Google’s $10MN bid for the corporate data of bankrupt Spirit Airlines initially appeared to be an unusual consequence of a company collapse. The deal now looks more like the emergence of a developing market where AI companies are increasingly willing to pay large sums for the data generated by businesses and their employees.
Google won the bankruptcy auction for Spirit’s internal business data over AI training company Mercor. AI models can be trained on enormous quantities of publicly available information, but internal business data captures information that AI developers increasingly struggle to obtain fresh sources for—the messy reality of how humans work and interact. Operational records are particularly valuable for developing AI agents designed to perform real workplace tasks.
Micro1, an AI data company that subsequent to the auction offered $12.5MN for Spirit’s data, has made this opportunity part of its business model. The company says it helps businesses monetize their operational data through licensing arrangements and is seeking information such as standard operating procedures, knowledge bases, CRM data, project histories, QA processes and decision-making patterns.
Micro1’s interest in Spirit also illustrates the growing competition for this information. The company’s founder and CEO, Ali Ansari, stated in a post on X on August 18 that “in the last 11 days, we’ve committed more than $20,000,000 to license real operational data to seed our RL environments. scaling on the realism dimension is just beginning.”
This week, on August 24, Ansari added:
“[M]ore than 1,000 companies have signed up to get paid for their anonymized data to train models just in the last few weeks. [A] massive TAM opportunity for the entire economy has emerged almost overnight.”
The claim provides a striking indication of how quickly the market may be developing.
Ansari has been making a broader argument about the commercial value of human-generated data. In a January post, Ansari claimed that “human data will be a $1 trillion/year market,” its value rising as AI systems become capable of automating more economic activity.
Micro1 is actively recruiting businesses into its data partnership program, with compensation based on the quality, uniqueness and value of the data provided. Its published examples include customer-support documentation, ticket workflows, QA processes, customer operations and knowledge management, indicating that the information generated by a contact center or customer-service operation can contain precisely the kind of real-world operational knowledge AI developers want to replicate.
Such context can be far more useful than another tranche of public web content.
Micro1 is far from the only company aiming to claim a share of that market. Scale AI is inviting submissions of non-public datasets to its Dataset Collection Initiative to help train models to analyze data, seeking material such as marketing performance, social media advertising campaigns, system logs, investing and scientific research. It offers contributors payments depending on the quality and use case of the dataset.
Trust Will Become Part of the Data Deal
The growing demand for non-public datasets could turn years of accumulated enterprise data into a potentially valuable asset.
CX teams are sitting on particularly rich datasets. Contact center transcripts, call recordings, agent notes, CRM histories, support tickets, quality-assurance records, workforce-management data and internal knowledge bases can collectively provide a detailed map of how an organization responds to large volumes of real customer situations.
Until recently, much of that information was valuable primarily because it helped the business operate. But AI’s insatiable need for data changes the calculation.
The Spirit case is different from conventional licensing deals involving media archives, stock photography or video libraries. And it is being sold through bankruptcy, meaning information created for one commercial purpose is potentially being repurposed in another.
“The consequences of this acquisition are chilling,” privacy expert Ron Zayas, Chief Executive Officer of Ironwall by Incogni, told CX Today.
“Understand that while it is customary for companies buying assets out of bankruptcy to get the data that they need to utilize those assets, in this case Google is buying the data. They claim they will use it only in aggregate to train AI engines, but that is a voluntary commitment. They are likely under no legal agreement to do so.”
The Association of Flight Attendants-CWA (AFA), representing Spirit’s former flight attendants, pointed out in objecting to the sale that Google could make inferences about employees from the anonymized data. The company could also potentially monetize customer behavior around travel, Zayas noted, adding “this is how Google makes money.”
There is also a reason some observers are skeptical of Google’s involvement in particular.
Google has faced significant privacy scrutiny before. In 2024, it agreed to delete billions of records of private browsing activity as part of a settlement of a class action over the Chrome browser’s Incognito mode, following allegations that the company had continued collecting users’ browsing data despite the privacy expectations associated with the setting.
And in 2025, Google agreed to a $1.375BN settlement with Texas Attorney General Ken Paxton over allegations that it unlawfully tracked and collected sensitive user data, including geolocation, Incognito searches and biometric information. Paxton described it as the largest recovery against Google by a state attorney general enforcing state privacy laws.
Those cases do not establish that Google will misuse Spirit’s data, but they explain why skeptics are likely to scrutinize its assurances around data handling closely.
For enterprises considering similar deals, a data buyer’s track record may become as important as the price it offers.
Customer Data Could Become Part of the Equation
A company could potentially license selected operational data to an AI developer, generating a new source of revenue while contributing information that helps models understand real business processes.
Recognizing the value in its customer data, Spirit is looking to sell off its list separately and expressly removed it from Google and Mercor’s early auction bidding.
The complication is that customer data does not become less sensitive because another company sees commercial value in it.
Enterprises considering data licensing would need to establish what information can legitimately be shared, whether customers or employees could be identified from supposedly anonymized datasets, what contractual restrictions apply, how long the recipient can retain the information and whether it can be reused for other purposes.
The Spirit dispute indicates how complicated those questions can become when the organization selling the information is already in bankruptcy proceedings. The risk is different when a company has not deliberately decided to build a data-licensing business.
In the Netherlands, recent reports indicate that customer data from several bankrupt solar panel companies appears to have ended up in the hands of sustainability advisers who used the information to contact former customers. The businesses making the calls were able to cite specific details about customers’ solar installations, while the source of the information remained unclear.
This raised suspicions that customer lists from failed companies may have been traded online, illustrating another dimension of the data market: customer information can become vulnerable when a company collapses, even where the customers themselves have done nothing to change how their information is shared.
For enterprises, that makes data stewardship relevant beyond normal day-to-day security and privacy controls; they also need to consider what happens to customer and operational information during an acquisition, divestment, restructuring or bankruptcy and who will control that information after the original business has disappeared.
The automotive industry provides another example of how valuable operational and behavioral data can become.
Connected vehicles can collect detailed records about where people go and how they drive, creating datasets that could be useful to insurers, advertisers, mobility companies and AI developers.
The privacy risks have already attracted regulatory attention. In May, California regulators fined automaker General Motors (GM) a record $12.75MN over allegations that it sold hundreds of thousands of Californians’ location and driving information to data brokers Verisk Analytics and LexisNexis Risk Solutions, bringing in roughly $20MN from the data-sharing arrangements.
The case shows how information generated as a by-product of a service can acquire value far beyond its original purpose. For AI developers, vehicle data could offer a rich source of real-world information about navigation, driving behavior, vehicle performance and human decision-making.
The issue of whether a company should be able to sell or repurpose behavioral data will become more important as AI developers look for increasingly specialized datasets. The commercial value of information may give businesses a stronger incentive to monetize data they already hold, while regulators and customers may expect clearer boundaries around how that information is used.
Enterprises Need To Decide What Their Data Is Worth
The emergence of a market for operational data could change how enterprises think about information, providing a new way to generate value from data they already hold.
But selling data for AI development also means giving another company access to information that may contain years of employee activity, customer interactions and operational knowledge.
The first question may be whether data can generate revenue. But before that, organizations need to establish whether they have the rights, permissions, governance and security controls required to commercialize it. Before licensing enterprise data, organizations will need to establish what can be sold, what must be excluded, how it will be deidentified, who verifies that process and what restrictions apply to subsequent use.
This could mean CX teams taking a closer look at datasets that have historically been treated as operational exhaust. While useful for training and evaluating AI agents, they could also expose customers and employees if handled carelessly, creating a tension between the potential value of data and the responsibilities attached to it.
The more commercially valuable operational information becomes, the greater the incentive to collect, retain and monetize it, and the greater the consequences when ownership or control changes.
It will become harder to avoid the strategic question of whether valuable operational data should be treated as a revenue-generating asset, a protected corporate resource, or both?
AI developers are increasingly willing to pay enterprises for their proprietary data. The question for business leaders is whether they are ready for that information to become an asset that others want to buy.
