What to send an AI company first: a safe sample of your data

The first thing you send an AI lab is a small, de-identified slice of your operational data that captures a complete request-to-outcome workflow. It must be safe, free of personal identifiers and any confidential details, and it should represent the larger set you hope to license.

What a safe sample looks like

A safe sample is a handful of records that let a buyer follow a business task from initiation through resolution. Typical examples include

• a support ticket together with the final resolution note,
• a sales-lead entry linked to the quote that closed the deal,
• a project-task record paired with the final deliverable status.

Each record should contain the fields that describe the request (date, category, requester), the actions taken (steps, agents, timestamps) and the result (outcome, closure reason). You do not need thousands of rows; a few dozen records or a short thread is enough, as long as the full chain is present for the buyer to see how the reasoning progresses.

How to pick the right slice

Ask yourself three quick questions before you export anything:

1. Does this slice cover a full end-to-end process?
2. Is the data typical of what you generate every month?
3. Can any personal names, email addresses or customer-specific codes be removed without breaking the chain?

If the answer is yes, you have a candidate. Imagine you run a 60-person logistics firm. Choose three recent shipment-issue tickets, each with the initial request, the internal chat that decided the reroute, and the final delivery confirmation. That set shows the workflow, reflects current operations, and stays small enough for a quick review.

Not sure your company is a fit? Take the free check

Removing personal and confidential details

Before any broker touches the data, it must be de-identified. The steps we follow, and that most AI labs require, are

• strip all direct identifiers such as names, email addresses, phone numbers and employee IDs,
• replace remaining indirect clues (exact dates, locations, unique order numbers) with broader ranges or stable pseudonyms,
• have a human reviewer attempt to re-identify any person or company; any failure means the file is ready.

We offer de-identification as a separate service; see the de-identification page for details. In most brokered deals the AI lab pays for the cleaning, but the exact arrangement is worked out in the contract. Whichever party performs the cleaning, you will receive a copy of the cleaned file to confirm it meets your standards before any buyer sees it.

Packaging the sample for the broker

When the sample is ready, follow these steps:

1. Save the records in a common format such as CSV or JSON, avoid proprietary binaries that are hard to inspect.
2. Draft a one-page summary that states the data type, the time span covered, the number of records and the business process illustrated. Do not attach the raw data to this summary.
3. Upload both the file and the summary through the secure portal on our site.

The broker never receives raw files until you approve a quote, and nothing moves until you give the final go-ahead.

The quick fit check on our site

Before you invest time in cleaning, you can see whether your company meets the basic criteria with our free fit check. Companies need at least 30 full-time employees; there is no upper limit, and larger firms usually generate more data to offer. The check asks a few basic questions and returns an instant indication of eligibility. Start the check on the sell page and then use the button below to begin the process.

Once you have a safe sample ready, we will share a one-page description with several AI labs, collect their quotes, and bring the best offer back to you. The negotiation, any additional de-identification work, and workflow-licensing specifics are all worked out together with you before any data leaves your environment.

The next step is to run the fit check, prepare a small de-identified slice, and click the button below.