Compliance first: We never take HIPAA/PHI data or personal customer data. Everything is de-identified before it leaves your control.
Home / Insights / Analysis
Analysis

The Small-Business Data Dividend: Why Your Operational Records Are Worth Money Now

The Small-Business Data Dividend: Why Your Operational Records Are Worth Money Now

Small businesses are sitting on years of email, tickets, invoices and job records that AI builders cannot scrape from the web. Here is what the data dividend is, why it exists now, and how a small company can collect it without giving up customer privacy.

For most of the internet era, the businesses that got paid for data were the giant platforms: social networks, search engines, publishers with archives going back decades. The licensing deals reported since 2023 mostly fit that pattern. But the next wave of AI is about doing work, not just writing text, and the records that show how work really gets done live inside ordinary companies. That shift creates what we call the small-business data dividend: a one-time, time-limited chance for owners to be paid for the operational history they already have.

By OpenDataPolicies Editorial · October 20, 2026 · 7 min read

What the data dividend actually is

The data dividend is the value an AI or machine-learning builder will pay for a licensed, de-identified copy of the records your business generates just by operating. It is not a new product you have to build. It is the exhaust of the work you already do:

  • Email and team chat showing how requests arrive, get clarified, escalated and closed.
  • Support tickets with the question, the back-and-forth and the resolution.
  • CRM and project records showing how deals and jobs move from stage to stage, and where they stall.
  • Accounting transactions showing the rhythm of a real business: billing cycles, categories, seasonality.
  • Industry software such as dispatch boards, matter management, repair orders or work orders.

Each of these is an example of what we call operational data for machine learning. None of it is the kind of thing that shows up on a public web page, which is exactly why it is scarce.

The word "dividend" matters. A dividend is a payout on something you already own. You built these records over years of operating; licensing a de-identified copy lets you collect on that work without selling the business or changing how you run it.

Why the dividend exists now and not five years ago

Three forces have converged.

First, public web text is being used up. Researchers at Epoch AI estimated the stock of quality-adjusted, human-generated public text at roughly 300 trillion tokens and projected that, if trends continue, models will be trained on all of it sometime between 2026 and 2032 (Epoch AI). OpenAI co-founder Ilya Sutskever put it more bluntly at NeurIPS in December 2024: "We've achieved peak data and there'll be no more" (The Verge).

Second, the open web is closing. The Data Provenance Initiative found that in a single year (2023–2024), roughly 5% or more of all tokens in the widely used C4 dataset, and 28% or more of its most actively maintained critical sources, became fully restricted by robots.txt (arXiv: Consent in Crisis). When the free supply shrinks, licensed supply gets more valuable.

Third, AI is moving from chat to agents. Sutskever predicted in the same talk that next-generation systems will be "agentic in real ways." An agent that books jobs, answers support tickets or reconciles invoices needs examples of how those tasks are done in real businesses. That is operational data, and it sits overwhelmingly inside private companies. We go deeper on this in operational data vs web data.

Put those together and you get a buyer's market for public text and a seller's market for private, workflow-shaped data.

Why small companies are not too small

A common assumption is that only enterprises with millions of records have anything worth licensing. In practice, buyers care about several things besides raw volume:

  • Uniqueness. A 12-person HVAC company's dispatch history shows patterns no web crawl contains. Uniqueness is often worth more than size. (See how buyers price data.)
  • Depth over time. Five or ten years of continuous records show seasonality, staff turnover and process change.
  • Vertical coverage. Builders training models for a specific industry need many examples from many similar businesses. A small firm in a well-defined vertical can be exactly the slice they are missing.
  • Clean structure. Small companies often run on one or two well-known systems such as QuickBooks, ServiceTitan or Google Workspace, which makes export and de-identification straightforward.

According to founder Jeff Cline, small companies have gotten over $100,000 for their operational data. Results vary; not a guarantee. The point is not that every business will see that figure; it is that size alone does not rule you out.

Why the dividend is time-limited

The dividend is not permanent, and that is the part most owners underestimate.

Training data is most valuable before a model has learned what it teaches. Once a set of models has been trained on, say, thousands of examples of how a dental front office handles rescheduling, the next thousand examples add less. Buyers will still want fresh data, but the premium for being early fades. Founder Jeff Cline's shorthand for this is "unique data is king": whoever holds the keys to unique data profits while it is still unique.

There is also a competitive angle. If businesses in your vertical license first, the buyers' appetite for that vertical gets filled. Being early in an industry tends to beat being large later.

That is why we tell owners to at least find out what their data is worth now, even if they decide not to sell. A free bid request or a quick run through the valuation calculator costs nothing and tells you where you stand.

What you keep, and what never leaves

A data dividend should not cost you your customers' trust. The structure that makes this work is simple:

  • You license a de-identified copy. You keep the originals and keep running your systems exactly as before.
  • Personal details are removed before anything leaves. Names, contact details, account numbers and other identifiers are stripped or replaced. Our de-identification techniques explainer walks through the methods.
  • Some data is simply off the table. We never take HIPAA/PHI data or personal customer data. Everything is de-identified before it leaves your control.

For healthcare and dental practices, that means admin-side workflows only, like scheduling and billing operations, never patient records. See healthcare practices for how that line is drawn.

Before you sell, it is worth running through a data rights checklist: confirm you own or control the data, check vendor terms, and think about employee notice. This is informational, not legal advice; talk to your own counsel about your situation.

How small businesses typically get paid

The mechanics are less mysterious than most owners expect.

  1. Inventory. List the systems you use and roughly how much history each holds. The data readiness checklist helps.
  2. Header sample. Share the column names and record counts, not the content. A header sample lets buyers judge fit without seeing anything sensitive.
  3. Bids. According to founder Jeff Cline, bids typically arrive within 72 hours if the data is correct and a header sample is provided, and ODP seeks offers from multiple buyers so you can take the highest bid. Results vary; not a guarantee.
  4. Agreement and delivery. The purchase agreement sets the license terms, exclusivity, payment and delivery. Data is de-identified before it is transferred.
  5. Close. According to Jeff Cline, deals typically close in about 90 days. Results vary; not a guarantee.

Requesting a bid is free, and you can say no to any offer.

Who is buying, in general terms

The buyers are AI labs and model developers, plus companies building industry-specific AI tools and agents. We do not name them. What we can point to is the public record: Google's reported $60 million-a-year deal for Reddit data (The Verge), Reddit's disclosure of $203 million in data licensing contracts (TechCrunch), and Shutterstock's reported $104 million in 2023 revenue from AI licensing (AOL / Business Insider). Those are large platforms, not small businesses, and none of them is an ODP buyer. But they show that serious money is changing hands for data that models cannot get anywhere else.

Our view is that the next chapter of that market belongs to operational data from ordinary businesses. As Jeff Cline puts it: "Every industry is a geek away from being uberized." The businesses that hold the records showing how their industry actually works are the ones with something to sell. We expand on that thesis in every industry is a geek away from being uberized.

The bottom line for owners

If you have run a business for a few years on email, a CRM, a ticketing tool or accounting software, you probably have a data asset you have never priced. The window to price it at a premium is open now and will not stay open. Finding out costs nothing.

Start with the valuation calculator for a rough range, or go straight to a free bid request.

Our compliance promise: We never take HIPAA/PHI data or personal customer data. Everything is de-identified before it leaves your control.

Frequently asked questions

Is my small business too small to sell data?

Usually not. Buyers weigh uniqueness, history and vertical coverage as well as volume. A small company with several years of records in a well-defined industry can be valuable. Requesting a bid is free.

What is the small-business data dividend?

It is the payment a business can receive for licensing a de-identified copy of the operational records it already has, such as email, tickets, CRM and accounting history, to AI and machine-learning builders.

Do I give up my customer data?

No. We never take HIPAA/PHI data or personal customer data. Everything is de-identified before it leaves your control.

How much can a small company earn?

According to founder Jeff Cline, small companies have gotten over $100,000. Results vary; not a guarantee. Value depends on volume, uniqueness, recency and industry.

Sources

  1. Epoch AI — Will we run out of data?
  2. The Verge — Ilya Sutskever on peak data (Dec 2024)
  3. Longpre et al. — Consent in Crisis (arXiv 2407.14933)
  4. The Verge — Google cut a deal with Reddit for AI training data
  5. TechCrunch — Reddit says it's made $203M so far licensing its data
  6. Business Insider via AOL — Shutterstock's AI licensing revenue

Updated November 19, 2026.

OpenDataPolicies.org connects businesses with buyers of de-identified operational data. We are not a law firm, broker-dealer or investment adviser, and nothing on this site is legal, tax or financial advice. Statements about amounts, timing and outcomes are Jeff Cline's statements based on past transactions. Results vary; nothing here is a guarantee of any bid, price, buyer or closing. Every deal depends on the data, its quality, its rights and buyer demand at the time.