Skip to content
Outrich Data

Data type

License your product usage data for AI training

Product usage data can be licensed for AI training only where your customer agreements permit that use. Outrich Data, as a direct buyer, starts with your terms of service and DPAs, then considers de-identified event sequences with user, account and content fields removed. Customer content stored in your product is your customers' data and is never included.

What counts as product usage data

Product usage data is the event stream your software records as people use it. Each event says that some user in some account did something at a certain time: opened a screen, ran a search, created a record, hit an error, used a feature, invited a colleague. Sessions, feature flags, plan tier and device details ride along.

It must be kept apart from customer content. The documents, messages, files and records your customers put into your product are theirs, and you process them on their behalf. Telemetry about how the product was operated sits in a grayer zone that your contracts define. Apps directed at children, and usage records from them, are excluded.

Why AI labs want it

A model that operates software needs to know how people really move through it. Usage events show the path a user took to finish a task, where they hesitated, which error sent them to the help page and what they tried next. AI labs use those sequences to train and evaluate agents that complete tasks inside applications.

The streams also record behavior over months. An account that adopted three features and renewed looks different from one that stalled after setup. That history supports work on recommendation, onboarding guidance and churn prediction. Public sources hold almost none of it, because event data rarely leaves the vendor's analytics stack.

Removed before licensing

What comes out first.

  • User IDs, emails, device IDs, cookies and advertising identifiers
  • Account, workspace and company names, replaced with codes
  • IP addresses and precise location, reduced to country or dropped
  • Free-text event properties such as search terms, record titles and file names
  • Any customer content captured in event payloads or session replays
  • URLs and referrers that carry tokens, record IDs or customer subdomains
  • Events from accounts whose contracts prohibit the use, excluded in full

Data title search

What gets checked.

  • Data processing agreements. A DPA normally limits you to processing customer personal data on the customer's instructions. Licensing it to a third party is usually outside those instructions.
  • Usage data clauses. Many SaaS agreements let the vendor use aggregated or de-identified usage data for its own purposes. Read whether that wording reaches disclosure to others, and for which customers.
  • Privacy policy and end users. What you told end users about analytics governs reuse. Under CCPA a transfer for payment can be a sale, with opt-out rights attached.
  • GDPR. Pseudonymized events remain personal data. A lawful basis and purpose compatibility are needed for EU users. Review with counsel.
  • Sector rules. Usage data from health, education or financial products can fall under HIPAA, FERPA or GLBA depending on your role.
  • Analytics vendor terms. Check the export and use terms of the tools that collected the events.

Who signsThe head of product or CTO sponsors it and the general counsel approves, often with the privacy lead or DPO. The data engineering team builds the extract from the warehouse.

Value

What moves the price of product usage data.

FactorEffect on value
Event granularityFine-grained actions in order are generally worth more than daily counts per account.
A tracking planDocumented event names and properties let a lab understand the stream without guessing.
Product complexityMulti-step professional workflows are scarcer than consumer scrolling and tapping.
Known end statesSequences that end in a known result, such as task completion, upgrade or cancellation, support evaluation.
Account coverageThe share of accounts whose terms allow the use sets the real size of the dataset.

Questions

Licensing product usage data: what companies ask.

Our terms let us use aggregated data. Does that cover licensing it?
It depends on the exact words. Some clauses allow use only to improve the service, some allow any use of de-identified data, and enterprise customers often negotiated their own version. Outrich Data reads the clause with you, and your counsel decides which accounts are in scope.
Is customer content ever part of this?
No. What your customers store in your product belongs to them, and you hold it as their processor. Only telemetry about product operation is considered, and event properties that captured content are stripped.
Are hashed user IDs anonymous?
Generally not. A hashed ID still links one person's events together, and long sequences can be distinctive. IDs are replaced with fresh random codes per dataset and rare patterns are reviewed. The free re-identification risk checker on this site shows the idea on a sample.
Can individual enterprise customers be left out?
Yes. Exclusion works per account. Accounts with restrictive DPAs, regulated customers and anyone who objected are removed in full before events are processed.

Find out if your data qualifies.

Eleven questions, about five minutes, no call required. You get an answer on screen and a written follow-up.

esc