The Troveo Business Data License Agreement, Explained

You license defined business data to Troveo. You keep ownership. Troveo prepares the data and licenses it onward to AI developers under its own agreements with them. You receive a share of what those developers pay, 60 percent, with no costs deducted first. You decide which systems are in scope and which are out, you can pull data from future deals, and you can choose whether the license is exclusive. In return, the license you grant is broad enough for buyers to actually use the data, and it cannot be undone for deals already signed. Everything below is detail on that paragraph. If you want the wider context first, our guide to how AI data licensing deals work covers the structures every deal in this market uses.

"Licensed Business Data" is the operational record of your company: email, chat and collaboration messages with their attachments; documents, presentations, spreadsheets, databases, reports, tickets, calendars, and project-management material; sales, marketing, customer-service, financial, operational, HR, and technical records; software code, scripts, configurations, logs, and technical documentation; and the metadata, schemas, labels, and linkages that connect all of it. What it is not is audience-facing media: a film library or a music catalog is licensed under a different agreement.

Two limits matter. Only data you actually deliver or authorize access to during the term is covered, so nothing is licensed by accident. And the Rider, the schedule attached to the agreement where deal-specific terms live, names the systems the data comes from and the systems it excludes. That is where "selectively" becomes contractual: you are not handing over the company, you are licensing named exports from named systems. What makes that kind of data valuable to buyers is covered in our guide to enterprise operational data.

The rights you grant are wide on purpose, because the buyers are AI developers and the ways they use data keep changing. The agreement lists them: training, pre-training, fine-tuning, aligning, grounding, testing, evaluating, and benchmarking models; building datasets, environments, benchmarks, task sets, retrieval systems, and agents from the data; and commercializing what results. Anything built from your data, such as a dataset, an environment, or a model, is a "Derivative Work," and the buyer or its customers may own it. Troveo owns the labels, annotations, and metadata it adds during preparation.

The rights are "sublicensable through multiple tiers," which is the clause that makes a marketplace work at all. Troveo signs agreements with buyers on your behalf, and those buyers may pass rights to their own customers and service providers. Troveo does not need your approval for each deal, because a marketplace that had to get per-deal consent from thousands of owners could not sign anything. What Troveo does commit to is that every buyer agreement obligates the buyer to follow the law and stay inside the rights granted, and that if you spot unauthorized use and tell Troveo, it works with the buyer to remedy it. The agreement is honest that model outputs cannot be fully controlled, and says so in writing.

This is the clause that gets the most questions. The license is irrevocable during its term, agreements Troveo signs with buyers can run past the end of your agreement, and ending your agreement does not unwind them. The reason is physical rather than legal: once a model has been trained on data, the data cannot be removed from it. No buyer will pay for training data under a license that could be revoked after training, because the alternative is retraining from scratch. Every serious licensing deal in this market, at every scale, carries some version of this term.

What you keep is the important half. Ending the agreement stops Troveo from signing any new buyer deals. Deals already signed continue, and so does your share of the payments from them, for as long as they run. And the Rider's removal right lets you name files, records, or categories to keep out of any future deal, prospectively, at any time. You cannot claw data back from a model that already learned from it, but you can decide what goes into the next one.

The agreement says it directly: as between you, Troveo, and the buyers, you retain ownership of the licensed data. You are granting a right to use a copy for defined purposes, not selling the asset. You keep operating on your own data throughout. The one thing you do not own is what others build from it: derivative works belong to whoever made them, and you have no ownership interest in, or right to restrict, those results. That is the standard trade in data licensing, and it is why the practical question is never "do I still own it" but "what am I letting people build."

Payment is a revenue share. You receive 60 percent of the license fees Troveo collects from buyers for your data, and the fees are counted before any costs are taken out, so nothing is recouped against your share. Fees Troveo charges buyers for its own data-preparation services are separate and not part of the split. Your share on a given delivery becomes payable once the buyer has confirmed in writing that it accepted the data and that it meets the technical specification; that acceptance step is why deliveries are prepared carefully. Payment follows within 45 days of Troveo receiving the buyer's fee, with a minimum payout threshold of 500 dollars for US licensors and 2,500 dollars outside the US, below which amounts accrue until they clear it. You have an annual audit right over the calculation of your share, on notice, at Troveo's offices, with third-party confidential information redacted. There is no universal price per record in this market and the agreement does not promise one; what buyers pay depends on the scarcity, context, and readiness of what you hold, which our guide to selling data to AI companies covers from the owner's side.

The base agreement leaves the choice open, and the Rider settles it per deal. Under a non-exclusive license you remain free to license the same data elsewhere. Under an exclusive one you do not, in exchange for the premium exclusivity commands with buyers, since it denies the data to their competitors. The Rider also offers a middle path: exclusivity that converts to non-exclusive at your election if Troveo has not generated an agreed minimum in payments to you within the first year, and optionally in each year after. That is exclusivity you only keep giving while it is producing. Which option is right depends on how unique the data is and how much optionality you want to keep; the tradeoff is the same one every owner in this market faces, and our glossary entry on data exclusivity covers it in short form.

Privacy work happens before anything ships, and the agreement sets out the mechanism. For personal information in the data, you do one of three things: remove, redact, or de-identify it using industry-standard methods; work with Troveo or a vendor Troveo designates to do that; or, where it is not commercially practicable, tell Troveo in writing what remains and where it is, so it can be handled. Sensitive identifiers, meaning government ID numbers, driver's license and passport numbers, dates of birth, and financial account or card details, and biometric data get stricter treatment: you confirm you looked for them with reasonable diligence and either removed them or disclosed them, and nothing containing them is knowingly delivered without Troveo's written approval. For any personal information that does remain, you confirm you hold the notices and consents that allow it to be licensed.

The reason is the same one that drives the whole market. Buyers pay for licensed data specifically because it comes with documented, lawful provenance, and the value of operational data is the record of how work got done, not information about the people who did it. Our guide to AI data provenance explains what buyers check and why.

Before delivery you provide a few written disclosures, each of which exists to prevent a problem that surfaces later. You list any material third-party intellectual property or materials inside the data, with the licenses or consents you rely on to include them. You disclose any prior license, sale, or transfer of a material portion of the data, especially to AI developers, data brokers, or anyone using it for model training, so a buyer is not paying for exclusivity that already left the building. You confirm the data is predominantly human-originated and identify any material categories that were generated or substantially assisted by AI, because buyers now screen for synthetic content. And you confirm no open-source or similar license attaches to the data, which matters most for code, where a copyleft dependency can attach obligations to everything it touches. If your data includes repositories, our guide to selling source code to AI labs covers that rights work in detail.

The representations and warranties are the part counsel reads twice, so here is what they amount to. You promise the data does not infringe anyone's rights and is not encumbered by a conflicting grant; that it was lawfully collected, including any personal information; that you own it or have secured every consent needed to license it, with nothing further owed to anyone; that it triggers no union or guild obligations; that it contains no malware, tracking, or disabling devices; that it does not conflict with your other agreements; that no open-source license applies; that AI-generated portions are disclosed; and that it is not freely public. You also agree not to bring claims against Troveo for a buyer's misuse of the data.

If some of the data turns out not to comply, the remedy is targeted: you replace, remove, or fix that data at your cost, or refund the share you were paid for it. Behind the warranties sits an indemnity: you cover Troveo and the buyers for losses arising from a breach of those promises, from gross negligence or willful misconduct, or from exploitation of the rights, and you add Troveo as an additional insured on relevant policies. These are the terms buyers require of Troveo, passed through to the source of the data, because a buyer's own legal exposure depends on the chain of rights being clean; our explainer on rights-cleared training data shows why that chain is what they are paying for. The Rider can add a matching indemnity from Troveo to you, for Troveo's own breaches or misconduct, and a mutual liability cap with the usual carve-outs. If a term in this section concerns you, that is the conversation to have before signing, not after.

Most of what a licensor negotiates lives in the Rider, and it is worth knowing what is there before you assume something is fixed. It names the systems in scope and the systems excluded. It carries the removal right, so you can withdraw files, records, or categories from future deals on written notice. It settles exclusivity, including the opt-out described above. It can list authorized buyers, so Troveo may sublicense only to named companies, or excluded buyers, who cannot receive your data without your written approval. And it can add the Troveo indemnity and the mutual cap. The base agreement is a template; the Rider is where a deal takes the shape of your company. A structured process for deciding what belongs in it, from inventory through rights review to scope, is laid out in our owner's guide to licensing company data for AI.

The initial term is three years, renewing in one-year periods unless either side gives notice thirty days before a period ends. Troveo may end the agreement early, including for convenience; the practical effect for you is that no new buyer deals get signed, while existing ones and your payments from them continue. The agreement and its terms are confidential on both sides, and public statements about the collaboration, Troveo, or any buyer go through Troveo's approval, which is standard where buyers themselves require confidentiality. Troveo may reference the collaboration in its own trade communications. A non-circumvention clause runs for the term and one year after: you do not deal directly on these rights with a buyer Troveo introduced or was actively in discussions with. That is what protects the work Troveo does to find and qualify buyers on your behalf. Disputes go to mediation and then confidential arbitration.

None of this is worth a lawyer's hour until you know whether your data is worth licensing. Troveo's free data value assessment takes about five minutes and scores the AI-training value of what your company holds, and the business data program page covers how the process runs from inventory to payment. Why a company would run this model at all, with examples at every scale, is in our guide to data licensing as a business model. When you are ready to talk terms, or want to walk through the agreement with a person, talk to us.