Data types & sources
Where the valuable data lives
Every operating company produces it. Most have no process to license it. Here’s what buyers look for, system by system — and what makes each category worth more.
Messages — Slack, email, Teams
The texture of real work: how requests come in, how people reason about them, how decisions get made. Long-running threads that resolve real problems are the core of this category. Value rises sharply when threads connect to what happened next — the ticket, the commit, the outcome.
Valuable: thread depth · resolution outcomes · specialist domains
Codebases — Git, GitHub, GitLab
Commit histories, PR reviews and issues show engineering judgement: what was tried, what was rejected, why. Repos with rich review cultures and linked issue tracking beat raw code dumps. For agent training, the review and debugging history is often worth more than the final code.
Valuable: PR discussions · issue→fix chains · internal tools
Tickets & support — Zendesk, Intercom, Jira
Naturally structured request → investigation → resolution chains, with timestamps and outcomes. Some of the most directly usable workflow data: it maps cleanly onto evaluation tasks and RL environments. Escalation threads and edge cases are prized.
Valuable: resolution chains · escalation records · outcome data
Documents — Drive, Notion, Confluence, wikis
Institutional knowledge: SOPs, runbooks, proposals, post-mortems, specialist documentation. On its own it's reference material; linked to the workflows that produced and used it, it becomes training signal for how expertise is applied.
Valuable: runbooks · post-mortems · niche expertise
CRM & systems of record — Salesforce, HubSpot, ERPs
Deal histories, pipelines, decision and escalation records over time. Longitudinal records show how businesses actually make decisions across quarters — hard to replicate, highly valued for business-reasoning tasks.
Valuable: multi-year history · decision trails · negotiation records
Raw & everything else — exports, logs, databases
Dumps, logs, internal databases, project archives — including data from winding-down companies. Messier formats need more preparation, but rarity and depth can make them very valuable. If it records how your business works, it's worth listing.
Valuable: rarity · historical depth · wind-down availability
Connectedness beats volume.
A single story traced end-to-end — a request raised in Slack, tracked as a ticket, fixed in code, documented in the wiki, closed with an outcome — is worth more than large volumes of isolated documents. If your systems link up like this, say so in your inventory: it’s the single biggest value lever.