Google buys Spirit data for AI training, say executives

Google has acquired Spirit Airlines' digital assets, drawing attention to the value of customer-service data for training large language models, industry executives said.

The transaction stands out because customer-service records contain large volumes of natural-language exchanges across calls, emails and chats. For model developers, they offer access to real-world conversations, including queries, complaints, booking changes and other service interactions that reflect how people actually communicate.

Adil Tahiri, Chief Technology Officer at Konecta, said the case is notable because of both the airline's circumstances and the legal framework governing the sale. Those factors, he argued, may limit the extent to which the deal becomes a template for a broader market in customer-service data.

"Google's acquisition of Spirit Airlines' digital assets, including customer-service data to train its LLMs, puts a whole new spin on the phrase, 'This call will be recorded for training and monitoring purposes.' But could the move signal a broader trend of AI companies seeking to partner with or acquire customer-service organisations? Customer-service data is undoubtedly a treasure trove for training LLMs. An airline's data may be particularly valuable because it is not only rich with real conversations but also highly varied, spanning multiple languages, demographics and regions.

"Before Anthropic and OpenAI begin a wider hunt for customer interaction datasets, it is important to understand why this case may be unique. The Spirit Airlines deal, pending judicial approval, took place in the US, where data-protection laws are arguably less stringent than in Europe. Moreover, because the airline is defunct, its estate, as the data controller with no ongoing service obligation to customers, was focused on recovering value for creditors rather than on the longer-term implications of selling the data.

"In Europe, the situation would be very different. Although customers may be told that calls are recorded for 'training' purposes, selling those recordings to a third party to train its AI models represents a change in how the data is used. Under GDPR, that would create significant legal and regulatory barriers, making European customer-service organisations unlikely targets for a similar deal.

"This is where a business process outsourcing model differs fundamentally. Providers use each customer's data exclusively to train and improve that same customer's own AI agents, and the data is never shared or pooled across different customers. This is contractually enforced, as the data remains tied to serving the customer it came from, in full compliance with applicable contracts and regulations.

"But does that mean customer-service data will become the next gold rush in the US? Not necessarily. Google reportedly paid more than $10 million for Spirit Airlines' data, beating out Mercor, a company that specialises in supplying training data to AI firms, which shows the market already exists. The high valuation could encourage other companies to demand a high price for access to current and ongoing customer data, but that will inevitably raise questions about return on investment. The key question now is how Google will use the data and what tangible benefits it will ultimately deliver before this signals any kind of trend," Tahiri said.

Data value

The sale has sharpened debate over which kinds of data are most useful in the race to improve AI systems. Public web content has long been a major source for training models, but service records offer something different: direct exchanges between customers and agents, often tied to practical tasks, emotional tone and problem resolution.

That makes contact-centre data attractive not only to model developers seeking scale, but also to companies looking to improve their own internal AI tools. Analysts and software providers have increasingly argued that call transcripts and chat logs can help businesses identify recurring issues, refine automated responses and support staff handling more complex cases.

David Fischer, Chief Revenue Officer at Luware, said many organisations already hold data that could be used for those purposes without buying external datasets. He described the Google-Spirit deal as evidence of the monetary value now attached to conversational records.

"The fact that Google has acquired a large dataset of deidentified airline customer-service interactions to support AI development highlights just how valuable conversational data has become. For Google, that data is clearly a gold mine, worth more than a staggering $10 million. But most organisations are already sitting on this opportunity within their own contact centres. Every call, chat and customer interaction can fuel an AI feedback loop, helping teams identify recurring issues, improve self-service, better support agents and continuously refine the customer experience.

"Businesses do not necessarily need to spend millions to buy vast external datasets to realise this value. With the right privacy safeguards, governance and human oversight in place, they can responsibly use the interactions they already have to make both their service and their AI better over time," Fischer said.

Regulatory split

A central question is whether similar sales could happen at scale, particularly outside the US. Tahiri pointed to a sharp divide between American and European data rules, arguing that a change in the purpose of collected customer data would face tougher scrutiny under GDPR.

That distinction matters for companies in customer service and outsourcing. Service providers often process data on behalf of clients under contracts that limit how information can be used. In that model, one client's call recordings or chat logs cannot simply be pooled with another's to train a broader third-party system.

The economics may also prove less straightforward than the headline figure suggests. If businesses begin to view historic service records as saleable assets, prices could rise quickly, especially for datasets spanning different languages, regions and customer groups. Buyers would then need to show that the quality and diversity of those records produce gains that justify the cost.

For now, the Spirit transaction is being watched as a test of how far AI developers will go to secure proprietary data, and how regulators, customers and companies respond when service interactions become part of that market. The reported price of more than $10 million has already turned a routine store of operational records into a closely watched asset class.