Planning a Perfect Trip with an AI Assistant — Part 1

12 min readApr 8, 2024

This is part 1 of a two-part series in which I unpack the impact of Artificial Intelligence(AI) on the online travel planner’s complicated customer journey. Click here for Part 2 of this series

The recent developments in Artificial Intelligence(AI) have already started shaking up the Travel industry, with several disruptions getting introduced in the various stages of a digitally savvy traveler’s customer journey. This blog focuses specifically on what’s getting redefined in the Dreaming and Planning stages of this complicated customer experience map, through the assistance of AI assets available to users.

Currently, the burden of using travel technologies to research and plan for a trip effectively lies on the user’s side — thanks to most of their interfaces being super dated. For example: The tiring UI of inputting dates, # of Pax, and # rooms to search and book accommodation is at least a quarter-century old! So, it is refreshing to witness the dawn of a new era where the burden of task execution will gradually get transferred from the user to the technology platforms. With the advent of AI, it is almost becoming a norm that the tools must perform the extra hard work of understanding and fulfilling user needs. While the users can seamlessly converse in their natural language and easily accomplish any research and planning task!

Nonetheless, it is still going to require some level of skill from the user front. Shifting from crafting the right query to crafting the perfect prompt. And for the systems, it’s going to be transforming from understanding a query intent to grasping the trip intent. The static planning loop of “search, select, browse, research, and compile” would inevitably evolve into a conversational exercise where travelers can “converse” with the system, sharing information like a destination, dates, interests, dynamics, etc via natural language and get back a very personalized itinerary packed laden with vivid descriptions.

The question is, are the super-hyped AI tools strong enough already to plan a seamless trip on their own? Can they be your ultimate travel sidekick like they claim to be? Here are my observations from the testing I performed using some of the available tools to validate if they can reduce the endless hours of trip research, and help in crafting personalized travel plans easily and effortlessly!

Breaking down Trip Planning

Before we get started, let us break down the complex task of trip planning. The Planning phase of the Customer Journey Funnel(snapshot below) is easily the most cumbersome process in the flow.

As we all know, designing a dream holiday can be both exciting and challenging. It’s easy to get bogged down in research and preparation, potentially spending as much time planning as you will enjoying your destination. From choosing the right destination to finding the best accommodations and activities, the process could easily spiral into a time-consuming and stressful activity. You could get immersed in a maze of guidebooks, travel blogs, reviews, and web searches — especially when composing a bespoke itinerary that suits your preferences and constraints.

From the travel planner’s perspective, this phase can be further broken down into various ‘micro-stages’ including:

Discovering Where to travel( when the destination(s) has not yet been decided)
Calculating How long is an ideal timeframe to spend there
Identifying When is the best time to travel there
Figuring out Which is the best way to get there
Learning What are the things see, do, and experience there
Choosing the Where to stay
Packing all these into a Perfect itinerary with the most optimal route and appropriate pitstops!

From the app or platform’s perspective, the system should be intelligent enough around the following capabilities to credibly serve the user across the various stages called out above:

Understanding the Traveler
Grasping the Trip Intents
Gaining a holistic view of recommended items
Suggesting a personalized plan
Refining and adjusting the plan based on user feedback

Let’s further dissect these two pillars, and explore the various things that need to be done correctly on both sides of this task execution flow (user side and system side).

What the user needs to do

The biggest ability AI tools would potentially empower a typical traveler with is the ability to be self-reliant and independent when it comes to planning and composing an itinerary. Not having to depend on the support or expertise of a local travel planner or operator to tailor-make an itinerary that caters to all their desires

But there is a bit of a challenge the users are faced with. Compared to traditional travel tech, in the AI universe, It’s not just about choosing a preferred tool but also mastering how to use it to achieve the best results. In the Generative and Conversational AI world, a well-engineered Prompt is what matters the most if you need to extract the most out of any AI platform you choose to use. So, let us walk through the various components of a memorable getaway a typical planner might be researching.

Before getting into the various components of a memorable getaway users might be researching, let us take a step back and look at the typical user personas on this front. Travelers come in many different flavors, each with their preferred approaches to trip planning and at various stages of the journey toward confirming their dream vacation. The below snapshot captures the most common(of course, there are a lot more types) personas.

The next section unpacks the Open-minded Traveler persona, who is still higher up in the customer journey funnel, and analyzes the various micro-stages they might go through.

A. Discovering WHERE to travel

Unless you are a destination-committed planner, the first step when planning a trip is often deciding on the perfect destination. When undecided on the location, users might want recommendations based on specific preferences, such as the type of destination (beach, city, hill station, etc), activities they enjoy(diving, surfing, trekking, etc), interests they have(nightlife, gourmet, shopping, historic exploration, etc), or traveler composition(family trip with kids, girls trip, solo travel, bachelor party, etc). Here is a sample blog that captures this micro-stage more in detail. Below is a sample prompt that I tried out to discover a potential destination for my next holiday:

Test Verdict: Among all the apps Layla felt most human, interactive, and conversational. However, it was way off on the recommendations! MindTrip and Vacay brought back decent results but were not compelling enough. Microsoft’s Copilot seemed a bit clueless recommending irrelevant locations. Most of the tools grapple with the same issue of limited choices. Through manual research, I was able to discover a wide range of destinations that matched my travel criteria. However, most of the tools restricted their recommendations to a very narrow(and restricted) set. This round certainly goes to the Humans and the Machine is far behind(albeit intelligent enough!) I eventually picked a location from the output of Gemini because of how it had broken down in detail and why it selected the places it chose based on various signals I had provided. Sweet! I chose Oahu, Hawaii. Now the next step was to figure out when to visit.

B. Identifying WHEN to travel

From here on the customer journey map of both the Experience Focused Explorer and the Destination Committed Planner personas would be pretty similar and go through the next few micro-stages in more or less the same way. Though I had started researching for my upcoming holiday in June, I wanted to be sure if it was indeed a good time to visit! So I fired the next question to my new travel planning buddy Gemini and asked ‘her’ the following:

My expectation here as a traveler was for the AI assistant to capture all critical factors that might impact my travel logistics and judiciously advocate whether the time I have chosen is the most optimal. And preferably advise alternate times as required. Below are a few factors that might potentially impact the ‘delightfulness’ of the trip, that the system needs to consider:

Weather: The most important factor, without a doubt. You dont want to visit a beach destination during rainy weather, a tropical destination during the peak of summer, or the northern hemisphere during the harshest part of winter — unless that’s your jam! The system should be well informed about this for at least the popular travel spots.
Seasonal Factors: Most destinations have a peak season and an off-peak season marked by a consistent ebb and flow of visitors. The high season typically overlaps with the time window when visitors can best experience what a destination is well known for. For example: viewing the Cherry Blossoms in Japan or Fall Colours in the Pacific Northwest, witnessing the Northern Lights in Scandinavia, or participating in popular festivities like the Octoberfest in Bavaria or the Carnival in Rio de Janeiro. Usually, a shoulder season is most optimal as it would let you experience the destination well enough without having to spend too much or competing with an overwhelming tourist crowd.
Dynamic factors: This involves being aware of things like events and conferences taking place at the destination — especially entertainment and sports events. These will heavily impact the prices of accommodation and transportation costs.

A miscalculation on these factors will not just impact the overall cost of the trip negatively, but also reduce the charm and desirability of a destination as a result of tourist overload.

Test Verdict: Layla and Mindtrip came back with a good enough response to this prompt. But once again Gemini was the most convincing with a detailed breakdown of why and how it came up with the response. There are also other apps like Claude which are not travel-specific but do a decent job. But I would still give this round to Humans as the AI bots suffer from too many oversights. T.

C. Figuring out which is the best way to travel

Now that I had settled on June and the destination of Oahu in Hawaii, the next step was to figure out how to get there from my home base of Seattle. So I fired the following prompt into the tools, with enough instructions around the various flexibilities weaved into it:

Test Verdict: In this scenario, I was a bit surprised by what Gemini fetched in response to my well-defined prompt. Given the amount of information it might have accumulated via Google Flights data, I was expecting more comprehensive recommendations. But it ended up sharing nothing more than the airlines I could explore and with some super generic tips! Layla was certainly better with very relevant tips on days of the week as well as sharing alternate airports to consider. It even had a clear call to action button for booking the flights. Mindtrip was even more specific on the recommendations and even suggested a specific flight option. But in general, the apps are still a long way from recommending a perfect itinerary that a human expert could have done with the appropriate amount of research. This round also goes to the Humans!

D. Learning about what to do at the destination(s)

This is by far the most demanding micro-stage in the planning process. This is the stage where AI-driven tools diverge from traditional travel platforms. Instead of overwhelming the users by showing a generic buffet of all the things to see, do, and experience at a destination, they aim to offer a nicely personalized bento box that only contains the items most relevant to the users based on the implicit and explicit signals they have shared. A range of signals that include:

Interests: Primary interests of the Traveller(s)
Traveler Profile: Information about the traveler’s party (any kids, elderly travelers, or pets?)
Time: Amount of time they have on the ground
Budget: Optional factor

For instance, if you’re planning a trip to Paris and adore art, the platform should suggest visits to the Louvre, the Musée d’Orsay, and various art galleries. If you’re a food enthusiast, it may recommend renowned restaurants and local food markets.

Here is the prompt I tried with the tools to tease out a sample set of recommendations based on my interests, and things I shouldn’t miss experiencing on the ground. In this case, I packed an additional nudge in the prompt encouraging the bots to ask me for more clarifications if required.

Test Verdict: The itineraries shared by Vacay and MindTrip were quite disappointing with both of them suggesting something barebones and without doing enough justice to the destination! Microsoft’s CoPilot was not bad, but it was nothing more of a skeletal list of things to see and do than a well-packaged itinerary. Gemini was much better showing what looked more like a real itinerary (with even influencing pictures thrown in!). But the clear winner in this case was Layla — which created a detailed plan broken down into various days and multiple embedded links to book and book tours, tickets, and accommodations. On the flip side, I am confident a lot of us can come up with a more robust plan if we spend enough time to research — the emphasis being on ‘time’! Nonetheless, Layla and Gemini did provide a fantastic starting point that could be expanded into a full-blown itinerary. So, for this scenario, AI tools could function as great assistants and complement humans rather than replace them — by reducing the time spent in planning.

But interestingly, none of the tools took my cue to ask me questions to clarify the requirements if required. Something any human agent or expert would have immediately jumped on. But instead, they jumped directly into stitching up itineraries for me with the bare minimum specifications they had! Something the machine needs to learn and get better at, if they want to compete with humans on this front.

E. Choosing WHERE to stay

Now, that I have a destination to visit, with a high-level idea of things to see and do on the ground, the next step is to get recommendations on where to stay. To keep it simple, I didn’t ask for accommodation options tailored to my tentative itinerary, but more of a generic advice. Like a local expert would do if I provided the following criteria:

Trip intent — the primary intent of the trip. Like a Solo Business Trip, a Family vacation with kids, or a Romantic Trip with a loved one.
Accommodation Type: the preferred style of stay(like boutique hotel, budget hotel, hostels, heritage hotel, vacation rental, etc). The type of stay would indirectly also provide an implicit signal on the budget you are willing to spend

Here is the prompt I used to ask for accommodation recommendations:

Test Verdict: Layla spewed out just a list of hotels without any indication of why it picked those, or plotting them out across days or locations. Not very helpful! MindTrip was no different, but at least had links to the hotels it surfaced (through which I could at least get a good grasp of the hotels).

Gemini came up with a well-distributed itinerary that covered the variety of interests I shared along with price ranges. It also surfaced what I could do at the different locations, and even provided some tips. But unfortunately, it failed to understand my preference for ‘affordable boutique luxury’ as the prices were way too expensive!

So, in summary, I wasn’t comfortable trusting any of these tools(or others I tried) to make a call on which places to stay. It also seemed like these platforms had not ingested enough of the travel media content(like guidebooks, magazine content, etc) which planners like me would be inspired & influenced by to make a decision. There is room to cover here for sure.

F. Putting it all together

Finally, once a user has successfully managed to identify the best places to visit and the quintessential things to see, do, encounter, and experience (based on his/her taste), the next step is to plot them all out into the perfect journey. It is important to note that other traveler personas might not be going through these individual micro-stages but instead engage with the more detailed prompt(like the sample below). The apps need to efficiently address both types of scenarios with equal finesse.

A full-blown prompt for planning a Multi-Day Itinerary

In Part 2 of this article, we will shift focus to the other side of the equation, and analyze all the features a reliable and resourceful AI travel assistant needs to deliver.