Over the past year, the trend for hotel groups investing in AI to address communication and productivity challenges, has clearly accelerated. However, many hoteliers are still relatively new to technology, and often, the focus is mainly on functionality, to the detriment of what really matters: the quality and scalability of the data.
The main role of a conversational AI, such as Velma, is to understand customer requests and give the appropriate responses. This is the intelligence/knowledge binary which we saw in the article: “Sophism and artificial intelligence”.
In the vast majority of cases, the responses from virtual assistants are text formats in each language (or worse, automatically translated). This method, from another millennium, can ultimately work for one hotel, but becomes horrific on a large scale.
These text-based solutions face not only significant problems of quality but also huge issues of scalability, particularly when dealing with non-European languages such as Arabic, Chinese, Japanese, etc
Velma, the universal virtual assistant from Quicktext, is deployed in hotels in 76 countries, and has to date, generated a unique volume of more than 30 million conversations. Since 2017, this conversational data has made it possible to create a unique structured hotel database, the only one of its kind, globally : Q-Data.
This database currently includes 2600 information points (by comparison, a hotel website usually has a maximum of 250 information points). Moreover, this database is constantly evolving at the rate of around 300 new information points per year. This structured data model is the basis for creating AI quickly, efficiently and accurately for any hotel. In fact, the set- up time for each hotel is two weeks on average, and for groups (depending on size), from one to three months.
When you are processing information on 10, 50, 100, or even 1,000 hotels, maintaining consistent and accurate data in multiple languages, is a real challenge. We might as well say it right away : without structured data, it is pretty impossible to manage the data of a hotel group and totally impossible to maintain it.
This issue is resolved by using correct information data points in place of free text. Velma thus generates its responses from the Q-Data database, which can be easily collected and updated when required.
A good example is worth a thousand words: In a classic system, the check-in data could be the following.
Check in is at 3 p.m., you can benefit from an early check in at noon for 30 Euros.
This data must be written and translated into all languages… I invite you to test this simple sentence in automatic translation, you will be rather shocked. In Q-Data: all that is required is to simply fill the database with the 3 values 15:00, 12:00, 30. And automatically, thanks to the previous input from our in- house linguists, Velma will be able to give the answer in 36 languages.
And now let’s assume that this check in data is the same for 40 hotels in a group, we simply need to indicate where required that this data is generic and thus applies to all hotels in this group.
And here we have only talked about 3 points of information, Q-Data has 2600 !!!.
Imagine the same titanic task when updating information… We perfectly understand why it is impossible because quite simply no one will do it.
You will tell me, all this is archaic, you just need to scrape the hotel data (web, pdf, powerpoint etc) and the generative AI will do the work. Well actually no, that is not the case, for several reasons:
The result is that the system will likely ‘hallucinate’ (give nonsense answers) and you will have no way of knowing the origin of the problems.
You may try to say to me; the data of 30-room boutique hotel in London and a 2000-room resort in Punta Cana do not have much in common. Indeed, this is why Q-Data was designed for a hotel group of 4000 hotels with 4000 rooms. And those who can do more, can do less. And so the difference is that probably the resort will use more information points than the boutique hotel, but the database is essentially the same. Incredible but true.
In addition, when information points are added to Q-Data (around 300 per year, following the study of the conversations by our data analysts), they are automatically available to all other hotels in the world. In other words, Velma in your hotel will also benefit from feedback from other hotels all over the planet. This is why you will hardly ever have to create new dialogues, on new subjects, because it is certain that another hotel in the world is likely to have already needed that particular piece of data.
If you ask a classic AI in one hotel: “does this hotel have a swimming pool?”, it will probably answer you well. This is called a vertical search.
Now if you ask an AI from a hotel group: “I’m looking for a hotel with a swimming pool” – what is required here is a horizontal search on all the hotels in the group. And the only way for an AI to answer something like that, is for its horizontal search to be based on a structured database, such as Q-Data. This is why today Velma is the only AI in the hotel industry globally, which is capable of vertical and horizontal searches.
But Velma goes even further than that, because she is also able to search on several criteria at once. For example, ” I’m looking for a hotel with parking, in New York, with a budget of less than $300 per day”… And with Velma the best AI Hotel chatbot, all of this can be handled easily, even in such a natural conversational manner, because this search should also be manageable verbally.
You have now understood that the real problem with AI is less the cognitive capability of the AI, and much more the capacity of the AI to possess the data and that only by structuring this data can you guarantee quality responses in the short, medium and long term.
In case you still have any doubt, ask Velma, because #VelmaWorks because #VelmaKnows…
AMEN
© Image: Shutterstock
