | CEO and CFO jointly |
| 7. Outcome review | 2024-03-18 | Review comparing forecast and actual carrying cost | SharePoint library, citing ERP figures | Operations director, presented to the leadership team |

Three details make this trail strong. The memo kept every version, so the change between v1 and v2 is visible. The approval cites the memo version it approved, so a reader can tell what the CFO actually saw. And the reversal names its reason and links back to step 4, so the withdrawal is not mistaken for a new decision.

## What each system keeps, and what it quietly drops

The same trail can be complete in one company and broken in another depending on settings nobody remembers choosing.

- **Document versions.** Microsoft's documentation says SharePoint can store, track and restore files whenever they change, with versioning on by default in libraries but off by default in lists. A decision register kept as a list with versioning off holds only the latest state.
- **Approval history.** Microsoft also notes that the approval columns on a SharePoint list item show only the latest approval activity; the full history is in the Approvals app. A trail that relies on the column alone will show "approved" and hide the withdrawal.
- **Ticket history.** Atlassian documents that Jira records who changed a field and when, visible in the issue's History tab. Status changes such as "approved" to "on hold" survive there even when the ticket's current status says nothing about them.
- **Email.** Reversals often happen in a thread, not a system. If former executives' mailboxes were deleted, step 5 may be gone even when every other record survives.

For engineering choices, [architecture decision records](/data-types/architecture-decision-records) provide a similar chain inside a code repository, and [capex requests with post-implementation reviews](/data-types/capex-request-post-implementation-review) are a common finance-side example. The guide on [assessing operational decision logs](/guides/how-to-assess-operational-decision-logs) covers high-volume, rule-driven decisions rather than one-off ones like DEC-117.

## A partner's completeness check, at metadata level

Ask the owner or a senior manager these questions. None requires opening a record.

- [ ] Do major decisions get an ID, project code or ticket number that appears in the related documents?
- [ ] Is version history switched on where analyses and registers are kept?
- [ ] Do approvals happen in a tool that keeps history, or only in email?
- [ ] Is the authority policy (who can approve what, up to which limit) itself versioned?
- [ ] Are reversals recorded with a reason, or does the old approval simply disappear?
- [ ] Do outcome reviews exist for a meaningful share of decisions, and do they cite the original forecast?
- [ ] Roughly how many such decisions a year, over how many years?

Partners never ask for the records, samples or screenshots. Discovery stays at the level of what exists, how much, over how long, in which system and who owns it. When an owner is weighing whether to raise a licensing idea internally, [who to tell first about a new initiative](/guides/who-to-tell-first-about-a-new-initiative) helps with sequencing.

## Does the company qualify, and who can authorize a discussion?

The introduced company needs 50+ full-time employees at peak (contractors excluded), a documented operating history, rights to license its records and an authorized contact. Companies in the U.S., Canada and the EU qualify, as do companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 70% of the data is in English.

Decision trails raise two ownership questions early. Board and investor materials may be subject to confidentiality duties, and records involving customers or suppliers may be limited by contract. The person who can authorize a licensing discussion is the CEO (in an owner-run company, usually the owner), not the operations manager who kept the tickets. What stays unknown until SourceX reviews the opportunity: whether the trails fit current dataset needs, which records the rights review clears, what de-identification is required and whether any offer follows.

## What public demand signals show, and what they do not

Demand signals are public but should be read carefully. On 2026-10-09, micro1 announced a commitment to spend $1 billion over 12 months acquiring and licensing enterprise operational data. That is a company announcement of intent; it is not evidence that the money has been spent or that any particular company's decision records will be bought. SourceX licenses approved business data directly from companies for AI training and evaluation, and manages the agreement, rights review, data preparation, delivery and payout.

## Company compensation and the referral reward

If a company licenses its records, SourceX pays the company under its agreement. A referral reward, if one becomes payable, is calculated on what SourceX actually pays that company and is funded separately, so it never reduces the company's payout. Nothing is owed for an introduction or a signed agreement alone.

## Next step

> **Refer a company. No cap on rewards.** Default reward on a company's first qualifying transaction: the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap; further qualifying transactions first agreed within 12 months of the first Closing pay 15%. Eligibility and payment conditions apply, including the USD 100,000 Minimum Transaction, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.

If an owner can answer yes to most of the checklist, [register as a partner](/register), confirm the company against [who qualifies](/who-qualifies), and read the [program terms](/terms) before you make the introduction.

## Sources

- https://support.microsoft.com/en-us/sharepoint/data-and-lists/view-the-version-history-of-an-item-or-file-in-a-list-or-library
- https://support.microsoft.com/en-us/office/approvals-in-lists-document-libraries-2bd0954d-5797-4be3-b78a-846f26338e17
- https://support.atlassian.com/jira/kb/how-to-identify-and-list-when-a-field-value-was-changed-in-jira/
- https://www.micro1.ai/blog/micro1-commits-1b-to-enterprise-data-acquisition
- https://refer.sourcex.si/terms
