Earlier this week, Google made headlines with its plans to buy business data from now-defunct Spirit Airlines.
Per an Aug. 14 court filing, Google was the highest bidder at the auction for the data rights, agreeing to pay $10 million.
The proposed purchase includes employee emails, chats and messaging data, as well as data related to the airlines’ finance and accounting systems, aircraft operations, revenue, website analytics and loyalty program.
“We acquired part of an enterprise dataset from Spirit Airlines, which can be helpful in improving our products and AI models,” a Google spokesperson told PhocusWire.
Before being transferred to Google, the filing states that the data is to be deidentified by a third party, meaning it will “remove or transform” elements that could be linked to a consumer.
“We will not receive any personal information from this dataset,” the spokesperson said.
But while Google said the data will be used to teach its AI models, there may be larger implications for the air and travel industry.
A new source of information
If the sale is approved, the move gives Google access to proprietary airline data, according to Eric Léopold, founder and managing director of Swiss travel consultancy Threedot.
Timothy O’Neil-Dunne, principal at travel and aviation consultancy T2Impact, called it a “unique opportunity” for Google to understand the inner workings of the airline industry.
“That is great training data—something that they could have received from their own scanning of emails but not legally. This covers a large variety of different elements, but the global 360 view of an airline, and one that went through a crisis, is a unique opportunity.”
And the next highest bid ($7.5 million) being from Mercor—a company that purchases data and supplies it to frontier AI labs for training—is another telltale sign of the data’s value, according to Gemma Timmons, director of operations and chief of staff at aviation data provider OAG.
It’s also a tangible source for Google’s large language models (LLMs).
“AI models tend to hallucinate when they don’t know something, often because they were never exposed to actual facts or data,” Léopold said. “Coding or mathematics sit on public information and work well with LLMs. Aviation runs on larger proprietary datasets that AI models usually don’t see.”
Timmons added that Spirit's data is valuable to teach AI models about processes as well.
“It can preserve evidence of the sequence behind an outcome, not only the outcome: a problem surfaces, teams apply or depart from a playbook, a decision is made, a system changes and the result follows,” Timmons said.
That is great training data—something that [Google] could have received from their own scanning of emails but not legally.
Timothy O'Neil-Dunne, T2Impact
“That can be useful for training, but particularly for evaluating whether a model can reason across communications, databases, code and operational systems, handle exceptions and produce a defensible recommendation.”
Beyond AI, Google could get a look at a tech stack outside of its own, as the filing notes data from Microsoft SharePoint, OneDrive and Teams is included in the sale.
“Google would also gain enterprise-scale workflow data from a predominantly non-Google technology stack: Microsoft 365, SAP, Navitaire, UKG and other systems,” Timmons said.
“That provides a detailed view of how a large, regulated business works across its people, software and operating systems.”
Léopold said Spirit’s commercial data, including shopping and customer care, may be helpful to train Google’s models, but other internal data could be too specific to apply generally.
He also flagged limitations given Spirit’s role in the marketplace.
“Although the data will be useful to train Google's AI models, it still provides a limited view on airlines' commercial and operational activities. Spirit was a U.S. LCC [low-cost carrier], which leaves the non-U.S. and the full-service carrier operations open for learning,” Léopold wrote on LinkedIn.
Looking ahead
Timmons suggested a few hypothetical use cases for how Spirit’s data may serve Google’s business operations, including enterprise AI, potentially creating traceable dataset spanning communications and different parts of the business.
Another is an aviation use case, considering Google Cloud’s recent partnership with Ryanair.
“Gemini Enterprise will help automate decision making and optimize crew logistics, while DeepMind models will support fleet operations and maintenance planning. Spirit’s data could help Google build or evaluate similar capabilities elsewhere,” Timmons said, reiterating that the recent filing does not give a specific intended use.
The data could also be useful as Google, and the travel industry as a whole, move closer to agentic booking.
“Insights into the daily sales and operations of a major air travel operator will help ground a more robust AI agent,” Léopold said.
Timmons noted that Google is currently running a U.S. test of agentic hotel booking in AI Mode and has plans to add flight booking.
Insights into the daily sales and operations of a major air travel operator will help ground a more robust AI agent.
Eric Léopold, Threedot
And while the Spirit filing doesn’t mention this program, the data could bolster Google’s endeavor.
“The data could nevertheless support Google’s ability to execute on any such ambitions, particularly in the post-search customer journey. Pricing, booking-curve, transaction, refund, voucher and disruption data could help Google model how a fare or itinerary behaves through servicing and disruption—the key part of the customer journey an agent has to get right before it is trusted with a transaction,” Timmons said.
Google is also limited by the fact that Spirit’s data is historical as opposed to a live operating feed, she said.
For his part, O’Neil-Dunne isn’t convinced that Google has its sights set on agentic, as it already profits from “informing about bookings.”
Still, he said, “The purchase provides a treasure trove of historical data that allows Google to understand pricing decisions, not just observe them.
“This means in the future, Google can use this data to determine whether a price is a good or bad one.”
Lingering privacy concerns
Spirit’s court filing outlines the deidentification process, and Google reiterated that personally identifiable information will be scrubbed.
And data governance also appears to be tied to the value, according to Timmons.
“Mercor offered $10 million if it could conduct deidentification in house, but the sellers designated its $7.5 million bid using third-party deidentification as the alternate. The court declaration says privacy terms could be outcome-determinative.”
Léopold said he believes there is a “commitment to remove personal data,” noting that “Google already handles personal data at scale” via billions of private messages on Gmail. On LinkedIn, he noted that this might not necessarily be top of mind, considering the search giant's existing access to "millions of mailboxes and flight searches."
And not everyone is convinced privacy can be maintained.
After news of the intended sale broke, former Spirit Airlines flight attendants filed an objection in bankruptcy court, alleging that Spirit’s filing doesn’t address whether the contents of records will be confidential. Additionally, the union said the language surrounding deidentification says it must preserve “referential integrity across the data set,” meaning it will maintain links between records.
“With a group the size of the Spirit Flight Attendant population, our union has significant concerns that it may be possible that information about identifiable individuals or small identifiable groups can still be reconstructed and determined,” a statement from the Association of Flight Attendants-CWA reads.
O’Neil-Dunne also pointed out potential flaws with deidentification.
“There is essentially no guarantee of the ‘deidentification’ process and certainly no come back on anyone if there is a ‘re-identification’ process that Google might apply to use the data provided it doesn’t go outside,” O’Neil-Dunne said.
“Frankly, I am horrified, given the correlation of the data to Google’s massive comprehension of who we all are as individuals.”