Shawn Lim
May 10, 2024

Why international airlines want a piece of Air India’s chatbot technology

Leveraging gen AI to develop a chatbot has been an important facet of Air India’s digital transformation. The Silicon Valley-based chief technology officer of the airline talks to Campaign about the process of developing and besting the chatbot.

Why international airlines want a piece of Air India’s chatbot technology

In the 1950s, the Indian government took over the country’s flag carrier Air India, with JRD Tata, the airline's founder and Tata Group chairman, continuing as chairman for some time. However, after he stepped down, a period of stagnation began as the government management struggled to run the airline effectively. 

In January 2022, the Tata Group regained control of Air India and embarked on a major transformation journey. Last year, the airline placed the largest aircraft order in history, acquiring 470 narrow-body and wide-body planes from Airbus and Boeing to upgrade its fleet.  

Consequently, the airline is undergoing a period of digital transformation. 

Campaign speaks to Satya Ramaswamy, the chief digital and technology officer at Air India, on the sidelines of Qualtrics’ X4 Summit 2024, about the airline’s digital transformation which encompasses customer experience, backend systems, commercial and engineering operations, and a new digital backbone across the organisation. 

How Air India integrated generative AI

The airline started working with Microsoft in November 2022, after OpenAI’s ChatGPT models were made public.  

In May 2023, Air India launched its chatbot, initially named Maharaja and later renamed AI.g. It has performed spectacularly well for the airline, as since its launch, AI.g has answered around two million queries. It handles about 20,000 queries from 10,000 customers daily, and the accuracy of its responses is approximately 93%. 

Ramaswamy says that in his career, he has used technologies like Microsoft LUIS (Language Understanding Intelligent Service), but the performance of ChatGPT-based support is on an entirely different level for him.  

“The chatbot can handle general topics, airline-specific queries, and transactional questions. For instance, if a customer wants to make a reservation, it can gather the details and direct them to the appropriate page to complete the booking. It can also check the status of a complaint or help file a new one. Overall, AI.g can handle around 1,300 different topics,” explains Ramaswamy. 

“The chatbot has been a spectacular success for us for two main reasons. First, customers are more inclined to use the chatbot rather than wait for the call centre. Since the chatbot can answer 93% of the questions independently, escalations to chat agents are only about 1%, and even fewer are escalated to the call centre. This has made the chatbot very cost-effective and improved customer experience because consumers today prefer quick answers over browsing pages for information.” 

However, Ramaswamy admits there were challenges initially as Air India familiarised itself with this new technology. Since then, the airline has now developed expertise in this area.  

“We have reached a point where other major global airlines, including some leading American airlines, are contacting us to learn how to implement ChatGPT-based chatbots,” Ramaswamy says. 

For Air India, customer experience has three critical touchpoints. First, there are digital touchpoints. Second, there are physical touchpoints, primarily in the airport, where customers interact with the airline. Thirdly, the in-flight experience (IFE) is crucial, especially since its flights from India can take up to 16 hours.

Ramaswamy explains that the website, mobile app, chatbot, and notifications are all important in terms of digital touchpoints. Notifications are essential for keeping customers informed about their booking status, check-in counters, boarding gates, flight changes, and modifications.  

“The chatbot has proven crucial among these digital touchpoints because it understands customer queries exceptionally well without requiring extensive training on every variation,” says Ramaswamy. 

“Customers find interacting with our LLM-based (large language model) chatbot far less frustrating than traditional chatbots. As a result, even major American carriers have consulted us to better understand this.” 

For example, AI.g can handle questions like, 'I want to take my German Shepherd on the flight. Will you allow that?' Traditional chatbots would need explicit training on terms like 'German Shepherd,' but Air India never trained AI.g specifically for different breeds. Instead, the airline simply ingested its pet policy into the chatbot.  

With its ability to reason, the chatbot understands that a German Shepherd is a dog and a dog is a pet, so the pet policy applies. The chatbot comprehensively responds, outlining the physical restrictions and the full pet policy. 

Ramasamy explains that Microsoft’s technology, particularly Retrieval-Augmented Generation (RAG), has been a game-changer for Air India. RAG is a method used to improve the precision and dependability of generative AI models by incorporating factual information retrieved from external sources. Using RAG prevents the chatbot from generating inaccuracies or misinformation for Air India's customers.

In addition, the ability to ingest natural language documents and apply general knowledge to reason through chains of logic has been transformative for Air India. 

“We measure the chatbot’s success in two ways. First, we explicitly ask customers if their questions were answered satisfactorily. Positive responses exceed 80%, and sometimes customers don't respond because they have received the answer they needed and simply move on,” says Ramasamy. 

“Second, we track the percentage of queries the chatbot escalates to a human agent. Currently, only 7% of queries are transferred, meaning that 93% of the time, the chatbot handles the questions independently.” 

Ramasamy continues: “The chatbot also ensures a consistent customer experience, providing uniform answers and a consistent tone across various questions, unlike human agents who may offer varied responses. This consistency is crucial and has proven to be a game-changer for us.” 

Ensuring data privacy when integrating AI

As travel returns to normal, airlines are increasingly targeted by fraudsters, exploiting vulnerabilities in their operations through various tactics.  

Research from 2022 shows that airlines account for 46% of all fraudulent online transactions, making them particularly vulnerable due to their reliance on digital sales of high-value tickets. 

For example, some of the most common frauds are bookings made with stolen credit cards or through scam travel agencies, bad actors accessing frequent flyer accounts to steal points, and travellers disputing legitimate charges to avoid paying for trips. 

Airlines face challenges in ensuring data privacy while integrating AI, given stringent industry regulations and the critical importance of protecting sensitive customer information. Manual fraud prevention methods involving extensive rules are resource-intensive and often lag behind emerging threats. 

Ramasamy notes newer-generation technologies like RAG need to limit company-related data to the company and not expose it. He explains the systems of record that the chatbot interacts with, like reservation systems and Salesforce for customer service, are protected by existing data layers and privacy protections. The chatbot cannot bypass these systems to access anything beyond what is required to answer customer queries. 

In addition, the airline also has a bespoke layer of safety by ensuring that the chatbot, for example, does not hallucinate or provide inappropriate information. Air India uses both ChatGPT's built-in mechanisms and its own safeguards. 

"We constantly analyse the chatbot's performance. Suppose a user gives a thumbs-down to indicate dissatisfaction with a response,” explains Ramasamy. 

“In that case, we first conduct an automated analysis to group similar issues and then perform a human study to understand why the response was incorrect and how it can be improved. So, we have multiple layers of safety to ensure privacy and the appropriateness of the chatbot's responses.” 

Upcoming innovations with generative AI

Air India believes generative AI must be leveraged across the board, both on the consumer and enterprise sides. For example, the airline has developed a copilot for on-time performance (OTP) with Microsoft because OTP is a crucial parameter in the airline industry. 

OTP tells airlines the percentage of flights that took off and landed on time. It is a vital metric because it encapsulates customer experience and operational excellence in a single parameter. 

To ensure a plane takes off on time, a whole host of operational tasks need to happen seamlessly: the aircraft needs to be available, arrive on time, and be cleaned promptly; fuel must be ready, and the pilots and cabin crew must be available. 

This single metric shows that everything has been done correctly. From a customer experience perspective, passengers greatly value flights taking off and landing on time, which accounts for around 80% of airlines’ customer satisfaction score.  

Air India conducts extensive surveys using Qualtrics to monitor this. Initially lacking situational awareness, the airline now sends 1.5 million NPS surveys to customers via Qualtrics, receiving 70,000 responses that provide valuable insights into the pre-boarding, in-flight, and post-flight experiences. 

This data allows Air India to pinpoint issues based on departure points, aircraft types, and flight crews. For example, food-related problems at Mumbai Airport can be addressed promptly.  

“The OTP copilot pulls data from all operational systems to analyse the performance of flights taking off and landing. Users can query this information using natural language,” explains Ramasamy. 

“For example, a San Francisco airport manager or someone at the headquarters in Delhi could ask, ‘How was OTP in San Francisco today?’ and receive a natural language answer, such as, ‘Flight A1234 took off on time, but Flight B5678 didn't due to XYZ reason’. They can then dig deeper into the data to determine the specific cause, whether it was ground handling, catering delays, a slow check-in process, or another issue.” 

The copilot operates entirely within Microsoft Teams, meaning users can immediately call the airport manager or ground handling staff to resolve the problem.   

Ramasamy notes that dashboards have traditionally been static and may show the top three delays, but users’ questions are not static, as they might change depending on weather conditions.  

For example, suppose there was heavy snow in the eastern United States. In that case, a user might ask, ‘How many flights were affected by the snow today?’ or ‘How many flights were affected due to pilot availability?’ 

“The system's ability to handle such varied queries and retrieve all relevant data makes a significant difference and has been a game-changer,” explains Ramasamy. 

The road ahead for Air India’s digital transformation journey

Air India’s ultimate goal is to become the world's most technologically advanced airline, both on the consumer and enterprise sides. The airline believes it has a shot at achieving this because it has access to a vast pool of software talent in India. 

In addition, the airline has allowed Ramasamy, who was an entrepreneur based in Silicon Valley before joining Tata Group in 2010, to remain in San Francisco and build a team that can learn from the innovative ideas emerging from the Valley. 

Ramasamy is also part of Stanford University’s academic ecosystem and an advisor to many startups, which allows him to take ideas and implement them quickly. 

For example, the creation of AI.g happened quickly because Air India realised OpenAI had released ChatGPT, and the airline wanted to be among the first to adopt it. The airline saw ChatGPT’s potential and jumped in to make it happen with the help of partners like Microsoft, Qualtrics, SAP, and others. 

“Compared to traditional airlines, since we are building our infrastructure from the ground up, we can become the most technologically advanced airline in consumer and enterprise areas. On the consumer side, we have six key channels: the website, mobile app, chatbot, in-flight entertainment, flight data analysis, and notifications. We aim to excel in all six channels,” says Ramasamy. 

“Our mobile app features a unique flight status capability, showing the flight status and the status of the incoming aircraft. In India, where passengers often worry that they won't see the plane at the gate, this feature has significantly reduced tension and improved the travel experience. It won the Gold Stevie Award in Asia this year.” 

Air India is also approaching in-flight entertainment (IFE) uniquely. IFE is not usually considered a proper consumer channel, but with all aircraft becoming internet-enabled, the airline sees it as an opportunity for targeted promotions. By understanding the customer's viewing preferences, Air India can deliver tailored advertisements. 

The airline is also opening its Application Programming Interface channels to other consumer-facing properties and channels so they can access Air India’s data and provide value to their customers. 

“We have made significant progress on the enterprise side by ultimately building new systems from scratch without disruption. Our philosophy has been 'transformation without disruption.' We've achieved all this progress without any significant interruptions,” explains Ramasamy. 

“In a few years, we will be the most technologically advanced airline in the world and definitely among the best.” 

Campaign Asia

Related Articles

Just Published

3 hours ago

Revealed: Uncommon is majority shareholder in ...

Uncommon, now majority owned by Havas, backed former creative director Josh Tenser to launch Calling with Rani Patel.

3 hours ago

Healthcare and offshore betting ads emerge as most ...

With 85% of objectionable ads coming from digital media, the report also found that online safety continues to be a concern.

3 hours ago

Why Tessa Ohlendorf left agency life for artificial ...

The former managing director of MediaMonks started Fabric Folks to help agencies adapt to the new AI era.

3 hours ago

Why the creator economy could take over the ...

There is a distinct possibility the creator economy may be out to eat agencies’ lunch.