Planning a Perfect Trip with an AI Assistant — Part 2
This is part 2 of a two-part series in which I unpack the impact of Artificial Intelligence(AI) on the online travel planner’s complicated customer journey. If you haven’t read Part 1 of this series, I recommend checking that out first!
In the previous installment of this blog, we double-clicked into the user side of the AI-driven travel planning customer journey and looked at potential use cases travelers might run through. In this part, we will be looking at what capabilities the apps and platforms need to be well-versed in to respond to those needs effectively.
Hello World! How can I help you?
To start with, in the traditional travel planning universe the technology platforms treat the two sides of the search & discovery framework in a silo with an incomplete understanding. Neither do they have a complete handle on the users engaging with them, nor do they have a full understanding of what they are recommending & pitching to them(be it a luxury hotel or a long-haul flight!) If the new breed of AI tools want to be significantly better than the traditional tools, they need to be very tight on both sides of the recommendation bridge:
- Understanding WHO they are suggesting recommendations for
- Possessing a very holistic view of WHAT they are recommending
Nailing both sides is very critical since if recommendations aren’t personalized enough, AI-powered itineraries(or suggestions) risk becoming merely a rebranded version of generic search results. Below are the critical capabilities the AI platforms need to nail if they aspire to craft itineraries that are robust as well as highly personalized
A. Understanding the Traveler Persona
Getting a holistic read on who is asking for assistance is extremely important in any scenario where an expert is helping a person needing support. And in the AI domain, it is no different. Users might share them explicitly(via key phrases in the prompt they share or their profile) or implicitly(need to read between the lines!) Independent of how they are shared, below are some of the attributes to scan for:
- Demographics: The most basic attributes like Age, Gender, Nationality, Occupation, LGBTQ, etc along with trip-related properties like if they are traveling Solo, Couple, Family with kids, etc
- Psychographics: In the context of travel planning user personas, psychographics delve deeper than demographics to understand the “why” behind a traveler’s choices. It focuses on the psychological aspects that influence their travel decisions and preferences. Here’s how it applies to travel planning:
- Values: What is important to the traveler when they travel? Do they prioritize sustainability and eco-friendly options? Are they looking for cultural immersion and authentic experiences? Understanding their values helps tailor recommendations to their priorities.
- Personality Traits: Is the traveler outgoing and adventurous, seeking thrill-seeking activities and new experiences? Or are they introverted, preferring peaceful destinations and relaxation? Knowing their personality helps suggest destinations and activities that match their comfort level and travel style.
- Travel Fears or Concerns: What worries the traveler about planning or taking a trip? Is it safety concerns, language barriers, fitting everything into a budget, or feeling overwhelmed by planning logistics? Addressing these anxieties through recommendations and resources can increase their comfort level and confidence in booking a trip.
Here’s an example: Imagine two user personas: Sarah (32, professional, single,) and David (50, married with children). Demographics alone might suggest similar travel options: beaches, and resorts. Psychographics can reveal the difference: Sarah (outgoing, adventurous, values cultural immersion) might prefer exploring Southeast Asia with unique experiences, while David (family-oriented, prioritizes relaxation) might choose an all-inclusive Caribbean resort.
By incorporating psychographics, the platforms can create a more nuanced understanding of the target audience and generate plans that resonate with their deeper desires and concerns.
3. Preferences: How they typically prefer to experience a destination. Includes exploration interests (sightseeing, cultural experiences, foodie trails, outdoor adventures, etc.), activities interests (surfing, diving, trekking, etc), or even transportation preferences(flying, driving, cruises, etc.)
4. Fitness profile: Understanding how active or fit the traveler is. This coupled with Demographics will help in adjusting a plan appropriately. For example, younger people and kids can travel faster and cover the travel areas quickly but if you are traveling with older people, you can customize your plan accordingly. Sometimes, users might directly prompt with situations like motion sickness, or special health conditions, in which cases the tool needs to avoid proposing irrelevant activities.
5. Budget profile: What kind of luxury levels do they typically prefer: Budget, Mid-range, Luxury, etc. And this shouldn’t be purely about the price tag as for some users value for money is more important than the price alone, and might stretch if a great deal is available. An Intelligent app should be able to surface a recommendation like ‘great luxury hotel that sells at $200 usually is pegged at $100 now, great deal!’ to value-for-money hunters across various budget profiles.
6. Travel Planning Behaviours (optional): Peripheral things like what kind of Information sources (travel blogs, social media, guidebooks, etc.) influence decision-making or Booking Preferences (online travel agencies, directly with airlines/hotels, etc.)
Suppose some of these are not clear enough from the user’s prompts. In that case, I’d highly recommend the apps proactively seek clarifications around these(further inputs from users) before jumping into suggestions. For example: It would be good to know what kind of passport a user holds, before recommending a dozen destinations he/she cannot travel to easily without procuring a Visa!
B. Grasping all the Trip Intents
In a traditional Search experience (either a text-based search engine like Google or a Search form in OTAs like Booking.com) the frameworks fueling the information retrieval dont typically go beyond understanding the query. While Query Understanding is important, it is more crucial to comprehend the User Intent that leads to performing the search in the first place. Similarly, if AI tools need to fulfill travel-related information requests satisfyingly, then they need to decipher the Trip Intent of the user firing a prompt.
Grasping the trip intent is decoding driving factors like
- Purpose of Travel: Is it a leisure vacation, business trip, honeymoon, or adventure expedition?
- Travel Motivations: What is driving the traveler to go on a trip? Are they seeking escape from routine stress, wanting to reconnect with family, or interested in personal growth through cultural exploration? Understanding their motivations allows for suggesting trips that fulfill those desires.
Understanding the trip intent is a crucial skill required for AI travel tools aiming to deliver personalized travel plans and itineraries. Here’s why:
- Personalization: Recognizing the intent behind a trip allows the AI to tailor recommendations to suit the traveler’s preferences, interests, and goals. For example, a business trip requires different accommodations and activities compared to a leisure vacation or a family reunion.
- Relevance: By understanding the trip intent, the AI can filter out irrelevant options and focus on suggesting activities, accommodations, and experiences that align with the traveler’s purpose. For instance, if the intent is a romantic getaway, the AI might prioritize suggesting cozy accommodations and intimate dining establishments.
- Contextual Recommendations: Knowing the trip intent enables the AI to provide contextual recommendations that enhance the overall travel experience. For example, if the traveler’s intent is adventure travel, the AI might suggest outdoor excursions, adrenaline-pumping activities, and off-the-beaten-path destinations. Or if the intent is a family vacation with children, the AI might recommend kid-friendly attractions, accommodations with childcare facilities, and family-friendly dining options.
It is important to note that even if the system has a clear understanding of the traveler’s typical preferences context of the trip being planned is a critical factor to be considered. A user might have traveled on plenty of business trips to a typical business destination engaging in conservative accommodation and transportation options. But might be planning an anniversary trip with a partner intending to go on a romantic getaway. So, the recommendations need to be adjusted contextually.
Overall, the capability to understand the trip intent is essential for AI travel tools to deliver personalized, relevant, and contextual recommendations, making the planning experience more enjoyable and efficient.
Below is a nice capture (from McKinsey’s whitepaper called Promise of Travel in the Age of AI). At the most focused level, hyper-segmentation takes into account multiple characteristics of an individual, creating a comprehensive view of a “segment of one.”
C. Gaining a holistic view of recommended items
This involves the system acquiring a deep understanding of the various things being recommended as a part of the plans to the users as destinations, hotels, F&B establishments, etc. Instead of getting shortsighted by only looking at static content(from traditional travel guides and owners of these items), the platforms should also assess broader User Generated Content(UGC) from across various channels(from sources like Google and TripAdvisor to Blogs and Forums). This will empower the app to not just recommend something based on ‘absolute’ value or score, but also weigh in other factors identified from the Traveler persona and Trip Intent. For example, a destination or hotel recommended in a specific season to a solo traveler interested in diving might not be the most relevant to a romantic vacation planner looking for a quieter holiday in a different season. Here are some of the factors the system needs to weigh in and accommodate in its recommendation logic.
- Static Factors: These are core attributes of a travel product that fundamentally define it. For a hotel, this could be things like star rating, amenities and facilities, style, etc. For an airline it could be things like full service, low-fare/budget, included ancillaries, etc. AI could significantly outshine traditional travel tech if it can assimilate beyond the basic definition and connect the dots. The ability to infer characteristics even if it is not mentioned. Like classifying a hotel as a boutique and tagging it as great for romantic couples based on various representative elements. Or drawing a similarity graph based on overlapping attributes to recommend alternate options with a lesser spend. Example: If a user is keen on a Marriott or a Hyatt property recommend alternates that match these in all aspects (and will provide a similar guest experience) but at a much cheaper cost. Even a human expert will not be able to nail this unless they have deep knowledge about all the hotels!
The AI platforms also need to have an accurate knowledge of all the operational aspects of what’s being recommended as part of the plan like Opening hours of Attractions and F&B establishments, Schedules of Flights and Ground transportation, or Border crossing restrictions.
For example, you dont want the AI assistant to recommend a specific restaurant for a day(or time of day) when it is closed or a train connection on a day when it is not operational!
2. Subjective Factors: These are dynamic parameters that need to be double-checked to ensure an ideal experience on the ground. Could be categorized into the following areas:
- Physical Hardship: The activity level and fitness required to experience an attraction need to be analyzed well before it is recommended. Especially in the context of the traveler profile. This way the tool can suggest strenuous hikes for adventurous travelers while offering scenic walks or cultural experiences for those who are older or prefer a slower pace. Here is an example: One of the AI tools I used to generate a plan for Edinburgh recommended climbing Arthur’s Seat — an extinct volcano that sits right in the heart of the city. While it is a unique attraction with magnificent views, there was no guidance or advisory about the climb involved! The hike up Arthur’s Seat is short but strenuous with some scrambling at the top. It’s achievable for most people with a good level of fitness, but not for everyone — and not without appropriate footwear. Also, an activity like this should be recommended at the right time of the day and not towards the end of the day.
- Weather and Seasonal Factors: The climate of the destination during the travel period is an important element to be weighed in. For example, avoid recommending strenuous activities in scorching deserts during peak summer months or have a very extended daily itinerary when daylight hours are limited during winter. Recurring weather patterns and predictable natural disaster risks in the destination also need to be. For example, avoid recommending travel during hurricane season in the Caribbean or monsoon season in Southeast Asia.
As much as avoiding, this should also involve offering the appropriate experiences and tips based on weather elements. For example, suggest snowshoeing or skiing during winter months in a mountain destination, or advise packing rain gear for a trip during the rainy season.
Seasonal elements include things like events, festivals, or other functions that might be taking place at the destination during the time the traveler plans to visit. For instance, suggest attending a tulip festival in the Netherlands during springtime or visiting a Christmas market in Europe during the winter holidays. At the same time, also warn the travelers about potential friction points to be expected due to the crowd.
Below are a couple of samples of carefully curated recommendations when a tool weighs in these factors appropriately:
A couple is planning a hiking trip to Glacier National Park in Montana. The AI assistant recommends traveling in July based on their preferences. However, it also informs them about potential afternoon thunderstorms typical in the park during that month. The AI can then suggest packing rain gear and checking the latest weather forecast before each hike.
A solo traveler wants to experience the vibrant culture of Rio de Janeiro, Brazil. The AI assistant recommends attending Carnival in February, a world-famous festival known for its extravagant costumes and lively parades. However, the AI also flags potential issues like high accommodation prices and large crowds during this peak season. The AI can then offer alternative options for experiencing Rio’s culture outside of Carnival or suggest booking accommodations well in advance if the traveler is set on attending the festivities.
- Dynamic Conditions and Closures: These are unplanned and atypical occurrences that could impact a travel plan. Like Transportation Disruptions, Temporary closures, Administrative restrictions, etc. For example, dont recommend hotels that are under renovation or Points of Interest that are temporarily closed during the time of visit.
By incorporating all these factors, the AI travel assistant can provide up-to-date and adaptable travel recommendations, ensuring a smooth and enjoyable experience for the traveler.
D. Suggesting a personalized plan
If an app has managed to crack all the elements called out above, it then needs to stitch them all into a full-blown recommendation. This is where it needs to apply all the ‘intelligence muscle’ acquired based on a complete assessment of the recommended items and the user type. Like the Edinburgh example mentioned earlier, make sure you stitch the right points of interest in the right order into an itinerary.
On top of this, a good itinerary not only needs to take into account what a visitor wants to see but also if it’s feasible. Meaning, taking into consideration logistical aspects like how the person will get there if the various points of interest would be open when they get there, and if reservations are needed, among other variables. This is where AI tools seem to be currently overlooking some key factors as opposed to human agents.
Finally, transparency and trust are key aspects of a good user experience with AI tools. It is highly advisable to include the appropriate citations and sources in the right places(like how apps like Perplexity or MS Copilot do) for the following reasons:
- Accuracy and Verifiability: Citations allow users to evaluate the credibility, reliability, and accuracy of the information provided by the AI tool.
- Deeper Exploration: Citations could offer users a starting point for further research on topics of interest. This empowers users to delve deeper into the information and gain a more comprehensive understanding, assuming the AI tools embed this within their canvas.
- Building Trust: Transparency builds trust with users. Knowing where the information comes from fosters a sense of confidence in the AI tool’s capabilities. And even prevent users from leaving the tool to check other trusted platforms like TripAdvisor for User Generated Content(UGC) or Publisher sites for Expert data
E. Refining the recommendations
What makes a good experience great, in any scenario that involves an agent finding and brokering a product or service to a customer, is the effective two-way conversation. A conversation that doesn’t just stop one-way with the customer providing their preferences to the agent. But a flow that involves the agent learning from each input, asking for clarifying inputs following each interaction, and eventually delivering exactly what the customer wants. The scenario could be a real-estate agent helping a buyer find the dream property within a few showings, or a department store staff assisting a lady to find the perfect dress for a special event! An efficient conversation is key.
Akin to this, in the AI world the two important features the apps need to have are:
- Proactively prompting the users to clarify and specify their likes and dislikes(before jumping the gun and surfacing recommendations prematurely!)
- Proficiently adjusting the suggested itinerary: with every input and pushback received from the user
This is exactly how human agents and travel advisors work to deliver a bespoke itinerary. And if the AI tools aspire to compete with them they have to have realistic conversations to refine various recommendations. Like modifying the places to stay based on unsatisfied responses or tweaking things to see and do based on preferences and clarifications.
While web search results essentially reset with every new query, the language models powering chatbots can carry on long conversations, remembering and responding to questions and feedback throughout. Unlike real travel experts, artificial intelligence models can’t get their feelings hurt. That means the users have the freedom to shoot down ideas if they don’t like them — in fact, it should be encouraged.
Where are we right now?
There is no denying that AI-driven travel platforms are starting to revolutionize the way we plan our adventures. And there is tremendous potential and endless opportunities for innovation on this front. Which is getting validated by the growing list of players jumping into this arena every few months. A snapshot of some of the players below
But are they full-fledged enough to create personalized itineraries, simplify the booking process, and offer tailored recommendations as convincingly as human experts? Certainly not. Or at least not yet!
Curating a comprehensive travel itinerary is more of an art than a science. A well-constructed journey plan is more than just a list of attractions, experiences, and accommodations. A delightful programme will need to string your day together in a way that makes sense geographically, logically, and thematically. The current AI tools dont seem to be factoring in the practical aspects and realistic logistics of the plans being suggested. For example, in this plan for a trip to the Balkans, the last segment recommended by the tool was not just impractical but almost impossible to do!
“On the last day of your trip, enjoy a leisurely drive from Makarska to Pula. Stop for a swim at one of the many beaches along the way. In the afternoon, arrive in Pula and explore the city’s Roman amphitheater, one of the best-preserved amphitheaters in the world. In the evening, enjoy a farewell dinner in Pula before returning your rental car and catching your flight back home from Sarajevo International Airport”
Something a flesh-and-blood agent will never suggest! While the current crop of AI tools can certainly help you get started with a base or supplement an itinerary, they have a long way to go before they can fully replace human-planned trips.AI tends to create an itinerary based on what it has learned from other travelers and the published content on the web. However travel goals differ from person to person, and AI doesn’t have the same kind of knowledge that an expert living in a destination does. On top of this, the tools suffer from a few other issues that need to be rectified if they want to be fully trusted. Some of them are:
A. Hallucinations: Also called confabulations or delusions, refer to instances where these AI models produce outputs that are incorrect, misleading, or entirely fabricated, despite appearing superficially convincing. Like providing irrelevant(and sometimes even non-existent!) accommodation options was one common occurrence in my testing was Gemini constantly popping up non-existent accommodations and attractions!
Another problem related to this is that most AI tools stretch too far to answer questions they don’t know the answer to, instead of honestly saying they dont know the answer! What is scarier is companies are starting to sell guidebooks generated by ‘still-learning’ and outdated AI models. The New York Times recently reported that shoddy and inaccurate guidebooks are starting to flood the market!
But even among these “I know it all, will answer them all” universe of AI tools, there are creatively cute players like Layla, that communicate very transparently about the reality rather than misguiding the users. Below is a lovely example of a scenario where the app sweetly suggested that I stop looking for certain types of accommodation in a place where I can never find them, and instead enjoy what was available there or expand my radius. A delightful experience compared to the other apps that were constantly whipping up imaginary hotels, or popping up accommodations that were irrelevant to my prompt criteria!
B. One-way street: With generative AI as with so many things in life, you get out what you put in. Asking smart questions can make all the difference to the quality of guidance you will get for your trip. Most of the scenarios I played around with clearly demonstrated the importance of asking the right questions. However, the AI assistants I tried seemed overeager to generate output without fully understanding the user’s intent or seeking additional context. This behavior leads to the AI making assumptions or generating material that may not be fully aligned with the user’s actual needs or goals. Instead of directly jumping into proposing an itinerary, it would be more effective to proactively reverse prompt the users to share some more details around preferences or clarify( just like how human travel agents would do).
Going back to the Balkans trip example I mentioned earlier, Gemini just charted out a plan without even asking me more details about what kind of towns I wanted to visit, what time the return flight from Sarajevo was, or even checking how I wanted to be transported(private car or public transportation, etc). So it ended up with a very incoherent list of towns that were not packaged correctly into an itinerary or even logistically possible!
C. Dynamic Blindspots: There are quite a few potential blindspots AI tools can have when it comes to travel planning. Regulatory and legal complexities are one example. Navigating visas, travel advisories, insurance requirements, and other regulatory/legal issues can be a challenge for AI systems, especially across different destinations. They also might lack deep local knowledge or contextual understanding of local cultures, customs, hidden gems, and insider tips that a human travel expert would have. This can lead to AI-generated travel plans that miss unique experiences or fail to account for local sensibilities. Unpredictable events are another example. AI systems generally have difficulty accounting for the serendipity and unpredictability that can arise during travel — things like weather disruptions, transportation delays, last-minute closures, etc. A human travel planner would be better equipped to adjust and adapt on the fly.
In summary, AI can be a powerful tool to assist with research and finding individual components like flights and hotels. But they struggle to integrate those elements into a cohesive, optimized travel itinerary. The human element is still invaluable, especially for complex, customized, or high-stakes trips. The most effective travel planning experience will likely involve collaboration between AI capabilities and human expertise.
The Road ahead
It is nice to observe that the current set of tools has already evolved ahead from taking a simple generative AI approach towards a more Interactive AI/Conversational AI mode. Most of the interfaces are smart enough for us to engage in a contextually smart two-way conversation via questions and prompts instead of a one-way ‘fire and receive responses’ flow. Nonetheless, In their current state of evolution, these powerful tools are far from replacing humans. Instead, they offer a valuable complement to human expertise, boosting productivity by significantly reducing research time.
Overall, most AI planning tools seem to deliver a workable starting point for itineraries, when addressing travel needs focused on a single destination. Nonetheless, they are low in reliability when it comes to a multi-destination travel plan. Like the ones focusing on multiple stops around a country(or multiple countries in a region). For example, when users ask for something like “Plan me a three-day trip to Delhi that is predominantly focused on history and sampling local food” most of them fetch a decent sample itinerary that includes attractions and restaurants(with even recommendations of dishes to try!), divided up by times of day.
But the minute you shift into a multi-destination itinerary and request something like the prompt below, you will most likely run into a lot of friction:
This is a segment of travel planning that is filled with a lot of gaps that the AI apps still struggling to bridge. Voids that can only be filled by human agents and local experts(or excessive traditional search & browse exercises) if you want to curate a customized travel plan. I am sure they will eventually learn the trick and catch up. But for now, you are better off with the traditional approach with these kind of itineraries.
Apart from improving on this front, the AI tools also need to get more robust in closing the deal. Currently, most tools end the journey after stimulating demand and pitching travel components to the customers. After this, the flow is mostly passed to an online booking platform or a human agent to close the sale. To be a full-fledged travel assistant, the AI tools need to go beyond supporting just the Dreaming and Planning stages. They should also be able to effectively address the needs during the Booking and Experiencing stages. Like plugging in directly with booking platforms for reserving flights, accommodations, activities, etc. Or integrating with customer service platforms to smoothly communicate with service providers. Unless these standalone AI travel apps nail the booking stage, the bigger players like Expedia, Booking.com, and TripAdvisor will swallow their customers with their AI bots embedded directly into their platforms!
Arguably, the well-established OTA’s potentially stand a better chance at winning over customers leaning more towards AI. Here are a few reasons why I believe so:
- Data Advantage: OTAs have massive datasets on user behavior, past bookings, travel preferences, pricing trends, and accommodation availability. This allows them to develop more sophisticated AI models that can personalize recommendations. Smaller startups might not have access to such vast amounts of data, limiting the accuracy and personalization of their AI tools. If they start aggregating internally captured data with data from third-party sources they could create a fuller picture of the customers that smaller startups might struggle to build. For example: nothing stops a player like Kiwi.com or Expedia from asking the users to register all their loyalty memberships(Hotels, Airlines, etc) within their platforms. This way, the AI tools they build can apply hyper-targeted prioritization when recommending hotels or flights!
2. Integrated Ecosystem: OTAs have a comprehensive infrastructure integrating various aspects of travel booking — flights, hotels, car rentals, and activities. Their AI can leverage this interconnectedness to create seamless booking experiences. Smaller startups may only have a narrow slice of the travel ecosystem limiting the scope of their AI’s capabilities
3. Predictive Intelligence: With access to unlimited historical data, the models they build can predict flight and hotel price fluctuations, aiding travelers in making informed decisions by identifying optimal booking and travel times
That said, smaller travel startups can still differentiate themselves by focusing on niche or underserved audiences, or by developing AI-powered features tailored to specific pain points. But in terms of overall customer journey fulfillment, the scale and resources of major OTAs give them a distinct advantage when it comes to leveraging AI to its fullest potential.
In closing, we have just entered an exciting era where the overall travel experience will get a lot of delightful innovations driven by the power of AI. Not just during the pre-trip part of the traveler’s journey, but even during the trip and post-trip. I can easily imagine a different universe where the airport and in-flight experience would have improved significantly via personalization and removal of friction points. Airlines tapping into Al to set ticket prices dynamically based on predicted demand. A macrocosm where Hotels offer a highly digitized, tailormade, seamless customer experience while employing Al to determine optimal room rates that maximize occupancy and revenue. Al could also boost efficiency for workload d planning, predicting staffing needs during busy check-in periods. If done right, the frontline workforce of the travel industry would look and feel like superheroes powered by AI! The teething issues the AI tools are facing will eventually be ironed out and we will be living in a hybrid reality of human-AI collaboration. Teamwork that would combine the Human strengths of Creativity & Personalization with the AI strengths of Efficiency & Scalability