How to get trajectory data for AI agents

Trajectory data for AI agents comes from licensed operational datasets that record complete workflows, the choices made along the way, and the final results. We match you with companies that own that data, take care of legal rights and de-identification, and deliver a quote that fits your project.

What trajectory data looks like for AI agents

Trajectory data is a step-by-step record of a business process, from the initial request to the outcome. Each step includes the system state, the action taken, any observations that informed the action, and the eventual result. Typical sources include support tickets that contain a problem description, troubleshooting steps and a final resolution; sales and CRM flows that capture lead creation, outreach actions, negotiation notes and the final win or loss; project-management logs that show task assignments, status changes, dependencies and final delivery; and accounting exports that detail invoice generation, approval steps and payment receipt. These chains give reinforcement-learning agents a realistic environment in which to practice decision-making. An isolated email archive lacks the action-outcome link and therefore provides far less training value.

Which companies generate useful trajectories

A seller needs enough activity to produce a meaningful stream of records. The baseline is 30 or more full-time employees. There is no upper limit; larger firms simply generate more data, which can raise the price but also enrich the trajectories.

Value-driving factors include:

• Years of continuous history on the same tools, which gives a consistent view of how work evolves.
• Data that surface reasoning, such as chat threads where a decision was reached.
• A niche or rare industry where comparable data is scarce.
• Clear rights to share the data after de-identification.

Any company that meets these criteria can be approached, regardless of its size.

Need licensed data for a model? Tell us what you need

How we package the data for RL training

When a seller agrees to share data, we follow a strict, privacy-first process. The raw files never leave the seller until a licensing agreement is signed. After signing, we export the data through a secure channel and run it through automated de-identification tools, followed by human review. All personally identifiable information and any content the seller marks as confidential are stripped out. Names become stable pseudonyms such as "employee 632" so the sequence of actions stays intact while identities are hidden.

The seller can review a sample of the cleaned data before the final hand-off. The exact method for handling raw files, the timing of deletion, and any other security details are written into the contract.

You license the trajectories with a one-time payment, retain ownership of the original data, and keep using it internally. Use rights are usually perpetual because a trained model cannot reproduce the raw records. If you need exclusive access, the standard term is 24 months; a shorter exclusive period is possible at a lower price.

Pricing and licensing basics

Approved deals typically fall between $100,000 and $2 million USD. With many buyers reporting a common range of $200,000 to $300,000 USD for a dataset that meets their needs. The price reflects the amount of data, its type, the rarity of the industry, and the length of the historical record.

There is no cost to the seller up front. The buyer pays a finder's fee only when a deal closes, and the seller's advisory fee is also due only at closing and only after it has been agreed in writing. All licenses are structured as data licensing, not a sale, so the seller retains ownership and can continue to use the data internally.

The steps to tell us what data you need

1. Click the button below and complete a short questionnaire that asks about the type of trajectories you need, the volume you expect, and any industry or workflow preferences.
2. We turn your answers into a one-page summary, no raw data is shared at this stage.
3. The summary is sent to a curated group of companies that match your criteria.
4. We collect the best quotes and the conditions each seller is comfortable with, then present them to you.
5. You pick the quote that fits your project, and we work with the seller to finalize the licensing terms.
6. After the contract is signed, the data is exported, de-identified, and delivered to you securely.

You stay in control of every decision. We handle the matchmaking, the legal checks, and the secure hand-off. Let us know the kind of trajectory data you are after by using the button below.