Challenges in Scaling AI Agents for Travel Industry

The Gap Between Demonstration and Deployment

AI agents often shine in controlled demonstrations, showcasing their potential capabilities. However, the real challenge arises when these agents are implemented in live scenarios, where they interact with actual customers and face unpredictable demands.

Research indicates that while 33% of travel operators are testing AI agents, only a mere 2% have successfully scaled them. This highlights a critical question: what hurdles exist when transitioning from a controlled experiment to full-scale production?

According to Erik Zahnlecker, the travel industry lead at Sierra, a conversational AI platform, production environments introduce a myriad of technical demands that go beyond initial demos. In a demo setting, agents function in isolation, free from the complexities of live systems which can impact performance and responsiveness.

Moving from Concept to Reality

The progression from a demo to production is multifaceted. Demonstrations typically utilize simulated datasets, while proofs of concept push the boundaries by testing an agent’s performance in a restricted environment. Full-scale production necessitates integration with real APIs, managing live interactions, and providing support for potentially millions of customer queries each year.

Internal support is also crucial; companies need to ensure they have a solid business rationale for this transition. Sierra engages in proofs of concept to showcase tangible returns on investment and to pave the way for broader adoption.

Deployment tends to occur in phases. Initially, a system might manage a specific set of customer queries or 5% of the traffic on a particular platform. As confidence and reliability improve, the system gradually takes on more responsibilities.

The Complexity Beneath the Surface

Sierra uses the “agent iceberg” analogy to illustrate the numerous components necessary for a production-ready AI agent. Above the surface are the elements commonly recognized, like large language models (LLMs) and retrieval-augmented generation (RAG). However, the foundational work crucial for success often lies below the surface.

This hidden infrastructure involves integrations with core business systems, runtime governance, context management, and ongoing testing and optimization. Key concerns include avoiding errors, ensuring compliance with brand messaging, and selecting the right operational models based on task complexity.

The rapid evolution of technology further complicates matters. For instance, Sierra regularly releases around 17 platform updates weekly, posing a challenge for companies to keep pace.

Travel Sector Specific Challenges

The travel industry presents unique challenges that can strain conventional service systems while demonstrating the necessity for robust AI agents capable of coping with real-world scenarios.

Legacy systems in airlines, often built on outdated mainframe technology, create barriers for AI integration. AI agents can serve as a modern, conversational interface, yet still need to interact seamlessly with these older systems.

Volatility in travel demand can also surge significantly during events like weather disturbances, overwhelming traditional support teams. AI agents, designed to scale, can efficiently manage these spikes.

Moreover, the travel sector’s multilingual requirements can make staffing challenging. For instance, Philippine Airlines employs an AI agent capable of switching languages mid-conversation, helping serve diverse customer demographics effectively.

Leveraging Talent for Optimal Outcomes

To effectively scale AI systems, organizations need the right talent behind them. Current data shows that only 3% of the travel industry workforce is AI-trained, highlighting a critical shortfall in AI expertise.

Erik Zahnlecker emphasizes that the focus should be on the type of AI expertise that travel companies truly need to leverage this technology efficiently. Sierra collaborates with in-house data teams to harness their existing knowledge effectively, ensuring the best value is derived from their investment.

Deciding Between In-House Development and Partnership

Travel operators face a fundamental decision: what to build internally and what to outsource. The general recommendation is to maintain core customer relationships while leveraging external expertise for advanced technology.

Zahnlecker notes that a successful relationship with AI providers requires clarity on operational functionality, independence of the company’s team, and aligned commercial interests between both parties.

A rigorous assessment should ensure that the chosen technology integrates seamlessly with existing systems while being adaptable to fluctuations in customer volume. Furthermore, collaborating on outcome-based pricing can help align incentives, leading to a mutually beneficial partnership.

By navigating these considerations, technology leaders can better position their travel organizations to capitalize on AI’s potential while ensuring operational reliability and scale.