Seller intent

How to license workflow and process data for AI

Last updated October 31, 2026/Seller intent

How to license workflow and process data for AI is about the trace of a job, not about the name of the tool that stored it. A useful trace shows a situation, the action someone took, the tool or step they used, and a result another person could check. Dispatch, field service, back-office review, a production handoff, a claims path, a design review: the industry changes and the shape stays. Written procedures tell people what should happen. The log, when it was actually kept, shows what did happen.

That shape is also what a reinforcement learning setup needs from real work: a state, an action, and a checkable result. What reinforcement learning environments need from real work is the technical page. This page is the owner's page. Having a log that resembles that shape is a commercial observation. It is not a statement that you may offer the log, and it is not a claim that the log is already an environment someone can train in. Turning a trace into an environment is later work, under a scope you would still have to approve. We do not skip from a curious export to that work.

The written cousin of a log is how internal documents and playbooks can be licensed. How an agent uses a sequence of real actions is how business data trains AI agents to do real work. A completed order sitting at the end of a process is how orders and invoices can be licensed. Permission for any of it, if permission exists, is a license as what AI data licensing means for a company uses that word. Your company's data could be worth up to $5M+. A line count in a log will not be multiplied into that figure on this page.

Why the action is the part public text does not keep

Articles and forums are full of advice about how a job ought to be done. They rarely keep the boring trace: which screen was opened, which check failed, which retry worked, which handoff was late. Agent builders and teams that test agents care about that trace because an agent has to act, not only talk. The same trace is why an environment can, in principle, score an attempt. Talk without an action is a weaker example for both uses.

A log can also be a worse privacy problem than a document. Process systems collect payloads: a customer's address inside a work order, a credential pasted into a step, a screenshot of an account, a free-text note that wandered far from the job. The commercially interesting skeleton is the sequence. The payload is often the part that must stay behind. We do not treat a more detailed log as automatically better.

What a process trace needs before it can be described

1. An action, not only a status. Moved, assigned, and closed are thin if the record never says what was done.
2. A result that can be checked. The job was completed, failed a test, was reopened, or was cancelled for a stated reason.
3. Repetition. One unusual project is an anecdote. The same kind of job across many weeks shows a practice.
4. A link you can explain between the step and the outcome. If the outcome lives in a second system, say so. Do not pretend the first system contains it.
5. Someone who knows which workflows are current. Old automations that no one trusts should not be offered as how the company works.

What this shelf is not

It is not a request for passwords, access tokens, or a live connection into production. It is not a claim that your logs are an environment today. It is not a promise that a research group wants your industry. The profile, if one is written, says what kind of trace exists and which limits you already know. Interest, if any, comes after that description, and it may never come.

What makes process data look large and still fail

• Heartbeats and login events counted as work. A system can emit millions of rows that never describe a decision.
• Payloads full of customer content, images of identity documents, or secrets pasted into a comment field.
• A trace the company does not have the right to offer. A customer agreement, a platform term, or a workforce rule can block a use that the log itself does not mention.
• Broken clocks and missing steps. If half the actions were never recorded, a reader cannot reconstruct the job, and we should not reconstruct it by guesswork.
• A request to stream production logs so the review can be more real. More real, in that form, usually means more personal and more secret. The free check does not need the stream.

This is not legal advice. A log the company finds impressive can still be unavailable. Value is the first lane. The right to permit a use is the second. Both can fail, and either failure is a complete answer. Other shelves sit beside this one and follow their own limits: how support tickets and chat logs can be licensed, how sales call transcripts can be licensed, and how CRM and pipeline records can be licensed.

How a trace is discussed without becoming our inventory

You are dealing with Big Data/Little Guy, operated by Unlockt, LLC, not with a buyer of your logs. We do not take a feed. The free check asks which processes are digital and how far back they go. Files stay where they are.

If a profile is worth writing, it names the workflows, the rough history, and the constraints. It does not attach the log. A match, if one appears, is a comparison with a stated need, not a delivery. Preparation happens only for a slice you approve, and preparation includes leaving out credentials and personal payloads. You can approve the skeleton of one workflow and refuse another. Refusal is a normal outcome, including a refusal of the entire shelf.

The marketplace is built for established companies, often about 50 or more employees, where the same jobs have been repeated long enough to see a pattern. A younger operation can still run the check. Nothing in the check creates an environment, a price, or a deadline. An estimate, when the check produces one, is not an offer.