The Hidden Asset on Every Balance Sheet: How Operational Data Becomes AI Revenue

Accountants have well-established ways to value a company's buildings, equipment, receivables and brands. There has never been a standard line for the record of how a company's people actually work: the procedures they follow, the decisions they make, the problems they solve and the outcomes they achieve. For most businesses, that record has been a cost. It takes storage, it needs securing and it rarely leaves the building.

AI has changed the economics. Models trained on the public web are good at general tasks, but they struggle with the specific, judgement-heavy work that fills a real business day. To learn that work, they need examples drawn from it. That is why buyers now describe a company's operational records as a revenue source. micro1's announcement (opens in a new tab) of a $1 billion, 12-month commitment frames it directly: the initiative gives businesses a new revenue stream from the operational intelligence they have built over years.

The clearest evidence that accumulated records can become revenue comes from companies that have already done it in public. When Reddit filed to list its shares in February 2024, its registration statement (opens in a new tab) described data licensing as a new line of business. The filing disclosed agreements with an aggregate contract value of $203.0 million and terms of two to three years, and said the company expected to recognise at least $66.4 million of that during 2024.

On the same day, Google announced an expanded partnership with Reddit (opens in a new tab) that gave it access to Reddit's Data API, describing the platform's content as a source of real-time, structured and authentic human conversation. The announcement framed the arrangement as a way to improve both Google's products and its AI models, which is exactly how buyers describe the value of operational records: genuine human work, captured in context.

Stack Overflow's partnership with OpenAI (opens in a new tab), announced in May 2024, followed the same pattern for developer knowledge. Years of technical questions, answers and peer review became a licensed resource, delivered through Stack Overflow's OverflowAPI, with attribution back to the community that created it.

These were large public platforms, but the principle carries over directly to private companies. In each case the value lay in a long, well-organised history of skilled people solving real problems. That description fits the support desks, engineering teams, claims departments and operations functions of many established businesses. The difference is that operational records are usually more specialised, which makes them harder for buyers to find anywhere else.

The structures also offer a useful reference. Multi-year terms, defined access and clear attribution are now familiar features of data agreements, and they give private owners a recognised template to work from rather than a blank page. For a board weighing whether its records could become an asset, the public precedents show that the model is proven, contractual and recurring.

The good news for owners is how familiar the qualifying records are. The buyer programmes in the market publish their lists, and they read like an inventory of a well-run company's shared drives. micro1's company data page (opens in a new tab) names six industries and the records it wants from each.

Beyond the functional records, buyers place particular value on two less obvious categories. The first is decision-making patterns: evidence of how teams weigh information and choose a course of action. The second is human feedback on AI outputs, which many companies now generate as a by-product of using AI tools at work. When staff correct, rate or rewrite an AI system's output, they are producing exactly the kind of signal developers use to improve models.

Mercor's data programme (opens in a new tab) describes the same opportunity from its side, telling businesses that the data they generate every day is exactly what frontier labs need. When several buyers describe the same records in similar terms, it confirms the demand is broad and lasting rather than a passing request from one developer.

Developer knowledge shows how quickly this category has matured. In May 2024 Stack Overflow and OpenAI announced a partnership (opens in a new tab) under which OpenAI uses Stack Overflow's OverflowAPI product to give its users accurate, vetted technical knowledge, with attribution to the Stack Overflow community inside ChatGPT. Years of questions, answers and accepted fixes became a licensed product with a named commercial channel.

AI development is moving from chat towards agents: systems that carry out multi-step tasks such as resolving a support case, processing an invoice or reviewing a contract. Training an agent requires more than text. It needs to see the whole sequence: the starting situation, the information gathered, the choices made, the actions taken and whether the result was good. Operational records are one of the few sources that contain that full sequence.

micro1 explains how it uses this material: it builds reinforcement learning environments from de-identified company data, where AI models and agents learn to navigate workflows, make decisions and complete complex tasks. In its own words, real business operations involve incomplete information, competing priorities and exceptions that require judgement, and training AI to handle that complexity requires environments that preserve the context behind how businesses work.

• A lone documentUseful background, but it shows no decision or outcome. Lower value on its own.
• A document with its historyShows how a piece of work changed through review and approval. Higher value.
• A complete case recordInput, reasoning, action and verified outcome together. The highest value, because it teaches a full loop of skilled work.

This is why a modest company with long, well-documented case histories can hold a more interesting dataset than a much larger one with scattered files. Depth and completeness count for more than raw size.

Licensing operational data does not mean opening the company's systems to a stranger. The leading programmes describe a staged, permission-first process in which the owner sets the rules before anything is shared. micro1 publishes four steps: an initial assessment, an evaluation of fit, a discovery call covering the data and privacy requirements, and a partnership agreement.

micro1 adds several specific protections. Personal information is removed where relevant, access is limited to a small number of authorised staff, and each customer's data runs through isolated pipelines so it is not mixed with other engagements. Some datasets go through synthetic rewrites that keep the structure of a workflow while loosening direct ties to the original records. Throughout, the company states that owners keep ownership of their underlying data.

Pricing reflects the same qualities that make data useful. micro1 lists the factors it weighs: dataset size and volume, workflow complexity, data quality, the domain expertise represented and the uniqueness of the operational knowledge contained. Its published tiers start at $100,000 for qualified enterprise data partnerships, rise to $500,000 and above for large-scale datasets or ongoing participation across several teams, and reach $1 million and above for highly unique proprietary data with significant value for frontier AI.

Qualified enterprise partnership

$100k+

Large-scale or multi-team participation

$500k+

Highly unique proprietary data

$1M+

Eligibility is practical rather than exclusive. micro1 looks for operationally mature companies with more than 30 employees, established documentation and high-quality operational records, with current demand strongest for companies in the United States, the United Kingdom and Canada and documentation in English. Companies outside those markets can still be considered, depending on the needs of active projects.

Ongoing participation can add to the value over time. A company that contributes across several teams, or keeps contributing as new records are created, moves from a one-off licence towards a recurring relationship, which is the structure that supports the upper tiers.

When a buyer assesses operational data, it is looking for signs that the records will teach a model something reliable. The clearest sign is structure. Records held in a consistent system, such as a ticketing tool, a CRM or a code repository, already carry dates, owners, statuses and links between related items. That structure lets a buyer follow a piece of work from start to finish without guessing, which is exactly what agent training requires.

The second sign is a recorded outcome. A support ticket marked resolved, a contract that was signed, an invoice that was approved or a bug that was closed tells the buyer how the work ended. Outcomes turn a set of documents into a set of examples, because they show which actions led to a good result. Companies that record outcomes as a matter of habit often hold more valuable data than they realise.

• Consistent systemsRecords kept in the same tools over time, with dates and links intact.
• Recorded outcomesA clear end state for each case, such as resolved, approved or closed.
• Time coverageSeveral years of history, showing how processes and decisions changed.
• Expert authorshipRecords written or reviewed by experienced staff, not just generated automatically.

Time coverage adds depth. A few months of records show how a process works today. Several years show how it handled exceptions, adapted to change and improved. That range of situations is valuable because real business work is full of exceptions, and models need to see many of them to handle new ones well.

Expert authorship completes the picture. Records written or reviewed by experienced professionals carry judgement that is hard to find anywhere else. This is where established companies hold a natural advantage over newer ones: their archives reflect years of skilled people solving real problems, and that accumulated judgement is the quality AI developers are working hardest to capture in their models.

None of these qualities requires a company to change how it works. They are the by-products of running a well-organised business. The task for an owner is to recognise them, document them and present them clearly, so that a buyer can see the value before any data is shared.

Many companies are already preparing their data for internal AI projects: building assistants for their own teams, automating workflows or improving customer service. The work involved, cataloguing data, cleaning it, protecting personal information and documenting its origin, is largely the same work required to license it externally.

That overlap means the two paths reinforce each other. A company that prepares its data well for its own use creates an asset that can also be offered to outside buyers, on terms it controls. Conversely, the discipline of preparing data for an external buyer, who will ask hard questions about quality and rights, tends to improve the data for internal use as well.

There is a strategic dimension too. Some organisations will choose to keep their most distinctive data entirely in-house, using it to build their own competitive advantage. Others will license some datasets while keeping others private. Making that choice deliberately, with a clear view of what each dataset is worth inside and outside the business, is one of the more important data decisions a leadership team can make.

In each case the starting point is the same: knowing what the company holds and what it can prove about it. Organisations that invest in that understanding create options for themselves, whichever path they choose.

It also changes how data is treated across the business. When teams understand that the records they create may carry value beyond their immediate use, they tend to capture them more consistently, store them more carefully and document them more thoroughly. Over time, that discipline compounds, and the asset grows in both size and quality.

The financial framing changes as well. A dataset that once appeared only as a storage cost can be assessed like any other asset: what it would earn under licence, what preparation it needs, what risks attach to it and how its value might grow as more material is added. Treating data this way gives finance teams and boards a clearer view of an asset that has historically been invisible in the accounts.

For acquirers and investors, the same lens is becoming part of how businesses are valued. A company that holds unique, well-documented and licensable data carries an option on a new revenue stream, and that option is increasingly recognised in strategic discussions. Owners who can demonstrate the quality and rights position of their data are presenting a fuller picture of what their business is worth.

Owners who want to explore the opportunity can do much of the groundwork internally. The steps below mirror what buyers assess, so time spent on them pays back in a smoother and better-valued process.

1. 01Map the recordsList the systems that hold operational records: ticketing, CRM, document management, code repositories, finance tools.
2. 02Check completenessIdentify where records show the full loop from input to outcome, and how many years they cover.
3. 03Confirm rightsCheck customer contracts, employment terms and privacy notices so the company knows what it can license.
4. 04Plan redactionDecide what must be removed or anonymised, including personal information and commercially sensitive details.
5. 05Set internal approvalsAgree who signs off, so the company can move quickly once a buyer is interested.

The opportunity is real, well funded and still early. The records a company has built up through years of skilled work have a market for the first time, and the owners who organise them carefully are best placed to benefit. The final study in this series explains why clean rights and consent push a dataset to the top of the price range.