Proof of demand

An announcement is a public statement of intent about a category of data; acceptance is a decision about one delivered dataset, made after it was checked. Between them sit published sourcing criteria, which describe the kind of company a buyer screens for. Only executed acceptance, followed by cleared payment, starts the money moving for a specific company, and only that can lead to a referral reward. Use the checklist below to sort what a client has heard before anyone builds expectations on it.

When a written introduction and a link click disagree, the evidence that matters is dated proof of who first validly introduced the company's authorized contact within the 90-day window. The terms credit an introduction, not a click, and credit is recorded when the company submits a verified application. A click shows a page was opened; a dated email to an authorized contact who opted in shows an introduction. Conflicting claims go to duplicate review, which weighs that evidence under the terms.

Yes. Public filings show AI developers pay for licensed data. One online platform disclosed about $203 million in AI data-licensing contracts over roughly three years.

SourceX assesses the data, then de-identifies it, through approved processing routes, before it is used or shared — personal identifiers and confidential details are removed or masked, and contracts define exactly how buyers may use it.

Reported AI data-licensing deals range from tens of millions per year to a five-year agreement reported at up to $250 million, depending on the data and the buyer.

Yes. Publicly reported deals include a five-year agreement reported at up to $250 million between a major news publisher and a leading AI developer.

Data licensing is a paid agreement where a company grants an AI developer the right to use its data for training or evaluation, under defined terms.

Buyers want real operational data — support conversations, documents, sales and finance records, engineering artifacts — because it teaches models how work actually gets done.

Public web text is a finite, heavily reused resource; researchers project high-quality public text may be effectively exhausted this decade, pushing buyers toward licensed data.

Epoch AI estimates the effective stock of public human-generated text at about 300 trillion tokens, and projects models could fully use it between 2026 and 2032.

Analyst firms estimate the AI training dataset market at about $2.8 billion in 2024, growing roughly 28% a year to near $10 billion by 2029–2030.

AI labs and data users want licensed data because it comes with clear rights and provenance, and because analysts report demand moving toward consented, domain-specific datasets.

Major AI developers have signed data-licensing deals worth tens to hundreds of millions of dollars, according to public reporting. This timeline lists the deals with links to sources.

Reddit's IPO filing disclosed $203 million in data-licensing contracts, mostly with AI developers. It is one of the clearest public data points that AI labs and data users pay for licensed content.

News Corp's agreement with OpenAI was reported at up to $250 million over five years, making it the largest publicly reported AI content-licensing deal.

Shutterstock publicly disclosed $104 million in AI data-licensing revenue in 2023, showing that licensing existing data libraries to AI developers is a real, reported revenue line.

Public reporting and industry disclosures show AI developers increasingly paying for expert-produced reasoning data in domains like law, finance and science, as open-web data is exhausted.

Reuters has documented an intensifying race among major AI developers to license proprietary data directly from companies that hold large private archives.

Industry estimates suggest AI developers collectively spend on the order of $10–15 billion per year on data and the human processes around it — a figure that is growing.

Any company with years of operational history holds data that reflects real work — support resolutions, sales conversations, financial processes — which is exactly what AI developers now seek.

Multiple lawsuits and regulatory actions are testing whether scraping content for AI training is lawful. Licensed data gives buyers legal certainty that scraped data cannot.

AI data-licensing deals typically define what data is covered, what uses are permitted, how the data is protected and anonymized, and how and when payment flows.

Operational data is the byproduct of running a business: support tickets, sales calls, documents, financial workflows. Because it captures real work, AI developers value it for training and evaluation.

Epoch AI, a research group, estimates that the stock of public human-generated text data could be fully used for AI training between roughly 2026 and 2032, pushing buyers toward private and licensed data.

The New York Times' AI content-licensing agreement with Amazon was reported at roughly $20–25 million per year, according to press coverage of the deal.

Data provenance means knowing where data came from, who consented to its use, and what rights attach to it. Licensed data has provenance; scraped data usually does not.

AI developers need real business data not only to train models but to evaluate them — testing whether a model can actually do support, sales or finance work.

Synthetic data — text generated by AI models themselves — is useful but has documented limits, which is why buyers keep paying for real licensed and operational data.

A company licenses approved business data directly to SourceX for AI training and evaluation. SourceX manages the agreement, rights review, data preparation, delivery and payout, subject to acceptance at Closing and agreed terms.

AI data demand is not limited to tech: reporting shows buyers seeking data from any industry where knowledge work is documented — support, sales, finance, legal, operations and more.

Consent — documented permission to use data for defined purposes — is what separates licensable company data from legally risky scraped data, and buyers pay for that certainty.

Analyst reports value the AI training dataset market at roughly $3.6 billion in 2025, growing toward $4.4 billion in 2026, with double-digit annual growth forecast through the decade.

Human-in-the-loop data records how real people perform tasks and judge outputs. As AI models take on business work, developers pay for this data to train and evaluate them.

High-quality training data is heavily English-dominated, so AI developers actively seek real business data in other languages — an advantage for international companies.

SourceX is the company's direct licensing partner and contractual counterparty for approved business data used in AI training and evaluation. It manages the agreement, rights review, data preparation, delivery and payout, subject to acceptance at Closing and agreed terms.

Before purchasing, AI labs and data users evaluate datasets on provenance, quality, coverage, rights documentation and preparation quality — criteria that determine whether company data is licensable.

Exclusivity — rights a competitor cannot replicate by scraping — is a major reason the largest reported AI data-licensing deals reach hundreds of millions of dollars.

Anonymization, or de-identification, means removing or transforming personal identifiers to lower the chance that individuals can be identified. SourceX de-identifies the data, through approved processing routes, before it is used or shared. No method is perfect: the supplier agreement records that re-identification may still be possible and that SourceX is not liable for it.

Public data exhaustion, legal pressure on scraping and enterprise AI adoption are converging, making the next few years the formative window for company data licensing.

Workflow data captures the sequence of steps, decisions and outcomes in real business processes. It is among the most valuable and hardest-to-obtain data for AI training.

Licensing grants a buyer defined usage rights while the company retains ownership of its data. Selling transfers ownership outright. Reported AI data deals are licenses, not sales.

A dataset is licensable when the company has authority over it, it has sufficient volume and history, it reflects real work, and it can be anonymized and documented.

No universal minimum exists, but AI labs and data users look for meaningful volume and multi-year history. Companies with 50+ full-time employees at peak typically clear that bar through normal operations.

Training data is bought by AI labs and data buyers — frontier model developers, enterprise AI teams building domain models, and data providers that supply them.

Fine-tuning is the process of adapting a general AI model to a specific domain or task using targeted examples — which is why AI labs and data users seek real business data.

Dataset pricing varies enormously — reported deals span thousands to hundreds of millions of dollars — driven by exclusivity, volume, quality, rights clarity and SourceX's current data needs.

RLHF (reinforcement learning from human feedback) trains AI models using human judgments about which outputs are better. It is among the most expensive data categories developers purchase.

Multi-year data histories show how real work evolves through changing conditions, making them more valuable than one-time snapshots for AI training and evaluation.

Dataset documentation describes what a dataset contains, where it came from, what rights attach and how it was prepared. AI labs and data users require it before any purchase.

The AI industry is shifting from models that answer questions to agents that perform tasks. Training and evaluating agents requires records of real work, data that exists only inside companies.

First-party data is data a company collected directly through its own operations. Because the company holds clear authority over it, it is the most straightforward category to license.

Properly structured data-licensing agreements protect the company through anonymization requirements, defined use limits, security obligations and retained ownership of the underlying data.

Post-training is everything after initial model training: fine-tuning, RLHF, evaluation and alignment. It depends on targeted, real-world data and is a fast-growing share of AI data spending.

Customer support data records real problems, real resolutions and the judgment of experienced agents, making it one of the most valuable company data categories for AI training.

Sales and CRM data records how real persuasion, negotiation and deal management work — capabilities AI developers cannot learn from the public web.

Company documents — policies, SOPs, contracts, reports — show how organizations encode and apply knowledge, making them a sought-after AI data category.

Recorded business calls capture real spoken professional communication — tone, timing, negotiation — a data category AI developers cannot obtain from text sources.

Finance and accounting data records structured, rule-based professional work — reconciliation, month-end close, reporting — that AI developers need real examples of to automate.

Engineering and IT data — tickets, incident reports, technical documentation — shows how real software systems are built and maintained, data that coding-focused AI development needs.

HR and recruiting data records how organizations evaluate candidates, onboard employees and manage people operations — a valuable AI data category when rigorously anonymized.

Contracts and legal workflow data capture structured, high-stakes professional reasoning, making them among the most valuable categories of company data for AI.

Domain-specific data captures the vocabulary, processes and judgment of a particular field. General web data cannot substitute for it, which is why AI labs and data users license it from companies.

A data-licensing transaction typically takes from several weeks to a few months, depending on how ready the company's data is and the buyer's evaluation timeline.

Training is when a model learns from data; inference is when it uses what it learned. Company data serves both — as training examples and as evaluation and grounding material.

AI research consistently shows that smaller, high-quality datasets can outperform larger, noisier ones — which is why real, well-documented company data commands value.

Data structuring converts raw system exports into organized, documented datasets that buyers can evaluate and use. It is a standard part of how SourceX prepares company data.

Because models need fresh data as the world changes, AI labs and data users increasingly structure licensing as ongoing supply relationships rather than one-time purchases.

A data assessment inventories what data a company holds, where it lives, its condition and its potential value to buyers. It is the first step in every SourceX engagement.

As AI products are sold into enterprises, they must understand real business work — which requires training and evaluation data from real businesses.

After a company applies, SourceX reviews the application and assesses the data for selection. As the company's direct licensing partner, SourceX licenses approved business data for AI training and evaluation and manages the agreement, rights review, preparation, delivery and payout, subject to acceptance at Closing and agreed terms.

Companies selling data directly to buyers face legal exposure, privacy risk, pricing disadvantage and buyer-vetting burdens. SourceX licenses approved business data directly from companies for AI training and evaluation, managing the agreement, rights review, data preparation, delivery and payout, subject to acceptance at Closing and agreed terms.

A SourceX referral partner introduces qualifying companies to SourceX, their direct licensing partner for approved business data used in AI training and evaluation. Partners earn 25% of the Actual Supplier Payout: eligible licensing compensation SourceX actually pays the referred company, excluding separately itemized taxes and expense reimbursements. A reward becomes payable after SourceX has paid the referred company under its signed agreement and the payout conditions in the program terms are met, including finance approval and your tax and payment details. The cumulative cap is $100,000 per referred company. SourceX funds rewards separately; they do not reduce the company's payout.

Alignment data teaches AI models to behave helpfully, honestly and safely. It is built from human judgment and preferences, which AI developers purchase from specialized suppliers.

Internal messaging archives show how teams actually coordinate, decide and solve problems together — a category of work data that exists nowhere outside companies.

AI labs and data users are the organizations that purchase licensed data for model training and evaluation: AI labs, enterprise AI teams and data providers. SourceX licenses approved business data directly from companies and supplies licensed datasets to SourceX and data users under agreed terms. SourceX's current data needs and agreed licensing scope remain part of the process.

Referral rewards are 25% of the Actual Supplier Payout: eligible licensing compensation SourceX actually pays the referred company, excluding separately itemized taxes and expense reimbursements. Rewards are capped at $100,000 cumulatively per referred company, subject to any signed Schedule A or addendum. A reward becomes payable after SourceX has paid the referred company under its signed agreement and the payout conditions in the program terms are met, including finance approval and your tax and payment details. SourceX funds rewards separately; they do not reduce the company's payout.

Qualifying companies typically have 50 or more full-time employees at peak headcount, several years of accumulated operational data, and clear authority to license it.

AI data demand grows because more models are being built, more applications need domain data, and legal pressure is shifting buyers from scraping to licensing.

A data pipeline is the path data takes from a company's systems to the buyer: extraction, anonymization, structuring, documentation and secure delivery.

Rare and edge-case examples — unusual problems, exceptional situations — appear only in long real operational histories, which is why they are valuable for AI training.

Model evaluation tests whether an AI system actually performs real tasks correctly. It requires real task-and-outcome data, which companies hold and buyers purchase.

Operations and logistics data records how companies coordinate physical work — scheduling, routing, fulfillment, exception handling — a category with almost no public equivalent.

Data exclusivity defines whether a buyer receives sole access to a dataset or whether it can be licensed to others. It is the largest single driver of AI data deal value.

Most partners' existing networks already contain qualifying companies. The skill is recognizing the profile: 50+ employees, years of history, documented operations.

AI training data is the material models learn from. As public web data is exhausted and scraping faces legal pressure, training data is increasingly licensed from companies.

Business email archives record years of real professional communication — negotiation, support, coordination and decision-making — at a scale no other source matches.

A data license agreement defines what data is covered, what the buyer may do with it, how it is protected, what is paid and how long the rights last.

Marketing data — campaign records, content performance, customer journeys — shows what actually works in real markets, data that AI marketing products need.

The simplest introduction: companies hold data AI labs and data users pay for, SourceX handles the entire process, and assessment costs nothing. The partner tools draft the message for you.

Data monetization means earning revenue from data a company already produces. For most established companies, licensing data to SourceX is the most practical path.

A company can only license data it has clear authority over. Rights review happens early in every SourceX transaction so deals are built on solid ground.

RLAIF (reinforcement learning from AI feedback) uses AI-generated judgments to train models. It reduces some human data needs but still depends on real data as its foundation.

AI models trained on old data lack knowledge of the current world. Fresh data keeps models current. SourceX usually pays a one-time fee for the dataset in the agreement; some agreements set recurring deliveries, each paid as the agreement states. There are no royalties or usage fees.

Foundation models are large, general-purpose AI models that other products are built on. Their development consumes the largest share of training data spending.

Data referral programs pay partners when companies they introduce complete qualifying transactions. SourceX pays 25% of the Actual Supplier Payout—the eligible licensing compensation it actually pays the referred company, excluding separately itemized taxes and expense reimbursements—capped at $100,000 cumulatively per referred company, subject to any signed Schedule A or addendum.

Personal data is any information that can identify an individual. SourceX de-identifies it, through approved processing routes, or it is excluded, before the data is used or shared.

AI models must perform across industries, languages and situations, so buyers pay for dataset diversity — data from many kinds of companies, not just volume from a few.

Data licensing transactions use controlled access, encrypted transfer and contractual handling obligations. Security requirements are written into the license agreement itself.

Data licensing requires volume and multi-year history. limited-history startups rarely have either, which is why the qualifying profile is established companies with 50+ full-time employees at peak.

Data valuation weighs SourceX's current data needs, volume, history, quality, exclusivity potential and rights clarity. A SourceX assessment gives a company a grounded picture of what its data could be worth.

Web data shows what people choose to publish; operational data shows what people actually do at work. AI labs and data users pay for the difference.

Every partner receives a unique referral link. When a company applies through it, the first valid introduction within the 90-day window is credited once the company submits a verified application; duplicates are reviewed. You can track the referral's progress in your partner portal.