Cognitive Computing

What can cognitive computing do in airline operations?

The safe and efficient management of an airline is an extremely complex task with several groups of individuals and their machines working in close coordination as one large socio-technical system.

Airlines – driven by intensifying competition and the need to stay profitable – turned to operations research decades ago as a means to deal with this complexity, and vigorously deployed advanced optimization techniques to power business decision support. Today we have, as a result of decades of sustained investment, dozens of mission-critical expert systems applying cutting edge mathematical techniques in diverse airline business areas such as demand forecasting, network design, route & schedule planning, revenue management, crew planning and assignment, recovery from irregular operations, and air traffic flow management.

However, a characteristic common to all decision support systems of the past is that they invariably use a hardcoded set of static rules. Consequently, they work well as long as the environment they are deployed in is identical to that they were designed for – typically static or slowly evolving environments; any sudden or sharp change in operating environment produces results that are suboptimal at best and inadequate or misleading at worst, and the systems rapidly become inoperable.

The world we live in – highly interconnected and constantly changing – is referred in research as a dynamic environment where actions are required to be made in real time, depend on previous action outcomes, and on parameters that are continuously changing. Often in such an environment the information needed to “optimally” solve a problem is not known, not known precisely enough, not in a format that can be readily fed into the expert system, or the volume is too large to handle. As a result the decision support systems are quickly cast aside and people quietly get back to working with thumb rules and judgment.

Examples for this abound in the industry; in airline context we need to look no further than the Cost Index – a powerful tool to optimize flight performance by trading off time costs against fuel costs – hardly put to full use, because real-life complexities and rapidly changing environments have prevented airlines from reliably establishing the cost of arriving late at a destination. Recovery from schedule disruption is another example. Although as much as 10% of all scheduled airline operations are considered to be irregular[1] (i.e. different from the plan), several airlines even today manage disruptions with very little computing power: the drivers and interdependencies of a real-life operation are seen as too complex to model accurately into a system, or t

The information needed by a system to analyze and recommend actions is too large to be practically made available in time.

To unravel this riddle, researchers looked at several disciplines in search of inspiration. And human cognitive skills provided a clue. From the cross pollination of ideas between human cognitive science and artificial intelligence emerged Cognitive Computing – as an answer perhaps to the search for expert systems with intelligence that is more ‘human’ and less ‘artificial’: computer systems modeled after the human brain, that can process natural language (NLP) and unstructured data to interact with humans in human-like ways, and that learn and grow smarter with time so they need not be hardcoded with static rules.

These human-like capabilitiescognitive could have a profound impact on how an airline makes routine operational decisions. Here is some crystal ball gazing on what cognitive computing can do in airline operations:





  • Dynamic re-routing of aircraft in response to unfolding weather conditions:

Aircraft routing is done today using computer-generated flight paths based on best available weather forecasts at one hour before takeoff or earlier. Air traffic controllers might choose to make changes to flight plans filed by airline dispatchers and implement reroutes to account for weather uncertainty – these often result in further increasing already large buffers factored into the flight plans. Controllers and dispatchers are busy people and would very likely miss opportunities for more efficient routes when weather changes as flights progress along planned routes.

To illustrate the size of the problem, adverse weather affected delayed flights consumed 740Million additional gallons of jet fuel at a cost of $ 1.6 billion, in the US alone in 2007.[2]

What if we could do real-time re-routing of aircraft enroute using live data streams of weather systems and airspace traffic?

This is perhaps not very far-off from reality. We already have today the Dynamic Weather Routes[3] – a search engine that continuously and automatically analyzes in-flight aircraft in enroute airspace and proposes route amendments for improving efficiency while considering traffic conditions and other real-time airspace constraints. The DWR facility was under trial by NASA and American Airlines at Fort Worth, Texas, as of October 2012.

Airline flight operations would detect an upcoming storm system 30minutes ahead an enroute flight using DWR, and feed the cognitive computing-enabled flight planning system at the control center with DWR recommendations. They initiate an impact assessment of a rerouting on dependencies such as fuel on board, schedule impact, cost of delay, crew assignments etc. to arrive at a decision to reroute within a workable timeframe. The system would analyze patterns of past rerouting instances on dependencies and come up with a recommended option among reroute alternatives for overall operations efficiency.

 The flight operations agrees on the new route with crew piloting the aircraft, and the flight plans on the cockpit FMS (flight management system) are automatically updated. The captain announces to passengers, “…we just took a detour and avoided a storm system; this will have no impact on arrival time as it will be made up for in remaining flight, and the aircraft continues to cruise. Relax and enjoy the flight.”

  • Recovery from Irregular Operations:

Disruptions seem to have grown into a natural aspect of the airline industry, and occurring with increasing regularity. While only a fraction of causes of irregular operations are within airline control[4], several airlines today lack a coherent support system to analyze and decide quickly on actions. Instead they react with a few broad strokes such as cancelling a block of flights or operating the entire fleet an hour behind schedule, and then wait and watch to see if that fixed the problem[5].

What if we could accurately pin-point a few specific actions that would fix entire operations and bring it back to normal, with minimum impact to everybody?

A cognitive computing system would simulate several dozens of recovery scenarios, “intelligently” compute the positives and negatives of each scenario in terms of impact to passenger, to employee and to the airline, compare these likely outcomes with similar instances in the past, and recommend to the user with degrees of confidence the best suited scenario and hence the specific actions and their sequence.

Over time, the system would ‘understand’ the airline’s preferences and tradeoff choices in impact to stakeholders, and generate options in real-time as the disruption occurs.

  • Rapid design and launch of products driven by Customer Insight:

The deluge of social media content we have today is a renewable resource for customer insight. We have the means to make sense of terabytes of data and analyze what is being said about our brand or product and who is saying it. Social Media Analytics uses NLP capabilities and distills out popular sentiment among consumers, and sometimes even their behavior and underlying motivations. This knowledge is available today. However these capabilities are used more to gauge and manage negative sentiment than to proactively engage with new offerings.

What is not clear yet is how we can use this knowledge to develop insights into customer needs and how that can translate into product opportunities.

What if we could spot an emerging product opportunity by analyzing social media data and cash in by rapidly designing and delivering the product to the consumer?

Social Media analytics would pick up a particular string of interactions among red-eye flight travelers and the pattern-spotting capability of cognitive computing quickly sizes up the number of ‘such’ travelers doing frequent red-eye flights with the airline. If this segment size is significant for the airline, the system goes further: it interacts with the travelers in real time to understand their latent or unmet needs, and also trials potential product ideas for their reaction – again by leveraging its NLP capabilities. After establishing an interest for a product idea, it makes recommendations on what new product will interest a segment, how strong is the propensity to buy and at what price.

The system also looks at past such instances and makes an estimate on the nature of the new demand – whether it is likely a fad or a wave or a tsunami.

  • Fuel loading decisions by Airline Dispatch crew:

Flight Planners today have access to advanced tools to compute fuel requirements for a flight: they take into account the best available forecasts of weather and traffic, taxi times and takeoff weights. However these computations are done using hardcoded rules and assumptions about cruise speeds, alternate airports and several others that significantly influence the computed fuel values. Also the tools typically do not allow comparison of actual fuel burns with plan and do not support validation of assumptions or pruning of buffers.

What if we could learn from experience and continuously improve our efficiency in fuel computation?

We have made a beginning here with IBM’s tool (developed together with Air Canada) that predicts the amount of discretionary fuel – or fuel the pilot or dispatcher chooses to carry on flight over and above plan – based on actual fuel burns in the past for the same aircraft and route under similar conditions of weather and traffic. The same capabilities of cognitive computing employed in this tool, viz. spotting patterns and machine learning, could be extended to other elements of fuel computation.

And we could have a smart system that is not hardcoded with static logic but learns from experience, to compute fuel quantities based on not only the latest forecasts of weather and traffic but also actual burns data of past flights in similar conditions.

Cognitive computing appears to hold the promise to transform the way decisions are made in airline operations. And in effect, pave the way for a smarter transportation, and a smarter planet.

Will we make this happen?

[1] Design of Support Systems for Dynamic Decision Making in Airline Operations: technical paper by Karen Feigh and Amy Pritchett, Georgia Institute of Technology

[2] Charles Schumer, Carolyn B. Maloney, “Your Flight Has Been Delayed Again, Flight Delays Cost Passengers, Airlines, and the US Economy Billions,” A Report by the Joint Economic Committee Majority Staff, May 2008.

[3] Jeff Osborne et al, Operational Evaluation of Dynamic Weather Routes at American Airlines, NASA Research Center and American Airlines

[4] Rich Coskey, “Bouncing back” – article in Ascend journal 2007, Sabre Airline Solutions

[5] Design of Support Systems for Dynamic Decision Making in Airline Operations: technical paper by Karen Feigh and Amy Pritchett, Georgia Institute of Technology

Global SME-aviation operations, Travel & Transportation Centre of Competence

More stories

The Shocking Truth About Cybersecurity Threats

Ransomware Carnage Cost Billions in 2017.1 Overall Monetary Damage Caused by Cyber Crime May Reach $6 Trillion per Year by 2021.2 174,402,528 Data Breaches* Occurred in 2017.3 Most IT security professionals know the above statistics and have read many more. All of these hazards can negatively affect a company’s operations, financial condition, and/or liquidity. These […]

Continue reading

World Rail Festival 2017: From digital disruption to digital reinvention

Key international railway leaders gathered for the first day of World Rail Festival in Amsterdam to discuss the digitization of the rail business model, how commercial growth can be achieved and how the rail industry can prepare to meet the demands of the travel ecosystem of the future. To win in the marketplace today, simply […]

Continue reading

Will Conditions Turn More Favorable and Allow Airlines to Soar?

Global economic conditions, foreign exchange rate fluctuations, extensive government fees and taxes, environmental regulations, jet fuel prices, fixed asset utilization, intense competition, weather disruptions, mechanical difficulties, IT infrastructure obsolescence, security. These are some of the challenges faced by the adventurous and lionhearted airline industry participants. Those daring to venture into this energetic and ultra-important industry […]

Continue reading