Data for reinforcement learning environments and reward models

Reinforcement learning needs a way to tell good work from bad. Enterprise records often carry that signal already: a ticket resolved or reopened, a deal won or lost, a QA score, a repair that needed a callback. SourceX sources data with those outcomes attached, for building RL environments and training reward models.

Last updated October 3, 2026

Request data

Why enterprise data

Outcome labels in business systems were recorded by the people doing the work, at the time, for their own reasons, which makes them hard to fake and close to the goals companies care about.

Linked records let you build environments from real history: the starting context, the actions available, and what actually happened next.

Relevant data

Data	Category	Modality	Availability
Helpdesk ticket threads with resolutions Multi-turn tickets with internal notes, tags, status history and the final resolution.	Customer support	Text with structured metadata	On request
QA-scored support interactions Tickets, chats or calls graded by human reviewers against a scoring rubric.	Customer support	Text with scores	On request
CRM pipelines with won/lost outcomes Accounts, opportunities, stage changes and activities through to the final outcome.	Sales and CRM	Structured data and text	On request
Outbound sequences with replies and outcomes Prospecting emails and follow-ups with real replies, meetings booked and deals created.	Sales and CRM	Text with structured metadata	On request
Task histories with context, actions and outcomes Work items from request to completion, with each action taken and the result.	Operations and workflows	Structured data and text	On request
Field-service work orders with technician notes and photos Jobs from dispatch to completion: diagnosis notes, parts used, photos and outcomes.	Operations and workflows	Structured data, text and images	On request
Pull requests with code review threads Diffs, review comments, requested changes, approvals and CI results for each change.	Software engineering	Code and text	On request
Incident, on-call and postmortem records Alerts, incident timelines, responder chat and the postmortems that followed.	Software engineering	Text and logs	On request
RFP responses and proposals with win/loss outcomes Requests for proposal, the responses submitted and whether each bid was won.	Sales and CRM	Documents	On request

What to include in your request

The outcome signal you want (for example resolution, approval, won or lost, QA score)
Whether you need full trajectories or only final outcomes
How many episodes you need and over what time span
The systems the environment should mirror

Questions

Which business data has built-in reward signals?

Support tickets (resolved, reopened, CSAT), QA-scored conversations, CRM opportunities (won or lost), outbound sequences (replies and meetings), field-service jobs (callbacks), pull requests (approvals and CI results) and RFPs (won or lost).

Can SourceX help build the environment itself?

SourceX sources the data and manages the licensing agreement. Describe how you plan to use it in your request so the package includes the fields your environment needs.

Need this data for a model?

Describe what you need: domain, volume, history, format and licensing terms. SourceX looks for companies that hold it and can license it.

Request data Talk to SourceX