Data First, AI Second

Business leaders are starting to realize how Artificial Intelligence can help drive and automate their business decisions. AI can radically enhance the commercial performance and operational efficiency of virtually any business by enabling resiliency within today’s fast-paced and highly dynamic markets. That said, most people don’t realize that to take real advantage of AI, one must first get their data in order. In other words, your AI is only as smart as the data it has access to.

Before we get into the complexities of correctly utilizing data for the purpose of AI, let’s first provide an overview of AI basics. The most common and relevant subset of AI is Machine Learning. Machine Learning is a data science technique that enables computers to learn and improve their understanding of how certain data impacts an outcome. Most importantly, Machine Learning enables computers to do such learning without being told explicitly how to find an answer.

The most advanced and cutting-edge form of Machine Learning that is almost exclusively applied at FLYR Labs, Deep Learning, mimics the function of the human brain and is capable of learning without supervision, simply drawing understanding straight from the data it is given.

Deep Learning is a particularly powerful tool for airlines when combined with today’s immense computing power available in cloud platforms such as Google’s. For example, while a traditional chip inside your laptop can handle around 4,000 calculations per second, special chips developed by Google and utilized at FLYR Labs can handle trillions of operations per second, re-optimizing entire airline networks in minutes while only consuming 2 watts of power.

A ‘Pod’ of Google Tensor Processing Unites (TPUs) utilized by FLYR Labs for highly specialized deep learning models that forecast revenue, load, and pricing strategy.

The unparalleled computing power and vast amounts of available yet underutilized data are perfectly suited to tackle the complexities faced by modern-day, COVID19-era airlines. Such advances and environment provide the perfect justification to rapidly deploy AI capabilities.

While airlines are eager to jump right into grand transformation programs, they often do so blind to the importance of a highly structured, extensively vetted, and well-maintained data foundation. AI is not a technology that simply plugs into existing systems and, voila, evolves you into the airline of the future, you require the right planning and infrastructure first.

The Baggage Behind Airline Data

Legacy solutions that used to inform airline business decisions ignore vast amounts of data that simply ‘sits on the shelf.’ For example, search activity across channels, events, promotions, ancillary revenues, schedule changes and more are either ignored or overly simplified. As if ignoring commercially critical information is not enough, the antiquated codebases behind legacy products can only forecast or re-optimize infrequently and are unable to cope with data sparsity. In short, they are always behind the eight ball.

Forced to work with what they are handed, airline analysts are frequently forced to deploy subjective, manual overrides to influence pricing. In fact, up to 60% of prices observed by consumers are the result of some sort of rule-based intervention by an analyst.

At FLYR Labs we have developed technologies that consider the complete context of an airline’s network as one giant interconnected source of knowledge, enabling us to surmount traditional challenges related to reactivity and data sparsity.

Historically, airlines have not maintained their data and data storage solutions in a consistent manner. When our team at FLYR Labs engages with a new airline client, we generally encounter a wide array of data formats and storage technologies, many dating back to the 90s and early 2000s and each hailing a different owner within the organization.

This lack of structure is keeping valuable data siloed across the airline. Even technology platform vendors that provide great underpinning data infrastructure don’t have the expertise or knowledge to evaluate, transform, store, and monitor the highly specific data types and formats found at airlines.

Establishing a Canonical Data Model

Getting your data into a well organized state pays off exponentially down the line. A well-managed data infrastructure can deliver critical insights across an airline’s departments, enabling optimization and alignment between network planning, marketing, crew planning, and other commercial teams within the organization.

To assure consistency and longevity of a data platform, one must establish a standardized format to organize data. We often refer to this as the Canonical Data Model (CDM). The CDM allows all the data to remain structured and well-governed while establishing clear relationships between previously unconnected data sources. In part because of the CDM, we can break past the silos and interoperability constraints that used to inhibit innovation and progress, regardless of age, size, and complexity of the airline’s underlying data.

The First Step Enables a Giant Leap

Once we have established the airline’s data infrastructure and data governance, we can start feeding data into our AI infrastructure. Only then can we truly move towards training and deploying AI algorithms that forecast and maximize revenue to automate and inform decisions.

Now that airlines are starting to increasingly adopt vertical Enterprise Software-as-a-Service solutions such as our Cirrus Business Intelligence platform, constant updates and improvements, high reliability thanks to cloud services, and ongoing access to innovation dictate the new standard.

Without having to request or pay for new capabilities, solutions such as Cirrus simply make new features available as they are invented, built, and released into the wild. The data platform at the foundation of our solution is the primary enabler of such rapid, production-grade release of new functionality.

Similar to how fly-by-wire controls transformed aircraft efficiency, reliability, and comfort by placing a computer algorithm between the pilot and control surfaces, Cirrus helps automate the most common analyst actions and directs focus to where it is most critical. For example, markets that are seeing a rapid decline in forecasted day-of-departure revenues by dollar value.

To the Future

Organizing the airline’s data to enable AI-driven transformation of revenue management and corporate decision making is merely the first step. We envision a bright future in which our Cirrus revenue, load, and pricing strategy forecasting is just the tip of the iceberg of what’s possible.

With Cirrus, we are unlocking ways to simulate and answer complex questions before they find their way into reality. “Which will generate more revenue, a departure at 7am or at 8am?”, “How is my revenue impacted when a low-cost competitor enters this market?”. What question will you ask?

Just remember, Data First, AI Second.