Common reasons a dataset is worth less than expected

Most sellers come to an evaluation with a number in mind. Sometimes the count supports it. Sometimes it comes in lower, and the reasons are rarely mysterious. The same handful of issues account for most of the gap between what a company feels its data is worth and what the evaluation finds.

Knowing them in advance helps in two ways. It sets realistic expectations, and some of them can still be fixed if you catch them early enough.

There is no firm number before counting, and that is deliberate. Typical licences range from $20K to $400K+, and where a company lands depends on connection depth, years of history, headcount and the number of systems. Those are measured during a read-only evaluation, not estimated from the company's reputation or revenue.

That is where many expectations go astray. A company can be well known, well funded and busy without much of its work having been recorded in a way that is useful for training agents. The evaluation counts the archive, not the company. For the full picture of how value is built, see how company data is valued.

A company that has existed for six years does not necessarily have six years of data. History gets lost in ordinary ways:

• A workspace plan or retention setting limited how much history was kept
• The team migrated tools, and the old system was shut down without an export
• A subscription lapsed during a quiet period and older data was purged
• Someone cleaned up channels, repositories or shared drives to tidy things up
• Early work lived in personal accounts that left with the people who owned them

Each of these trims years off the usable record. The sharpest drop usually comes after a shutdown, when tools get cancelled one by one to save money. We cover that timing in timing matters.

Value grows when the same project can be traced across several systems. A company that ran almost entirely on email, or entirely on one chat workspace, has a much flatter dataset than one that also used a tracker, a code host and a wiki.

Some companies did their real work off-tool. Decisions were made on calls, in meetings or in person, and the tools hold only the leftovers: calendar invites, a few follow-up notes, files with no discussion around them. That is common in services businesses and in very small teams, and it is nobody's fault. It just leaves less for an agent to learn from.

Direct messages create a version of the same problem. They are excluded unless the company chooses to include them and has the right to. In a workspace where most real conversation happened in DMs, public channels can look surprisingly quiet.

Even with several tools, the connections can be weak. If engineers never referenced ticket IDs in commits, if docs were written once and never linked from anywhere, if the CRM was updated in a batch at the end of each quarter, the systems sit side by side without describing the same work.

Buyers notice this quickly. A thousand tickets and a thousand pull requests that cannot be matched to each other are worth less than a few hundred that can. We wrote about why in multi-tool coherence.

Automated noise also dilutes a dataset. Bot alerts, notification digests and auto-generated tickets can make an archive look large while adding little that reflects human work.

Imagine a hypothetical 40-person agency with five years of Slack, a project tracker and a shared drive. On paper, that is three systems and five years. But if projects were tracked under internal codes that never appeared in chat, and files were stored by date rather than by project, an evaluator will struggle to tie a single client engagement together from start to finish. The volume is there; the connections are not.

Sellers set scope and exclusions, and that control is important. Every exclusion is the seller's call, and many are exactly right: a client with strict confidentiality terms, a sensitive channel, a folder nobody wants shared.

But some exclusions remove more than expected. Excluding the company's largest client can take out most of the engineering and support history if that client drove the work. Excluding an entire code host because one repository is sensitive removes the rest with it. A narrower exclusion, such as one repository or one client's channels, often protects the same thing at a smaller cost. It is worth talking those trade-offs through before the scope is annexed to the contract.

HR, payroll, medical and legally privileged folders are excluded by default. That rarely affects value much, because those folders are not what buyers are looking for.

If your company is still operating or has only just started winding down, a few steps protect value:

• Keep tools running until the data has been evaluated, rather than cancelling them first
• Check retention and plan settings so older history is not purged in the meantime
• Locate exports from any tools you migrated away from
• Think about whether broad exclusions could be narrowed without losing the protection you need

None of this takes long, and the evaluation itself takes under an hour of your time. Even a lower-than-hoped number is still a recovery from an asset that would otherwise be deleted, and payment goes to the company entity in two to four weeks.

If you would like a straight read on your company's data before deciding anything, book a 15-minute call.