Enabling Artificial Intelligence (AI) in business - A guide for airlines
The digital transformation wave is reshaping the airline passenger experience like never before. Airlines are investing heavily to harness maximum value from the vast swathes of data available, using a plethora of technologies available today. Among these, Artificial Intelligence (AI) is drawing significant attention. However, incorporating AI into mainstream customer facing operations is not easy. It requires a planned and strategic approach to help unlock the potential that lies beneath.
Successful AI implementations often follow a 3-step approach comprising of:
Let us have a closer look at each of them.
The objective of the discovery phase is to identify, prioritise, and document AI use cases. It can be further broken down into 3 stages
- Assessment - The first and foremost task is to conduct a comprehensive assessment of the current business processes and use cases within an organisation. Then we need to select a set of business scenarios that has significant impact on business performance and customer experience. An example would be the booking operations a flyer has to deal with. This would have direct impact on the sales revenue and hence become a good candidate for AI enablement.
- Brainstorming - The scenarios selected from the assessment stage are explored deeply to identify use cases feasible for enabling AI. The best way forward is to look at possible questions that flyers would ask themselves before every flight booking. Once the right questions are identified, Machine Learning (ML) algorithms can be deployed on the data that could provide insights to predict the exact thoughts or expectations of flyers. The following table is a depiction of what data AI systems process to arrive at predictions for customer experience.
|End User Thought/Expectations
|Airline inputs for Prediction
| I need to find the cheapest flight
| Past search and booking data, website engagements etc
| I usually travel on aisle seats and prefer them
| Past check-in data
| I am a vegetarian and prefer sandwiches
| In-flight purchases and engagement data
The questions that could be asked in this stage include:
- - Is user interaction really required for the scenario or how much could it be reduced?
- What are the ways in which a system could predict and aid flyers in achieving their objectives?
- What kind of data would be required to aid the flyer?
- Could the action done by the system be improved by providing training data and feedback?
- Can a device or sensor automate or provide data for the scenario?
- Feasibility analysis - Each scenario is further evaluated for feasibility of implementation and this requires the following parameters to be assessed comprehensively:
- Availability of data and feasibility of data collection for the scenario
- Infrastructure and development costs
- Impact to existing systems feasibility and cost
- Change management and operations impact
At the end of this stage, we get the final priority list of scenarios to proceed with AI enablement.
Now that we have identified and prioritised the use cases that can be the first candidates for AI enablement, it is time to document them. The key consideration while documenting is to ensure that all relevant dependencies and aspects are clearly captured for empowering teams to work more precisely on engineering AI capabilities. These aspects include, the benefits for each use case with AI enablement, complexity of implementation, impacted systems and stakeholders, estimated efforts and schedule, expected revenue impact, and the costs involved.
In the implementation phase, the key objective is to build real-time AI solutions with use cases identified in the discovery phase. A team comprising of data scientists, business analysts, and data engineers would be ideal for the implementation phase. Their roles are as follows:
- Business analysts - Work with consumer systems and manage expectations and outcome of the project to the stakeholders
- Data scientists - Assess the input features and their dependencies on the outcome and arrive at an efficient data model to be used
- Data engineers - Develop data collection routines, implement models, and provide data consumption and interface applications to work along with consumer applications
When we consider AI implementation, the dreaded thought of complicated algorithms could possibly lead us to drive back from proceeding, but today, we are fortunately at a very advanced technology ecosystem where the hardest of AI workarounds are encapsulated in libraries and even provided as services by multiple cloud vendors. Development teams engaged in building solutions just need to make the right use of it. The only area that needs focus is building a mathematical model where features are connected to business outcomes. Once the model is finalised and validated with test data, appropriate libraries/services could be used by data engineers to implement the model.
The key objective of the operations phase is to ensure that an AI implementation is sustainably operated for an airline to generate value from it.
Once the system goes live, the model performance will be monitored based on agreed performance parameters. Based on the prediction success percentage and feedback, the model will be continuously improved using methodologies like A/B testing where certain portion of traffic would be tested with modified models.
Artificial intelligence has matured dramatically over the past couple of years, but a recent SITA report1 points out that only 32% of airlines have major programs in place to tap the benefits of AI in their business operations. So, your competitors are probably still on their learning track. But, in another year or so, success stories are sure to pour in aplenty. The 3-phase approach explained here is ideal for airlines to drive return on investments from their AI investments, and if followed meticulously, can be the foundation for the next big success story in the airline sector.
About the author
Binu George, is an Enterprise and Solution Architecture expert who has worked with major clients in the airline and hospitality industries across geographies. He has attained a great deal of expertise and experience in the architecture, design, and implementation of mission-critical applications and digital transformation projects, creating significant business value and operational efficiency for the clients. He is an AWS Solution Architect - Professional and a TOGAF 9 certified professional.