What is your data worth?
A framework for understanding data assets, without assigning a price before the evidence is reviewed.
The framework
Value starts with evidence, not a figure.
Years of records, expert decisions, and private code describe how work in your field is done. Teams that train AI models and agents look for that kind of material.
What any asset is worth depends on what it is, who could use it, and what you have the right to share. This page sets out those three questions as a framework. It names no prices, because price is set in the market.
The idea behind it is Canon IV: your data is worth more than you think. Knowing what you hold comes before deciding what to do with it.
Asset categories
Start with the asset category
Most business data falls into one of these categories. The buyer types shown are typical, not promised.
| Category | Examples | Typical buyer type |
|---|---|---|
| Operational records | Structured transactions, schedules, and workflow outcomes | Vertical software teams |
| Decision and correction histories | Approvals, rejections, and corrections with the reason recorded | Model builders |
| Documents and expertise | Procedures, templates, and reviewed drafts that capture judgment | Model builders |
| Evaluation material | Labeled tasks with acceptance criteria and checked outcomes | Teams that evaluate models and agents |
| Industry observations | Rights-cleared trends, readings, and measurements | Research teams |
| Code and engineering history | Fixes, tests, and change histories in repositories you own | Teams that train coding agents |
| Models and agents | A model or governed agent with accepted results on your own work | Industry peers |
One business often holds several categories. A value assessment reviews each one separately, because rights and quality can differ between them.
The last category works differently. A model or agent is offered only after it has accepted results on your own work. An agent keeps its approval points, so a named person still reviews its consequential actions.
Four uses
Four ways an asset can create value.
The first use needs no outside party. The other three need clear rights and your approval of every term.
- Internal useYour own AI work: a model trained on your records, or an agent evaluated against your past decisions.
- Model training partnersTeams that train or evaluate models receive an approved package for a defined use, under terms you set.
- Data licensingA de-identified, documented dataset is licensed for stated uses, with provenance and use limits in the terms.
- ProductsA model or agent built and accepted on your own work is offered to others in your industry. Agents keep their approval points, so a named person still reviews consequential actions.
Buyer types
Who might use it
Proposals name buyer types, never promised buyers. You approve every placement.
Model builders
Task-specific examples and decisions with clear rights and provenance.
Vertical software teams
Structured information that supports a defined product workflow.
Research teams
Well-documented observations with permitted research uses.
Industry peers
Data, models, or agents that answer a concrete operating question.
Value drivers
What drives value
Six qualities decide whether an asset suits a use. Rights and consent come first, because nothing else matters without them.
Rights and consent
A documented right to use and share each asset, with client permissions and consents recorded. Without it, an asset stays out of any package.
Quality and structure
Consistent fields, few gaps, and records kept the same way over time. Well-kept data needs less preparation before any use.
Provenance
Where each record came from, when it was collected, and how it was processed. Any item in a package should trace to its source.
Uniqueness
Information that is distinctive and relevant, such as expert decisions that public sources rarely capture.
De-identification
Personal and client identifiers removed according to your privacy decisions, with the removal checked on samples.
Freshness and volume
Records recent enough for the intended use, in a quantity that suits the task. The right amount depends on the use, not on size alone.
What stays out of a package
Some assets stay private whatever their potential value. Records combined together can reveal more than any one of them, so packages are checked for combinations too.
- Assets without a documented right to use and share them.
- Records whose identifiers cannot be removed to meet your privacy decisions.
- Material your client contracts restrict, until permission exists.
- Anything that reveals pricing, methods, or client relationships you mark private.
Before you ask
Six questions to ask about each asset.
Your answers will not set a value. They show where an assessment would start, and what needs clearing first.
- Who created it, and under what agreement?Client contracts, employment terms, and vendor agreements decide what you may share.
- Does it hold personal or client identifiers?Identifiers must be removed to meet your privacy decisions before any package is built.
- Has it been kept the same way over time?Consistent fields and methods make records easier to document and to use.
- Could a public source supply the same thing?Distinctive material, such as expert decisions with reasons, tends to suit licensing better.
- Is it still being created?An asset that grows each month can support a refreshed package, where license terms allow it.
- Would sharing it reveal how you win work?Pricing, methods, and client relationships may be worth keeping private, whatever their potential value.
What to gather first
None of this needs to be complete. A rough version is enough to scope the work.
- The findings from Data Protection & Readiness, or your own rights review.
- A view of the data, code, models, or agents you would consider offering.
- A list of anything that must stay private.
- The person who will approve license terms.
- Any requests for your data you have already received.
The assessment
How a data value assessment works.
It documents your assets and their options. It does not assign a price or promise a buyer.
- Confirm rightsStart from the readiness findings, or your own rights and consent review.
- InventoryDocument structure, quality, freshness, and provenance for each candidate asset.
- Match formsMatch each asset to the forms that are licensed, with a defined use and buyer type.
- Owner approvalYour named owner decides what may be packaged, for whom, and on what terms.
- Package or keepLicense the approved package, build a product first, or keep it private.
Packaging and placement through Data & AI Monetization carry a fixed fee, with an optional revenue share where it fits. Your counsel approves the final license wording.
What the assessment gives you
A documented view of each asset and its options, so your owner can decide on evidence. The same documentation supports your own AI work if you keep everything private.
- The assets in scope, with the rights status of each.
- Structure, quality, freshness, and provenance, recorded per asset.
- The forms each asset could take, with a defined use.
- Typical buyer types and proposed license scope for your review.
- The assets excluded, and the reason for each exclusion.
Assets by industry
Where valuable assets often sit.
Typical applications across industries. They show where the framework applies, not past client work or results.
- Software company
Resolved defects as coding tasks
Fixes in repositories the company owns outright can become verifiable coding tasks for teams that train and evaluate agents. Secrets and client identifiers are removed first.
- Medical billing
Correction histories with their reasons
Coding and claim-correction decisions, each with a documented reason, can serve as evaluation data. Where contracts and consent allow, they are de-identified before packaging.
- Manufacturer
Inspection images and decisions
Labeled inspection images and the accept-or-reject decisions made on them can support teams training vision models. Customer part designs stay excluded.
- Construction
Estimating judgment
Judgment calls on takeoffs, allowances, and bids can shape a training environment, built only from rights-cleared project records. Client drawings and pricing stay excluded unless contracts allow.
- Energy services
Readings paired with diagnoses
Equipment readings paired with technician diagnoses and repairs describe field work in detail. Where customer contracts allow, they are de-identified and documented with known gaps.
- Education provider
Graded assessments
Assessments with rubrics and instructor grades can help teams evaluate AI reasoning within a subject. Learner consent comes first, and learner identities are removed.
In each case a named owner approves the package, the buyer types, and every term before anything is shared.
Get the full guide
Use the guide to describe your asset, review value factors, and prepare an evidence brief.
Questions
Can you put a number on our data?
Not as a figure. We document what you hold, the forms it could take, and the buyer types that license them. Price is set in the market, and we never promise a revenue number.
Does a larger dataset mean more value?
Not by itself. A smaller set with clear rights, consistent structure, and documented provenance can suit a defined use better than a larger one without them.
Do we need Data Protection & Readiness first?
Rights, consent, and privacy must be cleared before packaging, usually through Data Protection & Readiness. If you have completed your own review, we start from its findings.
What if the most valuable asset reveals how we win work?
We flag assets that reveal pricing, methods, or client relationships for your decision. Anything you mark private stays out, whatever its potential value.
How is the fee structured?
Data & AI Monetization is a fixed fee for packaging and placement, with an optional revenue share where it fits. Full engagement terms are finalized in a Master Services Agreement.
Do our clients need to agree?
Where your contracts or privacy obligations require permission, yes. The rights review identifies those cases, and any asset without clear permission stays out.
Can we start with one asset?
Yes. A single asset can be assessed, packaged, and approved on its own. Further assets can follow from the same rights review.
Will buyers see our raw records?
Buyers receive only the approved package, under the terms you set. Your records stay in your systems, and packaging works from approved copies.
Who approves the license wording?
We propose scope, use limits, and provenance terms for your review. Your counsel approves the final wording, and your named owner approves each license.
Can we use the assessment for our own AI work instead?
Yes. The documentation of rights, quality, and provenance supports your own models and agents. Model Training & Fine-Tuning can build a model on the same records.