Though it seems that AI is a newly discovered field, many principles and models were already discovered and developed in the mid 20th century. Already during the Dartmouth Conference in 1956 the foundation for Machine Learning and AI was laid. The most popular participants were C. Shannon, founder of the information theory and J. McCarthy, winner of the Turing and Kyoto award.
Still, no general intelligence was developed up-to-now. Instead, intelligent assistants are available to execute a specific task.
Already in 1996, Deep Blue, a chess computer developed by IBM defeated the at-that-time chess champion – G. K. Kasparov. Many people had already thought that the time had come that computers were superior to humans. Nevertheless, it turned out that computers are predestined for that type of calculations, which did not involve any kind of Artificial Intelligence as we know it today. Also, it came to mind that being human and interacting with its surroundings in a human fashion is a much more complex task.
In the meantime, many advancements in the field of data science led to a new golden area in machine learning and artificial intelligence. By applying principles and models like the LSTM and CNNs, human performance is achieved and even exceeded for some tasks like image recognition.
By adopting Reinforcement Learning principles, artificial agents were trained to play a specific type of game which cannot be solved solely by using brute force – like Deep Blue did. It has been shown, that while these artificial agents only relied on the principles learned during training, the best human players were outperformed. Still, especially in interactive tasks, human performance is still untouched.
One major problem of ML and AI algorithms is that now vast amounts of energy need to be provided for executing all processing steps and calculations.
Relating to the example of IBM Watson, which competed to previous champions of Jeopardy, a computer network of 90 8-core processors offering 4 threads per core combined with 16 TB of RAM were necessary to outperform humans. In comparison, though the human brain consumes approx. 20% of the energy provided per day, it is not even a measurable fraction.
Therefore, now many major tech companies are currently developing new architectures and computing paradigms especially designed for AI applications.
IBM TrueNorth or Intel Liohi are so-called neuromorphic processors, which try to imitate the behavior of the brain on hardware level.
In comparison to usual general-purpose processors, where most of the time major parts are active, neuromorphic processors provide huge amounts of processing cores with associated memory, that are operated only on demand. This way, the TDP can be reduced by a factor 10-1000.
We have structured all the advancements of AI in tourism around the customer journey. The customer journey consists of five consecutive steps, whereas according to studies the majority, except “on property” are carried out online today:
By a rising technologization of the travel industry many parties are involved in the customers journey, whereas it depends on the behavior of the guest itself, which parties are involved. Relating to AI, already information about the planning behavior of potential guests is of highest interest.
Considering that the tech giants have the resources for the development and execution of heavy-weight machine learning models and are normally involved in every journey step, they profited and still profit the most. This leads to a shift in dominance of the travel industry, whereas the hospitality industry suffers the most from these developments.
By aggregating user data using analytics services, which rely on machine learning meaningful insights are provided.
Being a key source of personalized information for the travelling industry, Google began to enter the market in 2011 by offering suggestions for travel destinations, activities and by providing an interactive metasearch engine for hotels.
The interface of a respective hotel shows reviews from many different sources, as well as the images and the location of the hotel. For booking itself, the user is redirected at a booking website – therefore Google Hotel is regarded as a metasearch website, rather than an online travel agency. Potential activities, shown on the bottom right are just listed according to the position of the hotel, while alternative hotels are displayed according to a listing created using machine learning methods.
Another important source for inspiration are social networks. Social networks are key information providers of personalized and aggregated information. While none of them entered directly the tourism market, they are regarded as one of the most important advertisement platforms. Still, it is to consider that users perform advertising by sharing their travel experiences with friends, but also in groups – only dedicated for this purpose.
Still we always needs to consider, where the so important booking phase is happening – as only then money is spent by the end consumer.
The next phase is dominated heavily by the large OTA (online travel agencies) which have become much more important in the last years and which are still developing at a very high growth rate. As well as social media and search engines, online travel agencies are using machine learning and AI to improve their services continuously.
Many of Bookings AI and ML ambitions focus on optimizing the customer experience, by accelerating processes and optimizing the HMI. Though not obvious, Booking is one of the hidden champions in AI, with more than 200 Data Scientists among 16.000 employees.
Machine learning applications are scaled across the company by the following approach:
Thereby, they combine offline and online sources. Its all about collecting data and training the algorithm and models.
Monitoring is performed live, whereas for some users’ new models are tested, for others not. Such that statistics of the gain of the new model can be abstracted directly from a live environment (online experiments), which in the online space is often called as A/B testing.
In 2017 Booking became one of the first companies to experiment with Neural Machine Translation (NMT). Meanwhile Booking is available in 43 languages, while most of the translations were performed by freelancers using the Google Translation Toolkit – which is a tatistical Machine Translation (SMT). SMT works by breaking up sentences into parts and then translating these parts one-by-one. For example English-Turkish translation does not work at all by using SMT, while the newly developed NMT glanced at this task
Another example is the creation of connected tours by using ML: Besides booking accommodation, Booking.com is leveraging technology to build connected trips that include tourist attractions, food and shopping. This means that users can book an „Experience“ using the Booking.com app.
Using AI and ML, based on the users’ individual travel intent, a personalized experience is created. Only insights abstracted from huge databases that include information of what travelers liked and did not like using ML tools, make this service possible. Combining this data source with the customer‘s previous travel preferences, as well as combination of third-party data like current waiting time at the most popular museum, assure that the suggestions are relevant.
Kayak processes more than 2 billion search queries a year. As the site further grows, Kayak has incorporated more sophisticated ML algorithms to help people refine their flight, hotel and rental car searches. Kayak saw AI as a mechanism to make its products smarter and more efficient, but it took years of learning to get to that point. One of the most important realizations is that AI should not be considered as a wonder weapon.
VP of technology Keller takes image processing as an example: in image classification only 95% of ML generated labels are correct, meaning that still evaluation of the results is necessary. For non-image problems it is even hard to reach 80% accuracy.
Keller says that „AI is about algorithms learning from already existing data. They‘re not going to generate any new solutions. They basically give the best answer of something you have seen before in a training set. As much as we try running our existing ML model with the belief that everything is still going to be fine, the better route is continuously measure and retrain when needed.“
In a more and more fragmented global market and world in general, Global Distribution Systems are more important than ever.
Realizing that the increasing demand for these services can only be covered by an increasing application of technology and especially AI, GDS companies invested vast amounts in these sectors and created partnerships with technology companies. Already today, ML is widely applied among the market leaders:
Automated information aggregation (Analytics and Intelligence Tools) is mainly based on ML now – f.e. Travelport captures more than 6 billion travel-related messages a day. Having these information, f.e. airlines can better plan their resources and flight plans
Amadeus says that their intelligence and analytics solutions uncover new opportunities for growth, highlight areas for performance improvements and anticipate trends and customer behavior, helping your business adapt and thrive in a fast-changing marketplace
A concrete example is Sabres SynXis Analytics Cloud platform, which was built for the hospitality industry. It helps hoteliers avoid lost revenue by analyzing data from operations, finances, room-stay production, ancillaries and rate-room-channel configurations. By using AI, a hotelier can design and test different predictive models by choosing a variety of pre-built machine learning algorithms
In order to keep up with those changes and reduce the dependability of technology giants, hotels and campsites but especially traditional travel agencies need to reinvent themselves and extend their web-presence.
The Nielsen Millennial Traveler Study from 2017 (performed in collaboration with SDA Bocconi) revealed that the millennial generation travels more than any other generation before. In contrast to past generations, where travel planning was characterized by in-person inter-action with a travel agent, most of the travel planning and booking takes place online now.
Although it is currently not common practice that journeys are booked using voice assistants, it might be widely spread soon. If that happens, search engines, social networks and metasearch-engines might lose part of their influence on the tourism market, as all research and booking steps necessary to book a journey are circumvented.
But, this requires too, that the properties are found directly in the search engines on the first position. Else the booking cannot be done on these sites, but will still increase the dependencies on small and mid-sized properties on large players.
The 2010 volcanic eruption in Iceland forced KLM to postpone many flights – this led to stranded passengers bombarding KLMs Facebook page. From that time on, KLM decided to increase their social media presence and recognized it as another channel for customer service.
Later, KLM realized, that they had another problem: the number of employees was simply too low to answer all customer concerns. Instead of scaling the social media team further, KLM turned to AI in order to automate customer queries in a few ways:
KLMs social media presence was extended to various platforms including Facebook Messenger, WeChat, WhatsApp and Twitter.
Sine the adoption of AI, KLM has seen many improvements:
In order to keep up with all those dramatic changes and the dominance of tech companies, hotels need to innovate themselves. Generally, personalization is one of the most powerful tools, to assert the position and increase the margin.
One possibility is the automation of repetitive processes, especially in customer service, in order to increase the efficiency of the company. An example for an automated service is the automated generation of offers. As accommodation requests do not automatically lead to a stay at the hospitality, it is crucial to adapt offers to the customers needs, to respond to requests quickly and to give the possibility to book directly on the personal webpage. By doing so, many parties involved in the booking process can be bypassed. Another positive effect is that many information related to the customer and user behavior is kept inside the company.
In order to keep up with all those dramatic changes and the dominance of Tech companies, hotels need to innovate themselves. Generally, personalization is one of the most powerful tools, to assert the position and increase the margin. One possibility is the automation of repetitive processes, especially in customer service, in order to increase the efficiency of the company. An example for an automated service is the automated generation of offers.
As accommodation requests do not automatically lead to a stay at the hospitality, it is crucial to adapt offers to the customer’s needs, to respond to requests quickly and to give the possibility to book directly on the personal webpage. By doing so, many parties involved in the booking process can be bypassed.
Another positive effect is that many information related to the customer and user behavior is kept inside the company.
Another aspect of personalization is the calculation of prices for every potential customer, also regarded as Revenue Management. Though personalized prices for every individual is still a future concept, there are currently already applications in use, that make prices dependent on several factors: Current travel season, Current occupancy, Prices and occupancy of competitors, if available Locality of the hotel , Travel trends.
As there is an unknown number of factors that still need to be considered, real personalization is an extremely complex task. A simple example for personalized price calculations are flight portals, where depending on whether users do research using an iOS based mobile or an Android based device, separate prices are calculated.
Still, there is a lot of untapped potential, for example the personal circumstances of potential customers are not considered now. By making use of analytics tools and including user data from search engines, from social media and other sources, the price calculation can be refined further.
Of course machine learning and AI methods are predestined for automation of price finding, as well as automated aggregation of vast amounts of data.
Although, it is excellent to use machine learning and AI for price setting, the same should happen for the offerings to the different clients. Here, yet, not a lot is done and therefore the properties do not really generate a lot of site revenues from food and beverage and services.
This untapped potential is incredibly high. First of all, properties are required to have digital systems in place, which allow their clients to easily make orders. And then these systems need to be integrated with AI in order to optimize the up- and cross-selling potential. So that the non-beer drinking client will not get offered a beer next to his pizza, but instead a glass of white wine.
Two examples of customer service automation and personalization, as well as enhancement of travel experiences are Connie from Hilton and the Henn na Hotel in Tokyo:
Connie is a robot concierge which was developed as a pilot by Hilton in cooperation with IBM. Set to operation in 2016 at Hilton McLean in Virginia, Connie makes use of a combination of IBM Watson APIs, including Dialog, Speech-to-Text, Text-to-Speech and Natural Language Classification, to enable it to greet guests upon arrival and to answer domain related questions about hotel amenities, services and hours of operation. By tapping into WayBlazer’s extensive travel domain knowledge powered by Watson, Connie can also suggest local attractions outside the hotel. The system is set up in such a manner, that iterative learning is provided, such that the more guests interact with Connie, the more it learns, adapts and improves its recommendations
An example of extreme automation is the Henn na Hotel in Tokyo, which was opened in 2015. The goal was to reduce the number of human-powered services and activities as much as possible. Some services, of course, still needed to be provided by humans. In 2019, the hotel administration stated that the reduction of half its 243-robitic workforce is inevitable, after customer complains did not abate. One major problem was, taking the front-desk as an example, that the robots were not even able to answer simple questions. This underlines the problem of closed-domain Text-to-Speech and Speech-to-Text. Another even bigger problem was the service intensity of the robots itself – leading to frustrated guests and frustrated service technicians.
So never forget – we are still at the beginning of using the full potential of AI. I personally believe, that we cannot allow, that we change the processes of people if we want the are accepting and using the technologies.
Many Marriott hotels now put Alexa voice assistants in their hotel rooms. If implemented properly, these assistants don’t only provide general information and services, but also information and services related to the hotel.
Room service, as well as billing of it can be automated to a certain extend – increasing the overall efficiency of companies. By making strong use of IoT technologies, following smart homes principles, completely new customer experiences can be provided:
In the traditional sense, the hospitality industry only provides accommodation. When relying on this credo, paradoxically though being the core of the travel industry, the market relevance of the hospitality industry will suffer further.
Considering that also private accommodation provided by Booking.com or Airbnb is regarded especially by millennials as a good alternative nowadays, the hospitality industry needs to underline its relevance.
Taking as a negative example the case that a hotel is not listed by search-engines, nor by metasearch-engines or OTAs, the only possibility to attract customers is by traditional travel agencies and by relying on regular. Therefore, a strong web-presence, as well as target-group-oriented-marketing is inevitable nowadays.
The automation or at least assistance of Hotel Property Management Tasks is another, yet still undeveloped area.
Taking again the example of an automated offering process, it requires a logic that automatically verifies the availability of free rooms, which correspond to an identified travel mission by using an interface to the local HPMS. By implementing such a mechanism, simultaneously, different strategies can be carried out to optimize the occupancy of hotels and dedicated room types. Considering that the automatically generated offer was accepted, and the fee was paid online, this information can be shared autonomously over the same interface with the HPMS making human interventions only an exception.
Another yet very relevant example, when using voice assistants to enable a personalized room service, the respective costs can be automatically assigned to the respective room and guest.
Chatbots, which are mostly based on AI, are already widely spread among the hospitality industry, flight industry and tourism industry in general for consumer services. Though the user’s adoption of this kind of technology depends strongly on the locality, the generation and the culture of a potential customer, it offers another communication channel.
Therefore, independent on the adoption rate, such technologies should be implemented – also as reassurance for a future higher adoption of these kind of services.
As already implemented by KLM, such applications should not only be limited to personal websites, but should also be implemented as service on our daily communication platforms like:
Especially for travel inspiration of regulars and the post stay, chatbots might be a good medium to maintain contact.
For tourism organizations, keeping up with travel trends always was a crucial task in order to succeed. Still, the difference in travel expectations among generations and the difference among the communication channels was never as big as today.
Therefore, the application of state-of-the-art methods for information gathering and information aggregation is inevitable. A common practice is the use of analytical tools provided by search engines and social media, but also the execution of extended marketing studies.
Unfortunately, only spare information about the current practices and methodologies are provided by organizations.
In order to be successful on a highly technologized travel market, especially the hospitality industry must keep up with these quick changes. Though many instruments exist and have been described during this article, the adoption is little spread.
As the research and booking process were nearly totally shifted to the virtual world, following the winner-takes-all-principle, there are still several processes untapped by big tech.
This includes the especially the travel inspiration and the on property stay.
To succeed in a vibrant travel world, hotel-customer binding needs to be further deepened and current best-practices of internal processes need to be re-evaluated in order to maximize the efficiency of companies.
By defining and following the concept of a highly technologically-integrated hospitality strategy can be carried out, such that all competitors can be circumvented by extending the number of regular guests that book directly on the personal webpage or contact consumer-service-staff directly – increasing the margin drastically and reducing the dependence on tech giants.
Gather as many data as you can. By aggregating data from many different sources, target-group-oriented-marketing needs to be carried out, in order to address a potential guest directly.
Of course, simple flat rate online advertising is not beneficial – by doing so the opposing effect is achieved, such that competitors on the travelling market are strengthened. Following the practice of smart advertising, also f.e. by making use of newsletters and special offers, on one hand the web-presence can be extended and the whole booking process can be circumvented
On property, the guests should be offered a personalized experience by making use of technological mediums like voice assistants that are smartly integrated in the hotel concept or through apps and web apps, which allow the customer to buy.
By offering guests to possibility to enter virtual worlds, also new personalized experiences can be created – related e.g. to the alpine hospitality industry, people that are unable to reach the top of a mountain could still be given the possibility to do so, by making use of VR or AR.
Following the principles of Automation-as-a-Service, AaaS, by defining and applying highly automated processes, staff can be shifted from administrative and repetitive consumer-oriented tasks to face-to-face consumer-oriented tasks – which leads eventually to more satisfied guests and more satisfied employees.
Also, for tourism organizations, personalization is key to succeed – which includes knowing its customer by making use of analytics tools, carrying out studies and combining customer information from different sources. By having this information, it is much easier to build up its right web-presence, which is essential nowadays. Also being engaged in target group-oriented advertising is easier realizable then. So, gathering data, is more important than ever.
Open Data Hub is an example of a publicly accessible data base, which provides information related to travelling and mobility. This includes hotels, hotel rooms and inventory, weather, parking, charging stations, activities, gastronomy, museums. Connecting all this data may lead to a huge benefit and allows tourism organizations to much better understand the market they are offering. We encourage all tourism organizations worldwide to follow a similar approach and to begin to gather data of their guests.
By making use of interactive communication tools like chatbots, but also passive communication tools that make use, for example of text analytics or question-answer-engines, personalization can be deepened further: Based on the information provided and shared by the potential guest, customized trips can be carried out. This may include: Identification of the travel mission, such that dedicated activities can be recommended. In accordance with the travel mission, an appropriate hotel or hotel room can be recommended and in best case, whole travel packages are recommended. Ideally, all recommendations are integrated in an internet booking engine, such that it can be booked by entering only the payment information and clicking on the confirmation button.
This article is based on the presentation “Artificial intelligence and the travel revolution”, which Hannes Lösch, founder and managing partner of Limendo, has given September 26th 2019 in Kochi, Kerala, India. If you would like to receive a copy of the presentation slides, please send us an email at firstname.lastname@example.org