How much is your company data worth? How AI companies price operational data

A few years ago, the question "how much is your company data worth?" barely had an answer. Today, AI labs are increasingly buying workplace data – Slack, email and support tickets – to train AI agents. That means operational data that has sat untouched in archives can suddenly have a price. But pricing is very different from selling a machine or a property. This guide explains how AI companies think when they value data, what levels occur and how you can make an initial sanity check yourselves.

Why AI companies pay for operational data

Large language models have already been trained on most of what is openly available on the web. What is missing is data that shows how work actually gets done: how a customer service team resolves a case, how a sales team moves a deal forward or how a project group makes decisions in a chat channel. That kind of data exists almost only inside companies.

For European companies there is also a language factor. Nordic and other smaller European languages make up a small share of global training data. AI Sweden built GPT-SW3 on a Nordic corpus, and OpenEuroLLM has had to generate synthetic data because of the shortage. In Norway, the National Library saw that licensed books and newspapers improved Norwegian models. In other words, real text in local languages from real businesses is in short supply.

How much is company data worth? Example price ranges

There is no open price list for training data yet, but a few price signals from the US give an indication. The amounts below are roughly converted from dollars to euros and should be read as examples, not commitments:

Two things are worth noting. First, the range is enormous – a factor of twenty or more between the lowest and highest levels. Second, the figures come from a different market, with different laws and different kinds of companies. A European deal can land both higher and lower, and many datasets can't be licensed at all. The only way to get a real figure is a concrete assessment and a concrete offer.

The factors that drive value

When an AI company looks at a dataset, it is basically asking one question: how much better will our models get from this, and how big is the risk? That breaks down into a number of factors.

Volume and history

More years give more complete sequences – from first customer contact to closed deal, from fault report to resolved case. An archive with ten years of history is usually more interesting than one with a year's history, even if the number of messages per year is the same.

Reasoning and complete sequences

Threads where people discuss, question and solve problems carry more weight than automated notifications, newsletters and calendar invites. For agent training it is the chain of steps that is valuable: question, analysis, action, outcome.

Language and domain

Business communication in Nordic and other smaller European languages is rare in training data. A dataset from a specific industry – for example technical support, logistics or legal administration – can be particularly in demand if few others can offer the same thing.

Rights and personal data

This is often where it is decided whether there will be a deal at all. Data without personal data, or where personal data can reasonably be removed, is much easier to license. Health data under Article 9 GDPR is entirely outside what we broker. Emails and chats from employees may also be covered by workplace privacy rules, and depending on your country, employee consultation requirements. Read more on our page about GDPR and anonymization.

Structure and quality

Data that can be exported in a consistent format, with timestamps, threads and metadata intact, takes less work for the buyer. Messy material with duplicates and broken attachments lowers the value.

Documentation

The EU AI Act (Article 53) requires providers of general-purpose AI models to respect copyright reservations and publish a summary of their training content. Data with clear provenance and documentation makes it easier for the buyer to meet those requirements, and that has value in itself.

What lowers the value?

Knowing what pulls the other way is just as important as knowing the value drivers:

• Unclear rights – for example if parts of the material belong to customers or partners.
• A lot of personal data that is hard to de-identify without the content losing its meaning.
• Platform terms – Salesforce has reportedly restricted bulk export and LLM training through Slack's API terms. How, and whether, you can export your own archives requires legal review.
• Short history or data consisting mostly of automated messages.
• Sensitive categories such as health data, which can't be licensed through us.

How license terms affect revenue

You are not selling off the data. A license is by default non-exclusive and limited in scope, time and use. That affects the economics in two ways.

A non-exclusive license means the same dataset can in principle be licensed to several buyers over time, which can produce several revenue streams instead of one. An exclusive license can command a higher price per deal but closes the door to others. Limits on time and use mean you stay in control and can renegotiate. More about how the agreements work under data licensing.

Checklist: prepare an assessment of your data

You don't need to send any files to get an initial assessment. But the better you can describe the data, the more accurate it will be. Go through the points below:

1. Which sources do you have? Email archives, Slack or Teams, support systems, CRM, call recordings, documents and intranets, ERP exports.
2. How far back does the history go? Give an approximate number of years per source.
3. How large is the volume? Approximate number of emails, messages, cases or hours of audio.
4. Which languages? Share of English, Nordic and other languages.
5. Which industry and domain? What is the content about?
6. Is there personal data? Customers, employees, third parties – and if so, how much.
7. Is there sensitive data? Health, trade union membership or anything else requiring particular care.
8. Who owns the rights? Are there agreements with customers or suppliers that limit use?
9. Can the data be exported? In what format, and do the platform's terms allow it?
10. Which internal decisions are needed? Board, management and, depending on your country, consultation with employee representatives about staff communication.

If you have answers to most of these questions, you have a good basis. If you don't, that is perfectly fine too – part of the assessment is helping you work it out.

From estimate to offer

Our process is simple. You describe your data in our form in about two minutes, without uploading anything. We assess rights, personal data and demand, and get back to you within 48 hours on business days. If there is a possible deal, you get a concrete offer to consider. Nothing is shared until a non-disclosure agreement and an approved offer are in place, and delivery only takes place after de-identification.

The assessment is free. We get paid through a share of the deal, and only if a deal actually goes through. That also means we speak up early if we don't think a particular dataset can be licensed.

For a shorter overview of how value is assessed in practice, see what is your data worth.

Summary

Company data has gained value because AI companies need to see how real work gets done, and data in Nordic and other smaller European languages is rare. Price signals from the US range from around ten thousand euros to well over a million – examples, not commitments. What determines where your data lands is history, reasoning, language, quality and above all the state of rights and personal data. Start with the checklist above, and you'll be well on your way.