Five categories
| Category | Typical systems | What it teaches an AI |
|---|---|---|
| Communications | Slack, Teams, Gmail, Outlook | How work is coordinated and handed off |
| Business systems | Salesforce, HubSpot, QuickBooks, Xero | How deals, transactions and accounts progress |
| Documents | Google Workspace, Microsoft 365 | How drafts become finished work |
| Projects and workflows | Jira, ServiceNow, Zendesk | How problems are triaged and resolved |
| Technical data | GitHub, GitLab | How software is written, reviewed and fixed |
What makes it useful
- Decisions. Why a quote was priced, a claim approved or a vendor chosen.
- Revisions. How a draft, design or forecast changed before it was final.
- Exceptions. What the team did when the standard process didn't fit.
- Approvals. Who signed off, in what order and on what basis.
- Outcomes. What happened next, and whether it worked.
These five signals are what separate valuable operating data from a pile of files. See the eight value drivers.
One support ticket, five signals
Support ticketBilling · Resolved
Customer reports a duplicate charge on the March invoice
Agent drafts a reply, then rewrites it after checking the ledgerRevision
The charge is outside the standard refund window, so it is escalatedException
Billing lead reviews the case and signs offApproval
Refund issued for the duplicate chargeDecision
Customer renews the next quarterOutcome
Work exhaust: what agents learn from
Businesses that may not realize they hold it
Agencies, law firms, consultancies and field service companies generate exactly these records every day. We look at them in professional services firms are sitting on AI training data.