What drives the value of a dataset
Five things move the number more than anything else: years of history, volume, outcome labels, modality, and rights. How each one changes what a buyer will pay, and what you can do about it before going to market.
Owners ask the same question in the first call: what is this worth? The honest answer is a range, and the range is wide, because the same company can be worth very different amounts to different buyers. But the variables that set the range are not mysterious. Five of them move the number more than everything else combined.
Here they are, in roughly the order a buyer reads them on a manifest.
Years of history
Depth is the first thing a buyer checks, and the one thing you cannot manufacture. A dataset covering six years of the same workflow shows how a process behaves across staff turnover, product changes, busy seasons, and at least one bad quarter. Two years shows a snapshot.
Continuity matters as much as length. Ten years with a helpdesk migration in the middle that dropped the first seven is a three-year dataset. Say so on the manifest; buyers will find it at sample review anyway.
The qualification we use starts at three years. Below that, the right advice is usually to wait and keep the records.
Volume
Volume is the number of records, and it scales with headcount and time. Twenty desk staff producing tickets, deals, threads, and documents for five years generate a body of work large enough to learn from. Five people for one year do not.
Buyers think about volume per workflow rather than in total. Forty thousand support tickets is a usable set on its own; four thousand tickets plus four thousand emails plus four thousand files is three thin sets. Concentration in a few deep workflows generally prices better than a little of everything.
The reported Spirit Airlines data sale is a useful public reference for how volume gets described. The asset was reported as roughly 100 million emails, 500 million chat messages, and 30 million lines of code. Most companies are two or three orders of magnitude smaller, and that is fine. The point is that buyers read volume in those terms, per system, and a manifest should give it to them that way.
Outcome labels
This is the driver that separates two otherwise identical companies. A record with an outcome shows not just that work happened but how it ended. A ticket with a resolution code and a satisfaction score. A deal marked closed-won or closed-lost with a reason. A dispute with a ruling. A code review that was merged or rejected, with the comments attached.
Buyers pay for outcomes because models learn from them. A million messages with no resolution teach a model how people talk. Fifty thousand tickets with resolutions teach it how problems get solved. The second is rarer and more valuable.
If your systems already capture outcomes, make sure the manifest says so, per system, with a rough share of records that carry them. If they do not, start now. It costs nothing and changes the price in two years.
Modality
Modality is the form the records take. Text is the baseline: tickets, email, chat, documents. Structured records such as CRM stages and billing events add value when they can be joined to the text. Audio, meaning recorded calls, is scarcer and prices higher when consent exists and transcripts are available. Code and engineering workflows are a distinct category with their own buyers.
Two things matter here. First, the mix: a company with tickets, the chat where the ticket was discussed, and the call that preceded it has something a single-channel export does not. Second, usability: a native export with attachments is worth more than a lossy one, and audio without transcripts is a project rather than a dataset.
Rights
Rights are the quiet driver. A buyer who is unsure whether you can license a set will discount it heavily or walk away, regardless of how good the records are. Three questions decide it.
What do your customer contracts and privacy notices say about using records for purposes like this? What do your employee policies say? Is any of the content licensed from someone else, such as a vendor's documentation or a partner's data, and therefore not yours to license?
Most companies can license most of what they have, with specific exclusions. The value comes from knowing which, in writing, before the manifest goes out. An unresolved rights question is the most common reason a strong manifest draws a weak number, and it is the cheapest to fix.
How the five combine
A buyer reads years and volume to decide whether there is enough to learn from. Labels and modality decide how much can be learned. Rights decide whether it can be used at all. The strongest sets are deep, concentrated in a few workflows, labeled, multi-channel, and clean on rights. Most companies are strong on two or three of the five and can improve a fourth before going to market.
What you cannot do is invent any of them. Do not backfill records, do not relabel history, do not guess at rights. Buyers verify at sample review, and a manifest that fails verification is worth less than an honest smaller one.
Why pricing comes before market
Because the five drivers are legible on a manifest, a set can be benchmarked against what vetted buyers are paying for comparable records before any buyer sees your name. That is how a first offer, an indicative and buyer-benchmarked range, comes back within 24 to 48 hours of a complete manifest. Binding terms follow sample review, and the range is a process target rather than a guarantee.
If you want a rough sense of where your company lands, the estimator on this site takes the first three drivers as inputs and returns an illustrative range. For a real number, the manifest is the next step.
Questions worth asking
Which of the five drivers matters most?
Outcome labels, in most cases. Two companies with the same years, volume, and systems can draw very different prices if one set of tickets carries resolution codes and satisfaction scores and the other only shows that messages were exchanged. Buyers are paying to learn how problems are decided and how decisions turn out, and labels are what make that learnable. Years and volume set the floor; labels and rights set how far above it the number goes.
My records are in one system and only go back two years. Is it worth asking?
Probably not yet, and we will say so rather than waste your time. The qualification we use is 20 or more desk staff, three or more years of records, and data in three or more systems, which is roughly where buyers start paying. The useful move now is to keep the systems you have, stop deleting history, make sure tickets and deals are closed with an outcome, and check what your customer contracts say about data use. In a year or two the same company is a different manifest.
Can I raise the value of my data before going to market?
Within limits, yes. You cannot add years, and you should never fabricate or backfill records. You can make sure what exists is exportable, that outcomes are recorded where they already happen, that transcripts exist for recorded calls, and that the rights position is documented rather than assumed. Resolving an unclear rights question with counsel before the manifest is often worth more than any other preparation, because it removes the discount a buyer applies to uncertainty.
Measured from a complete data manifest. A first offer is a process target, not a guarantee; buyers decide eligibility, price, and terms. Indicative value ranges mentioned on this site are illustrative, not quotes. Nothing here is legal advice; have your own counsel review any agreement before data moves.