How to become a data broker

Search for data brokers and most of what you find is about companies like Acxiom or the people-search sites: firms that collect records about consumers and sell them for marketing, background checks or fraud screening. That business needs huge collection pipelines, compliance teams and registrations in several states. It is not what most people mean when they ask how to become one.

The other job is closer to a business broker or a real-estate agent. A company is sitting on data it built for its own product: years of support transcripts, a codebase from a product it shut down, sensor readings, labelled images, pricing histories. Somewhere a buyer, often an AI lab or a data team at a large company, would pay to license it. The broker finds both sides, works out what the data is worth, gets the paperwork right and takes a commission when the deal closes. This guide is about that second job.

Generalists struggle because every dataset is judged on different things. A buyer of source code asks about commit history and licences; a buyer of call recordings asks about consent and speaker diversity. Pick one or two kinds of inventory and learn them properly.

• Codebases from closed or pivoted software companies, which AI labs license to train and test coding models.
• Text and conversation data such as support tickets, forum archives or transcripts, where the owner can show consent and can strip personal details.
• Domain data like logistics records, retail prices, satellite or sensor readings, which hedge funds and analytics firms buy as alternative data.
• Media libraries: images, audio and video with clear ownership, sold for model training.

A good niche has three things: buyers you can name, owners you can reach, and a rights picture you can check. If you can't list ten companies that would buy the data and ten that might hold it, keep looking.

Buyers rarely pay for volume alone. Before you pitch anything, learn the questions a serious buyer will ask, because those questions decide the price and whether a deal happens at all.

• Rights: does the seller own the data outright, or did it come from customers, partners or scraping under terms that forbid resale?
• Personal data: does it identify people, and if so, what did those people agree to?
• Exclusivity: will this buyer be the only one, or is the seller licensing it to others?
• Quality: how complete, clean, recent and well documented is it? Is there a sample?
• Delivery: what format, how big, and will updates follow?

Write these into a one-page intake sheet and fill it in for every seller you meet. It becomes the core of your pitch and saves you from bringing a buyer something they can't legally use.

Most data owners have never thought of their data as something to sell, so you won't find them in a marketplace. You find them by looking for the situations that leave valuable data behind.

• Startups that shut down or pivoted: the codebase and data outlive the product, and founders often want to recover some value.
• Mature companies with years of operational records in a niche that AI labs or analysts care about.
• Agencies and research firms holding archives built for clients, where the contracts allow reuse.
• Communities and publishers with large archives of original text, if their terms and their members' consent allow licensing.

Reach out to the person who can actually decide, usually a founder, CEO or head of data. Lead with a specific reason their data might be worth something, not a generic offer. Expect most to say no or not now; a small number of serious conversations is enough to start.

This is the step new brokers skip and later regret. Ask the seller to show where the data came from and what agreements cover it: customer terms of service, privacy policies, contributor agreements, open-source licences inside the code, and any contracts with partners. If personal data is involved, find out whether it can be removed or de-identified before sale and what the people involved were told.

There is no list price for data. Value depends on who would buy it, how hard it would be for them to collect it themselves, whether they get it exclusively, and how clean and documented it is. Start from the buyer's side: what would it cost them in time and money to build the same thing, and what is it worth to their product to have it now?

Give sellers a range, not a promise, and say what moves it up or down. A rough range keeps both sides realistic and protects you when a buyer's offer comes in lower than the seller hoped. DataBroker's app gives an estimate range for codebases, task data and raw data to help with this first pass; treat it as a starting point, not an offer.

Paperwork protects your commission and keeps both sides comfortable sharing details. The order matters: you want your fee agreed before you introduce a buyer, and confidentiality in place before anyone sees a sample.

Pitch buyers with the intake sheet, a small sample and the value range. Name the use the data serves for them, not just what it contains. Once a buyer is interested, keep both sides moving: answer questions quickly, push for a letter of intent before long legal reviews start, and use a data room so access is logged.

Deals take weeks to months. Expect diligence questions you didn't anticipate and be honest when you don't know an answer; buyers remember brokers who overstated what a dataset contained.

• Knowledge of one data niche and the buyers in it.
• A way to find and contact owners: research time, a CRM or spreadsheet, and patience.
• Template agreements you understand, reviewed by a lawyer.
• A business entity and bank account to receive commissions.
• Honesty about what you don't know, especially on rights and privacy.

You don't need a degree, a licence or capital to buy data up front, because a broker doesn't own inventory. What you trade on is trust, so the habits above matter more than any tool.