Support tickets, CRM notes and project files quietly record how a business really works.
Somewhere in your business today, someone is solving a problem the manual never mentions. A delivery turns up damaged, the customer is furious, and the policy says one thing while the account history says another. Your team member reads the thread, checks the order, gets an exception signed off and closes the case before lunch.
To you, that is a decent afternoon of customer service. To the people building AI systems, it is something far harder to find: a complete, written record of how a real business behaves when its normal process breaks.
That gap is turning ordinary company archives into a potential new source of revenue. Years of support tickets, project files and CRM notes, the material most owners think of as a storage bill, may be exactly what the next generation of AI needs to learn how businesses actually run.
The open web has been read, but real work has not
The first wave of large AI models learned mostly from openly available text: websites, books, forums and code. Much of that has now been used. Developers are looking for new sources of real-world data, and the richest ones are not public. They sit inside companies, in the systems where staff answer customers, approve exceptions and get projects over the line.
Many developers also prefer data licensed with the owner’s agreement over whatever they can scrape. That preference is visible across the industry. In November 2023, OpenAI publicly invited organizations to work with it on public and private datasets through its Data Partnerships program. It is a useful signal of where demand is heading, though it says nothing about any particular company and involves no relationship with SourceX.
For an owner, the practical question is simpler than the industry debate. Does your company hold a record of real work that someone building AI would want to study?
Why the exception matters more than the manual
A standard operating procedure describes the path you expect people to follow. Your actual records capture the detours: the missing document, the integration that failed, the customer who changed their mind, the decision to escalate. That messy middle is what AI systems struggle with, and it is precisely what a polished handbook leaves out.
The most useful records tend to connect a problem to a decision and an outcome. Support conversations that show the customer’s issue, the response and the resolution. Workflow histories that track tasks, approvals and results. Sales and CRM records, internal documents and knowledge bases, and engineering tickets with their code review history all belong to the same family.
Completeness counts for more than sheer volume. For illustration, picture a distributor with six years of delivery exception cases, each one linking a complaint to an investigation and a fix. Next to it sits a folder holding ten thousand copies of the same shipping policy. Both take up space on the server. Only one of them shows anything about how the business makes decisions.
How the money works, in plain terms
SourceX licenses business data directly from companies for AI training and evaluation, and pays them for it. The company signs with SourceX, not with an AI lab. SourceX handles the agreement, the data preparation and the payout, with a defined scope, de-identification and contractual confidentiality built in.
The public offer is that a company can earn up to $4M by licensing its data. That figure is an upper bound for the largest enterprise datasets, not a typical or guaranteed outcome, and any early estimate is indicative. What a specific collection is worth depends on which records are included, how complete and useful they are, what rights the company holds and how much preparation they need.
Most agreements pay a one-time fee for each agreed dataset, though some provide for recurring deliveries. The originals stay with you. The license can grant exclusive AI-training rights for a set term and lasting rights over the prepared, de-identified dataset, so it is worth understanding exactly what you are granting before you sign.
Nothing leaves the building until you say yes
The process is designed so that curiosity never turns into an uncontrolled export. You decide which systems, records and date ranges are in scope. A rights review then checks that the company owns those records and can license them, including any limits set by customer or vendor contracts.
After qualification, the company signs one Master Supplier Licensing Agreement. Each deal then gets a short Dataset Deal Confirmation that fixes the exact dataset, the net payout and the terms. Nothing is delivered until both documents are signed and SourceX gives the go-ahead. A company can walk away at any point before signing, at no cost.
Data is de-identified through approved processing routes before it is used or shared. No method is perfect, and the agreement says so plainly by recording that re-identification may still be possible. Large datasets can stay in your own storage or travel on encrypted drives. SourceX reviews delivered data for up to 30 days, and standard payment follows within 60 calendar days of a valid invoice unless the signed terms say otherwise.
Who is likely to fit
The strongest candidates are knowledge-work businesses that have had 50 or more full-time employees at their peak, not counting contractors. The company does not need to be growing, or even trading. Operating, acquired and wound-down businesses can all qualify, though not every company does. SourceX prioritizes the US, also works with Canada and the EU, and considers the UK and several other markets when at least 90% of the data is in English.
Some archives will not make the cut. Records may be thin, repetitive or tied up in third-party agreements. Finding that out early is still useful, because it tells you what your systems really hold.
A first step that takes an afternoon
You can make real progress without sending a single file anywhere. List the systems that hold your records, the years they cover, the rough volume and the person who understands each one. Note where ownership might be complicated, such as attachments supplied by clients under separate agreements.
Then take the fit check. It is a few quick questions, with no uploads and no commitment. Your team has already done the expensive part by handling years of real situations well. If you want to know whether that work could earn something new, start with SourceX and see whether your records qualify.
About the founder
Noah Loul is a four-time founder and the founder and CEO of SourceX, which licenses business data directly from companies for AI training and evaluation. He previously founded B2B Rocket, acquired by Blackpearl Group. His work focuses on building permission-based data infrastructure for AI.
