Project Snapshot
The airline industry has constant customer service demands. For our client, a major international airline, managing daily customer inquiries from different time zones and in many languages became increasingly expensive. In response, they decided to harness AI for airline operations and develop a round-the-clock customer service solution.
Between October 2024 and October 2025, software engineers from Intellias, a leader in travel technology and software development services, developed and deployed an advanced large language model (LLM) solution to process customer inquiries. An AI chatbot for airlines acts as a customer service assistant via a chat interface on the carrier’s website.
Unlike basic chatbots that provide only static answers, our client’s chatbot employs a multi-agent AI architecture. One way agentic AI is transforming the airline industry is by reducing manual labor on difficult tasks. For the airline’s chatbot, each agent within the system specializes in a functional area, such as booking or security verification. These agents use a centralized orchestration system to manage their shared memory, which helps the chatbot maintain context throughout long conversations. By calling the airline’s backend APIs, the chatbot handles customer service tasks in real time, such as processing refunds and checking passengers in for flights.
About the company
Our client is a major European airline that offers flights to more than 100 popular destinations worldwide, including in Europe, the Americas, Africa, and Asia. The carrier offers over 600 flights a day to more than 50 countries.
Business challenge
The client identified several areas where their existing support system did not meet their needs. A primary concern was high operating costs. For example, in customer service operations, personnel salaries accounted for a significant portion of their budget.
Furthermore, lack of scalability was a recurring issue. During peak travel seasons or in the event of unexpected cancellations, call centers became overwhelmed. As a result, customer service waiting times were long, the customer experience deteriorated, and brand loyalty faded. Consistency was also a problem. The quality of customer support often depended on an agent’s experience or current workload.
Additionally, providing 24/7 support in multiple languages required a massive global workforce that was financially difficult to justify. While the airline offered digital self-service tools, those platforms lacked the intelligence to perform multi-step transactions. If a customer wanted more than information already available in an FAQ, they were almost always forced to call a representative.
As a result, our aviation industry client encountered several significant hurdles with their existing customer support system related to:
- Transaction volume: The airline struggled to efficiently process distinct tasks without losing the conversation thread during periods with a high number of customer service inquiries.
- Resource availability: Customer inquiries needed to be classified into 200+ categories based on intent and were often misdirected, while the simultaneous processing of many inquiries caused the system to fail.
- Information retrieval: Customers frequently searched through extensive website content to find the answers they wanted.
- Data security: The client needed to ensure interactions with APIs were secure to prevent unauthorized access, especially when handling reservations and refunds.
- Performance monitoring: The client lacked a way to track detailed interactions and system performance to identify areas for improvement and ensure compliance with data policies.
Solution
While implementing AI for airline operations was new for our team, Intellias software engineers have extensive experience working with AI/ML models. They applied this experience to the emerging field of agentic AI in aviation to deliver a multi-agent system with retrieval-augmented generation (RAG) at its core. The system has specialized agentic workflows, and the architecture lets agents work as a team to achieve a higher rate of customer satisfaction. A system comprised of agents also eliminates the need for a monolithic customer service system.
Each agent has a specialized role:
- Translation agent: Manages multilingual requests by converting customer messages into Spanish (the internal processing language) and then translating responses back into the customer’s native language.
- Security agent: Serves as a safety layer by inspecting every incoming message for malicious intent.
- Intent classifier agent: Classifies customer needs according to a library of over 200 intent categories and routes each conversation to the appropriate specialist.
- RAG agent: Searches the airline’s website to provide accurate answers to questions about baggage policies or airport lounges.
- Transactional agents: Retrieve booking details via PNR numbers, manage refunds, process check-ins, choose upgrades, check flight status, and generate travel documentation requirements.
- Tone agent: Ensures that each response has a consistent brand voice and formatting regardless of whether the customer interacted with the system on the website or through WhatsApp.
When the system receives a message, it immediately scans for security threats and translates the text. If the message is safe, the intent classifier identifies its goal. An Amazon DynamoDB stores conversation files, Amazon Redshift provides analytics about them, and a custom framework counts LLM tokens and monitors expenses.
Business outcomes
By automating simple inquiries and complex transactions with AI agents, the airline achieved a new level of efficiency. The number of escalations to customer service agents dropped significantly, allowing personnel to focus on other tasks. As a result, the workload at the contact center became more manageable, and operational overhead costs dropped.
After implementing their new AI chatbot, the airline noticed a remarkable improvement in customer satisfaction. Passengers no longer had to wait in long phone queues for answers to basic questions because AI agents for airline customer service were available on the website 24/7.
Furthermore, the airline’s digital services became more inclusive for international travelers with the addition of multilingual capabilities. Because of the AI-powered customer service system’s scalability and flexibility, engineers can now add new agents without rebuilding everything. Finally, using cost-effective generative AI models for simple tasks and high-reasoning models for complex logic optimized the total cost of ownership.
KPIs: 50% reduction in agent escalations 24/7 global support 100% brand consistency
Technologies used
Amazon Bedrock, Azure OpenAI, LangChain, Python, FastAPI, Amazon DynamoDB, Amazon Redshift, Redis, AWS Fargate, Amazon Kinesis, AWS Lambda