How company data is valued: volume, coverage and connection depth

Founders who ask what their company data is worth usually expect the answer to depend on size: more messages, more files, more money. Size matters, but it is only one of three questions a buyer asks. The other two, how many systems the archive covers and how tightly those systems connect, often matter more.

It is tempting to think of data like a commodity, priced by the terabyte. That model breaks down quickly for operational data. A shared drive full of exported reports and duplicated slide decks can be enormous and still teach an AI agent very little. A modest archive where every decision can be followed from discussion to outcome can be far more useful.

AI labs license company data to train agents and build evaluations that look like real work. What they are paying for is not storage. It is evidence of how people inside a business plan, coordinate, make mistakes and finish tasks. Volume, coverage and connection depth are three ways of measuring how much of that evidence an archive contains.

Typical licences range from $20K to $400K+. Where a given company lands depends on the factors below, which is why we never give a firm number before the data has been counted.

Volume is the simplest factor. It covers the amount of material in each system, but in practice it is shaped by two things that are easy to reason about:

• Years of history. A company that has run on the same tools for six years has recorded more complete projects, more hiring cycles, more product launches and more customer relationships than one that has run on them for eighteen months.
• Headcount. More people means more conversations, more tickets, more pull requests and more handoffs between teams. A 150-person company generates a richer record than a 12-person one over the same period.

Volume has diminishing returns when it is repetitive. Ten thousand automated alert messages in a monitoring channel count for little. Ten thousand human messages in which a team works through real problems count for a lot. Evaluation looks at the substance, not just the row count.

Coverage measures how many parts of the business the archive touches. A company's work is spread across tools, and each one captures a different slice:

• Chat (Slack or Teams) shows the informal coordination and quick decisions
• Email shows external relationships with customers, vendors and partners
• Docs and wikis show plans, specs and how knowledge is written down
• Tickets and projects (Jira, Linear, Asana) show how work is broken down and tracked
• Code and pull requests show how software is actually changed and reviewed
• CRM and customer support show how the company sells and serves customers
• Finance workflows show approvals, invoicing and close processes

An archive with one system is useful for narrow tasks. An archive with five or six systems lets a lab study how work moves across an entire company. We look at this in more detail in six systems that make a dataset valuable.

Connection depth is the factor that surprises people, and it is often the largest driver of value. It asks a simple question: can the same piece of work be traced across several tools?

Imagine a 60-person logistics software company. A customer reports a billing bug through the help desk. A support lead flags it in a Slack channel. An engineer opens a Jira ticket, links it to a GitHub pull request, and a reviewer leaves comments. A product manager updates the release notes in the wiki, and the account manager emails the customer to confirm the fix. If all of those systems are in the archive, that is one connected story with a clear start, a set of decisions and a verifiable outcome.

That kind of story is what agents need to learn from. A model trained only on the Jira ticket sees a task description. A model trained on the whole chain sees why the task existed, who was involved, what trade-offs were discussed and whether it worked. This is the heart of why connected data beats big data.

Connection depth depends on anonymization being consistent. If the support lead is PERSON_014 in Slack but a different token in Jira, the chain breaks. That is why our pipeline gives every person, company, address, email, phone and account the same token across every tool. The links survive, and the identities do not.

Some factors reduce what an archive is worth, even when it looks large. Most of them come down to gaps in the record:

• Systems that were deleted or cancelled before the evaluation
• Large exclusions that cut out the most connected parts of the work
• Long periods where the team worked in tools that are no longer accessible
• Content that is mostly automated, duplicated or boilerplate

Exclusions are always the seller's call, and some are necessary. HR, payroll, medical and legally privileged folders are excluded by default, and sellers can add more, such as a specific client or channel. The trade-off is simply that each exclusion removes some of the record.

Valuation happens during a read-only evaluation, after the seller has set scope and exclusions and those have been annexed to the contract. The evaluation takes under an hour of the seller's time. We count what is in scope across each system, look at history and headcount, and check how much of the work can be traced across tools. Only then is there a real figure to discuss.

If you want a rough sense before talking to anyone, the value estimator gives a rough range from a few questions about your company. It is a starting point, not a quote, and it can help you think about which of the three factors is likely to matter most in your case.

When you are ready for an actual number, book a 15-minute call. We will tell you what an evaluation would look at, and you decide whether to go further.