­­­­­­­­

Traveltech has always been quick to adopt new systems - GDS platforms, online booking engines, mobile apps. Each wave promised efficiency and delivered it, mostly on the operational side. AI is doing something different. It is not optimising the existing system; it is replacing the logic underneath it. 

The shift is visible in the numbers. According to Amadeus, AI-powered tools are now embedded across booking, pricing, and customer service workflows at the majority of large travel operators. Machine learning engines handle dynamic pricing decisions in real time. Natural language interfaces are absorbing customer service volume that used to require large contact centre headcounts. And the pace of adoption has accelerated sharply since 2023, when large language models became capable enough to handle travel's notoriously complex, context-dependent queries.

For IT managers and C-level executives in mid-sized travel businesses - or in industries adjacent to travel, like corporate expense management, logistics, and hospitality - this is the moment to understand what AI is actually doing, where it creates competitive advantage, and what infrastructure decisions are now consequential. The companies that treat this as a vendor selection question are likely to make the wrong choices. The ones that treat it as an architectural question will be better positioned.This post covers the five most material applications of AI in travel technology today, with realistic examples drawn from the patterns now emerging across the industry.

Hyper-Personalisation: Beyond Preference Matching

The word "personalisation" has been in travel marketing for twenty years. What it meant for most of that time was relatively modest: remembered seat preferences, loyalty tier upsells, destination suggestions based on prior bookings. The underlying logic was rule-based. If a user had booked business class twice, show them business class options. If they had travelled to Mediterranean destinations, surface similar ones.

AI-driven personalisation is a different mechanism entirely. Modern recommendation systems do not apply rules — they build probabilistic models of individual traveller behaviour using continuous streams of input data: browsing patterns, search sequences, time of day, device type, how long a user pauses on a particular option before abandoning it, and how similar users with comparable histories have behaved. The output is a ranked set of options that the model believes this specific user, in this specific context, is most likely to book.

The business case is clear. According to research published by Amadeus, personalised travel recommendations delivered through AI increase ancillary revenue conversion rates and reduce the time from first search to completed booking. Neither of those outcomes requires the user to notice anything unusual — the interface looks the same, but the logic underneath it is no longer static.

Consider a realistic scenario: a corporate travel manager running bookings for a professional services firm. Historically, that function has involved policy enforcement — ensuring travellers book within approved fare classes and preferred suppliers. An AI-personalisation layer changes what "preferred" means. The system can learn that certain travellers consistently choose aisle seats on flights over two hours, that specific team members have a strong preference for hotels within walking distance of meeting venues, and that departures before 7am generate disproportionate expense claims for airport transfers. Rather than enforcing a blunt policy, the platform can surface options that are simultaneously compliant with company travel policy and aligned with demonstrated individual preferences. The result is fewer booking amendments, lower traveller friction, and better data on where policy and behaviour diverge.

The technical requirement is meaningful. AI personalisation at this level requires a data infrastructure that can ingest, clean, and model user behaviour in real time. For mid-sized operators, that often means a decision about build versus buy - whether to extend an existing CRM and booking platform with an AI personalisation module, or to migrate to a platform where that capability is native. DevPals works with travel technology teams navigating exactly this decision, and the right answer depends heavily on existing data architecture rather than on product marketing claims.

Disruption Management: The Invisible Recovery Layer

Flight disruptions are one of the most expensive problems in travel operations, and they are almost entirely unpredictable at the individual level even when aggregate risk is known. A severe weather system might delay 40% of departures from a hub airport — but which specific passengers are affected, what their onward connections look like, what alternatives exist, and what compensation they are entitled to are all questions that, until recently, required human agents to resolve case by case.

AI changes the economics of that problem. Platforms integrated with real-time flight data, GDS inventory, and passenger record systems can identify disruptions the moment they are confirmed, map all affected passengers, calculate each traveller's optimal rebooking based on their itinerary and fare class, and issue new tickets before the original flight has even been announced as cancelled. This is not a hypothetical capability. Cirium's analysis of AI deployment in travel operations documents this pattern as an active use case at major carriers and OTAs.

The scenario plays out like this: a connecting passenger lands at a hub thirty minutes late, which means their onward connection is no longer feasible given minimum connection time. A legacy system would surface this problem to a gate agent, who would then work through available alternatives manually. An AI-powered disruption system has already identified the problem, pulled the passenger's record, found the next available flight on a compatible fare class, and sent a rebooking notification — before the passenger has left their seat. If no alternative on the same carrier exists within an acceptable window, the system can also check interline agreements and surface options on partner carriers. The passenger experience changes from standing in a rebooking queue to receiving a push notification.

For enterprise clients with managed travel programmes, this capability has a direct financial value that is reasonably easy to quantify. Disruption-related costs - emergency rebookings at full fare, hotel accommodation, traveller time loss, and TMC service fees — are measurable. AI-powered disruption management reduces each of these. The harder question for IT managers is integration: these systems require real-time data feeds from airline inventory systems, which means API integrations that need to be maintained, monitored, and tested against live disruption events. DevPals has built resilient integration architectures for travel clients in precisely this context, and the lesson is consistent — the value of the AI capability is only as reliable as the data pipeline feeding it.

Predictive Pricing and Revenue Optimisation

Dynamic pricing in travel is not new. Airlines have used yield management systems since the 1980s. What AI adds is not the concept of dynamic pricing — it is the sophistication of the demand forecasting that drives it, and the speed at which pricing decisions can be made and updated.  Traditional yield management models worked with relatively simple inputs: historical booking curves, seasonality factors, and capacity data. They were recalculated periodically — sometimes daily, sometimes weekly. Machine learning pricing engines work differently. They ingest a far wider range of signals in real time: search query volumes on metasearch platforms, competitor pricing across distribution channels, weather forecasts, local event calendars, social media sentiment on destination popularity, and macroeconomic indicators that correlate with leisure travel demand. The model updates pricing continuously, and the pricing update is applied at the individual session level — two users searching for the same route at the same time may see different fares based on their browsing behaviour and inferred purchase intent.

According to research cited by both Amadeus and BuiltIn, AI-powered revenue management systems consistently outperform traditional yield management tools on revenue per available seat or room, particularly during periods of demand volatility. The improvement is most pronounced in situations that would cause a traditional model to behave conservatively — unexpected demand spikes, new route launches without historical data, and post-disruption demand redistribution.

For hotel and hospitality operators, the same logic applies. Rate optimisation systems now factor in demand signals from multiple sources simultaneously: search volumes on OTA platforms, booking pace relative to historical curves, local competitor rates pulled via scraping, and — increasingly — predictions about demand from events that have not yet generated bookings but that the model has learned to anticipate. A hotel that previously adjusted rates twice a day based on occupancy can now run continuous optimisation that responds to signals hours before they would have been visible in a traditional system.

The governance question this raises is significant and often underweighted in vendor discussions. When pricing is managed by a machine learning model, the model's behaviour during edge cases — extreme demand, sudden competitive changes, regulatory events — needs to be understood and bounded. IT leaders evaluating AI pricing solutions should ask vendors for documented examples of model behaviour during demand shocks, and should insist on human-in-the-loop overrides for pricing decisions above defined thresholds. The efficiency gains from AI pricing are real; so is the reputational risk of a model that prices opportunistically in ways that create regulatory or press exposure.

Agentic Travel Planning: The Virtual Agent Becomes Useful

Chatbots in travel have had a poor reputation for most of the decade. Early implementations were decision-tree systems dressed up with conversational UI — capable of answering a narrow set of FAQ queries and failing expensively on anything outside their training set. Customer satisfaction data on those deployments was generally negative, and many operators quietly retired them or reduced their scope to handle only the simplest queries.

The current generation of AI agents is a different technology. Large language models can handle free-text inputs of significant complexity, maintain context across a multi-turn conversation, access live inventory and booking systems through API integrations, and produce structured booking outputs from unstructured natural language requests. The difference in capability is substantial enough that the user experience is qualitatively different, not incrementally better, but functionally different.

A realistic scenario illustrates this. A travel manager at a professional services firm needs to organise a five-day team offsite for twelve people, coordinating flights from three different cities, hotel accommodation, ground transport, and a working dinner. Historically, this task would require a TMC agent spending several hours across multiple booking systems, with back-and-forth email to collect individual traveller preferences and confirm details. An AI travel agent can take a natural language brief — "twelve people, arriving from London, Frankfurt, and Dubai, four nights in Lisbon starting 14 October, need a hotel with meeting space, budget around £4,000 per person" — and return a structured itinerary with options, fare breakdowns, and a booking flow, within minutes.

Amadeus documents this pattern as an active development area, with platforms building agentic capabilities that can handle multi-leg, multi-traveller itineraries from free-text input. The shift has implications for TMC relationships, corporate travel policy, and the role of travel managers themselves — a function that has historically been defined by its access to booking systems and supplier relationships may find those advantages narrowing as AI agents provide comparable access at lower cost.

For IT managers evaluating these systems, the critical integration questions are around data security and policy enforcement. An AI agent that can book travel on behalf of employees needs to operate within defined policy guardrails — fare class restrictions, preferred supplier lists, approval workflows for above-threshold spend. Building those guardrails into an AI agent architecture requires more careful design than configuring them in a traditional online booking tool, because the input is unstructured and the agent's interpretation of policy language needs to be tested against a wide range of edge cases.

Invisible Hospitality: Smart Hotels and the Connected Stay

The hotel side of travel AI is developing along a different axis from the booking and pricing applications described above. The focus is on the physical stay experience — check-in, room environment, service requests, and concierge — rather than the commercial transaction that precedes it.

Smart hotel systems use AI to reduce friction at every touchpoint of the stay. Mobile check-in, already common, is now being extended to AI-powered room assignment that factors in guest preferences, floor plan, and current occupancy patterns to maximise the probability of an upgrade or preferred room type being available at arrival. In-room systems can learn guest preferences across stays — preferred temperature, lighting levels, preferred news sources on in-room displays — and apply them automatically on subsequent visits without the guest needing to configure anything.

The more significant development is in connected room environments. Sensors that monitor occupancy, temperature, and energy consumption allow properties to manage their physical plant significantly more efficiently - a room that has been unoccupied for four hours can have its HVAC dialled back automatically, with restoration timed to match the expected return of the guest. According to Endava and Intellias, smart hotel implementations using AI-driven building management are achieving material energy cost reductions while simultaneously improving guest satisfaction scores, because the system responds to occupancy rather than operating on fixed schedules.

The enterprise angle here is relevant for hospitality IT teams and for corporate clients negotiating preferred hotel programmes. Properties with mature AI infrastructure can offer corporate clients richer data on how their travellers are actually using accommodation - occupancy patterns, service requests, meeting room usage — that can inform travel policy and negotiations. The hotel that can tell a corporate account that their travellers consistently leave before the breakfast service and rarely use the gym is offering a more useful commercial relationship than one that can only report nights booked and rate paid.

Scenario: A Mid-Sized Tour Operator Rebuilds Its Tech Stack

To make this concrete: consider a mid-sized tour operator running a mix of packaged and semi-custom holidays, with a customer base of several thousand active bookers and a technology setup built around a GDS integration and a CRM that is five years old.The operator's problems are familiar. Pricing is managed manually against a rate card updated weekly. Customer service handles a high volume of status queries that could be automated. Conversion rates on the website are below industry benchmarks because the recommendation logic is basic. And disruption events — which happen several times per season — generate significant agent overhead and occasional customer compensation costs.

An AI transformation programme for this operator would not start with replacing the booking engine. It would start with a data audit: what signals exist, where they live, how clean they are, and what integrations would be required to make them available to a machine learning layer. From that foundation, a pricing optimisation module can be integrated — using existing GDS inventory data plus external demand signals — before any customer-facing AI is deployed. Once the pricing layer is stable and understood, personalisation and chatbot capabilities can be added incrementally, each tested against baseline conversion and satisfaction metrics before being scaled.

DevPals has structured programmes precisely for this kind of phased AI adoption — identifying where data infrastructure needs to be strengthened before AI can be effective, and building the integration architecture that connects AI capabilities to existing operational systems without requiring a full platform rebuild.

Conclusion

AI in travel technology isn't a single technology or a single vendor decision. It's a set of capabilities - personalisation, disruption management, predictive pricing, agentic planning, smart hospitality - each of which requires different data inputs, different integration architectures, and different governance frameworks to operate reliably.

The companies getting genuine value from these capabilities share a few characteristics. They invested in data infrastructure before investing in AI applications. They built clear governance around how AI systems make decisions and what human oversight looks like. And they treated AI adoption as an iterative programme rather than a one-time implementation.

For IT leaders in travel and travel-adjacent businesses, the takeaway is straightforward: the competitive window is not closed, but it's narrowing. Operators who build the data and integration foundations now will be in a position to deploy AI capabilities faster as they mature. Those who wait for the technology to "settle down" will find themselves integrating into a market where the leading operators have already compounded the advantages of early investment.The question worth asking now is not which AI vendor to choose. It's what your current data infrastructure can support, and what it would take to make it AI-ready.

If you are working through that question, DevPals can help. Our team specialises in travel technology architecture - from data infrastructure assessments to end-to-end AI integration programmes. Reach out to our experts for a diagnostic conversation about where your organisation stands and what a realistic path forward looks like.