How to Value Your Operational Data, Lessons from Spirit Airlines

Spirit Airlines collapsed and began auctioning its assets. What would you normally expect to see in an airline’s bankruptcy auction? Aircraft, equipment, software, maybe even airport slots. But what made the headlines was their data — not customer records or invoices, but operational data: emails and Teams messages.

Google offered to pay $10M for it. Google wasn’t the only one bidding. Mercor, an AI-powered hiring platform, offered $7.5M.

According to reports, the dataset comprises 100 million emails, 500 million Teams messages, and 30 million lines of code — the company’s day-to-day operations and decision-making conversations. A dying company’s internal communications are worth $10M.

The deal isn’t final yet. Airline staff filed a petition against it over privacy concerns. But why did Google want that data? More importantly, how can you estimate the value of your own data estate, and what makes that estimation so difficult?

What Makes Spirit Airlines’ Data So Attractive?

Traditionally, structured data has been perceived as high value because it feeds easily into analytical engines and machine learning models. Converting unstructured data into a model-friendly format was cumbersome, and most companies didn’t think the ROI was meaningful.

That perspective has changed. GenAI can now extract structure from massive unstructured datasets. What makes Spirit Airlines’ data particularly interesting, however, is reinforcement learning.

Reinforcement learning involves an agent interacting with an environment and learning from the feedback it receives. Over time, the agent optimizes for a challenging task. It has been highly effective in teaching AI to code, partly because thousands of public GitHub repositories provide a meaningful starting point, and agents can run code in a sandbox for instant feedback. That tight feedback loop is what makes reinforcement learning viable in that domain.

But coding is a special case. Many other AI applications don’t have such a convenient feedback environment — and few industries are more complex or lucrative than airlines. Engineers can simulate physical environments to model aircraft operations, but the decisions made during day-to-day airline operations are invisible to any modeled environment.

Engineers can simulate a storm and use reinforcement learning to navigate it. But deciding whether to fly through that storm or divert to a previous airport requires more than a survival probability. An operator must weigh varied factors and expert opinions. A physical simulation can’t capture that.

What if engineers had access to decades of conversations that led to exactly those decisions? That’s what Spirit Airlines’ operational data represents to Google. How did the airline survive the pandemic? How did it respond to sudden demand swings or supply chain disruptions? These are questions no outsider can answer from public data alone.

How to Value Your Operational Data Assets

Unlike physical assets, operational data has no direct market price. But traditional valuation principles still apply. The following focuses specifically on operational data, which has been largely overlooked because it had no marketable value until recently.

There are three primary approaches to valuing any asset.

The Income Approach

The income approach values an asset based on its expected return over its lifetime. In general terms: estimate the direct cash flow the asset generates, subtract operating costs over an expected period, and discount those figures at the rate of return you could expect from the next-best alternative. This works well for physical assets, intangible assets, and entire businesses — and it can work for data too.

For operational data, however, the approach runs into two problems. Operational data doesn’t directly generate cash, and there’s no obvious next-best-alternative rate to use as a discount rate. That said, there are ways to work around both.

Operational data doesn’t generate cash directly — but it saves it.

Emails don’t make money on their own, but without them the company wouldn’t have made revenue at all. Older emails matter less than current ones, and not every message carries equal value: client communications and key decisions are worth more than a year-end party announcement. Even low-priority emails, though, may have helped resolve issues faster or avoided legal penalties.

The challenge is assigning a monetary value to these. The following steps help:

Step 1: Identify different data asset classes

This doesn’t mean categorizing by file type. Even within email, threads with customers, insurance providers, and internal staff should be treated as separate classes. Performance monitoring logs and application logs should be treated differently, too.

Step 2: Find future use cases for each data asset class

Each class has different potential uses. Performance monitoring logs help teams plan compute resources. Application logs explain customer queries and support debugging. Map out the realistic uses for each class before attempting to assign value.

Step 3: Estimate the revenue it brings or the cost it saves

Look at recent legal or operational issues your company or similar companies in the industry have faced. How much was lost, or could have been lost? That gives a concrete estimate of what a specific use case for a given data class is worth.

Step 4: Estimate the probability or frequency of future use

One legal dispute doesn’t mean every client will dispute with you. The value of client communication email can’t simply equal your estimated worst-case loss. Calculate the probability of these events occurring. If you lack enough historical data, use a proxy organization or estimate at the industry level.

Step 5: Compute the value of each asset class

With use cases, probabilities, and per-event values in hand, you have enough to compute expected asset value. Sum the use case values to arrive at a value for each data asset class. A finance professional would go further by distributing expected value across years and discounting at a predetermined rate, but that level of precision isn’t always necessary for an initial estimate.

The Cost Approach

If you paid $1,000 for a laptop, one reasonable answer to “what is it worth?” is $1,000. Another is whatever the same model sells for today. Both are forms of cost-based valuation — one reflecting historical cost, the other replacement cost.

Most data assets can be valued this way. If you purchased data or invested in data collection activities, you already have a figure. Operational data is different because you don’t acquire it; you generate it continuously as your business runs.

Even so, the cost approach can still be applied to operational data in two ways.

Use storage cost as a proxy for value

You choose to store this data for future needs and pay ongoing storage costs. Your organization’s retention policy reflects an implicit judgment about that data’s value — you stop paying to store data you consider useless. If the data is critical, you also maintain replicas, which adds further cost. These figures, taken together, represent a floor value for the data.

Use insurance cost as a proxy for value

If data is critical to your business operations, it may already be insured. If not, request a quotation. You can then deduct a small percentage to account for the insurer’s premium and arrive at an implied asset value.

For operational data, the cost approach is generally the quickest way to put a concrete dollar figure on your data.

The Market Approach

The market approach asks a simple question: how much could you get for your operational data if you sold it?

In practice, companies don’t sell operational data — they protect it. Outside of a bankruptcy auction like Spirit Airlines’, comparable market transactions are nearly impossible to find. But when you do identify a proxy transaction, you can’t assume your data is worth the same amount without adjustment.

Adjustments should account for data quality, volume, and variety. Volume is straightforward. Quality matters too: if the proxy dataset scores higher on completeness, correctness, timeliness, and granularity, your data may be worth less. Provenance and usage rights are equally important. You can’t sell a dataset you don’t own or lack the rights to sell — a common issue with operational data. Who legally owns an email a customer sent you? Even if you do own it, do you have the right to sell it?

That question is exactly what stalled Google’s Spirit Airlines deal. The labor union argues that the airline’s operational data contains elements that belong to the staff, not the company — and that selling it without their consent would constitute a privacy violation.

Final Thoughts

Without realizing it, your organization may be building a significant data asset. Spirit Airlines’ bankruptcy auction made that clear: everyday conversations, software logs, and internally written code were valued by Google at $10M.

If a failing airline’s data is worth that much, the question of what your data is worth becomes harder to ignore. Traditional valuation techniques — the income approach, the cost approach, and the market approach — each offer a different lens. When more than one method converges on a similar figure, confidence in that estimate increases.

Each approach requires judgment and adaptation to your specific context. The right combination of methods will depend on your industry, data maturity, and how you intend to use the valuation — whether for a transaction, internal planning, or insurance purposes.