We’ve all been there: frantically checking a flight’s status, only for the information to be out-of-date or simply wrong. Whether searching, shopping, booking or traveling, we consumers wear many hats — passenger, traveller, customer. And for us, accessing accurate and up-to-date information is just as important as receiving it in a digestible format. Conversational AI excels at distilling information, but ensuring accuracy requires careful execution, particularly when relying on constantly updated data from multiple complex systems.
Think about it: the AI tool that guides your shopping experience on an ecommerce website is working in a relatively static data environment. Most of the everyday queries posed to ChatGPT aren’t context-dependent. But some operating circumstances and AI use cases are far more complicated. Airport environments, for instance, are rife with complex systems and dynamic data: flight schedules change by the minute, arrival and departure gate numbers can update without notice, and passenger services must adapt in real-time. In these unique, high-pressure situations, data quality and maintenance often determine whether AI assistants prove genuinely useful or just plain frustrating.
Building a Foundation with Stakeholders
Let’s start with the fundamentals: the data sources. The most important aspect of maintaining data quality for conversational AI systems is establishing close working relationships with all stakeholders. Many enterprise-level AI deployments involve dozens of interconnected teams, all of which need to be on the same page.
Take airports, for example, which must account for airline operations centres, ground handling services, security personnel and retail partners, among others. Each team manages critical data streams that passengers need access to through conversational interfaces. Without direct, consistent collaboration, AI systems risk delivering outdated or conflicting information that undermines passenger trust.
Partnership agreements must clearly define data sharing protocols, update frequencies and escalation procedures when systems fail. Teams should establish regular communication channels to discuss data anomalies, system changes and passenger feedback. When an airport updates its departure procedures, the AI system needs immediate notification to avoid directing passengers to the wrong gates.
For example, if Frankfurt airport temporarily limits its Fast Bag Drop service for a certain airline, the AI system should be able to access that information in real time. That way, a passenger asking about it can be directed accordingly, avoiding frustration. This collaborative approach means the difference between systems that adapt quickly to operational changes and those that consistently provide outdated guidance.
It Takes More Than a Connection: Maintaining API Integrity
Partner relationships are most important when considering structured data. In an airport environment, much of the information passengers seek from a conversational AI is specific, context-dependent and not publicly available. Flight details, arrival and departure times, gate assignments, passenger name records and itineraries represent the core data that is only accessible through airport systems. This information flows through partner APIs that must be connected to the conversational AI system during implementation. However, each connection requires ongoing monitoring to maintain accuracy and reliability over time.
API connections can fail for a number of reasons, from network issues to partner system updates that change data formats without notice. Monitoring systems should track response times, data completeness and format consistency across all connected APIs. Regular testing of API connections helps identify issues before they impact passenger experiences, while data validation protocols verify that incoming information matches expected formats and contains required fields. Finally, automated alerts can notify technical teams when APIs experience downtime or data quality issues, allowing for rapid response and system restoration.
Monitoring the Quality of Unstructured Data
But structured data ingested through APIs would only provide half of the picture to airport AI systems, and thereby to passengers. Conversational AI systems also collect unstructured data by scraping partner and relevant third-party websites. In airport environments, this is typically the airport website itself, but it might also include municipal government sites for transit schedules, food and beverage vendor sites for menus, and federal agency websites for passenger rules and regulations. This process requires careful guardrails to maintain accuracy and prevent the intake of outdated or incorrect information.
Web scraping presents unique challenges because website structures change frequently, content updates happen irregularly and information quality varies significantly across sources. AI systems can be trained to verify the data — this content validation should include checking publication dates, comparing information across multiple sources and flagging discrepancies for human review. Regular audits of scraped content help identify patterns in data quality issues and improve scraping algorithms over time.
Even with these preventative measures in place, this is the area that requires the most vigilance through ongoing analysis and user feedback.
Incorporating User Input and Feedback
User interactions are often the best source of real-time feedback about data accuracy and system performance, so any conversational AI system simply must have a process by which to collect and analyze this input. A natural language processing (NLP) pipeline transforms raw human language into structured data that machines can understand and act upon. This workflow enables conversational AI systems to code and structure the communication received from passengers.
In airports, passenger input feeds into the AI’s learning process to improve responses while also informing airport partners about passenger needs and preferences at any given moment. This insight enables a more responsive and personalised service, helping airports strengthen relationships with travellers.
When multiple users report the same incorrect information, it signals a data quality issue that requires immediate attention. Feedback loops should capture both explicit corrections from users and implicit signals like repeated questions about the same topic. This information helps identify gaps in the knowledge base and pinpoint areas where data sources may be unreliable.
And here’s where the real magic happens: revenue opportunities emerge through understanding passenger needs and preferences. A conversational AI could register and track an increased number of queries, say, for open tapas restaurants in the international arrivals terminal of Barcelona airport and communicate that insight directly to the relevant vendor through the airport. When conversational AI systems track which services passengers ask about most frequently, airports can optimise marketing for retail, food and beverage, parking, lounges and other services.
Curating a Reliable Knowledge Base
All of these practices contribute to building a comprehensive, accurate and reliable knowledge base from which conversational AI systems can draw. The quality of this knowledge base determines whether a conversational AI tool achieves high adoption rates, reduces strain on customer service staff and provides genuine utility to users — or gets disregarded after disappointing experiences.
Knowledge base curation requires ongoing attention to data freshness, accuracy and completeness across all sources. Regular audits should identify outdated information, gaps in coverage and opportunities to add new data sources that improve passenger experiences. The ultimate goal is to create a self-service model that enables passengers to get answers quickly while simultaneously reducing operational costs and freeing staff to focus on higher-impact activities.
We’ve been living with AI for over a decade, but we’re only just scratching the surface of what’s truly possible. In one of the world’s most complex data environments — air travel — AI is already making the journey in more than 80 airports more convenient for passengers and more profitable for airport operators. That’s not the limit of its potential either, but however far AI takes us, the commitment to data quality will determine how useful it is to everyday individuals like you and me.
Author bio:
Filipe Pereira is the co-founder and Chief Technology Officer of Airport AI, responsible for strategically guiding the technology that powers the world’s leading AI-powered customer service platform for airports. Under his leadership, Airport AI developed the market-leading conversational AI designed to improve the airport experience, which led to Airport AI’s recent acquisition by 15below to create a comprehensive, unified communication layer between airports, airlines, and their passengers. Prior to founding Airport AI in 2016, Filipe spent more than 3 years building web and mobile solutions for large enterprises and startups in Europe and Asia, including Gymondo GmbH, BNP Paribas Securities Services, and Red Ape Solutions.
