A small experiment with AI
Artificial intelligence (AI) is revolutionizing the way businesses operate. In the hospitality industry, AI can improve the guest experience in a variety of ways. One common way is to use AI to analyze guest reviews, extract information, and respond to them.
Guest reviews can be a valuable source of information for hotels. They can provide information about guest satisfaction with the services provided, as well as suggestions for improvement. However, analyzing guest reviews is often a tedious (and in many cases manual) task.
With AI, the process of analyzing guest reviews can be automated. Hotels can use AI to identify patterns in reviews, understand guest sentiment, and generate recommendations for improvement.
For example, a hotel could use AI to identify the most common topics mentioned in guest reviews. Cleaning of the rooms, customer service, or food quality. Once the hotel has identified these topics, it can use this information to take action and improve the service.
Another way AI can be used is to generate responses to guest reviews. Responses to guest reviews can help hotels build relationships with guests, show that they are listening to their feedback, and show that they are committed to improvement.
The best part is that AI can generate personalized, relevant, and helpful responses. For example, a hotel could use AI to identify positive guest reviews and generate a response of thanks. The hotel could also use AI to identify negative guest reviews and generate a response that addresses the issues mentioned by the guest, always with human supervision. In some of the tests I have done, AI has taken the liberty of proposing a refund of the reservation amount or giving free days for their next vacation (at least it tries to get the customer back…).
My Experiment
A few days ago I was at the Google Generative AI event in Mexico City. They were showing us the advances that exist today on the Google Cloud platform regarding generative AI. We were able to try, firsthand the AI tools along with Google engineers.
At that time I came up with a thousand ideas that could be applied to the hospitality industry, although I have been using Google’s IA systems for different processes in the company’s operations for a long time, the leap that is given with generative AI is spectacular and opens up new possibilities for developments that seem like science fiction.
I wanted to do a small proof of concept of a system that answers reviews from the different websites that we all know.
After setting up the platform and reading the reviews from the last month, the first thing I did was to program an analysis of feelings to assess the general sentiment of each review. Before anything, knowing what I am going to find, this will help me, for example, to detect what I can answer automatically or apply supervision depending on the different parameters that the analysis throws.
In the image, you can see the parameters thrown, the “score” tells us how good or bad the review is [-1.0,1.0] and the “magnitude” the charge of sentiment that there is in the text [0,….], a “score” negative with a “magnitude” high, is that we have someone very angry. With these values I created a new value called “happyness”, if it is positive it is good, if it is negative bad and the amount tells you how good or bad it can be.
Then I did an analysis by entities within each review, with this I can value the entities (words) within the reviews (the staff, the pool, the food, etc..) and know what importance they have in the good or bad rating of the review.
As we can see in the image, for this guest the most important thing when writing his review was the service and the bars, followed by the staff and the experience
I also included a summary, there are very long reviews that with a summary is enough to understand what it is about.
Finally, I generate three responses, each with a variation of the “creativity” and the number of words, curiously when it is more “creative” it needs more words to express itself.
Here we can see how it has responded, for obvious reasons I do not include the review but the summary can be seen, the first is the most serious (less creativity) and the last the most informal.
With the good reviews there is no problem, but what happens when the review is not good?
This is what surprised me the most, when the reviews are not good is when you really realize the power of AI. The answers are surprisingly good (note!!, they have to be reviewed, it proposes actions that are sometimes difficult to carry out), so much so that as a guest I would think about giving the hotel a second chance.
Response
Sentiment analysis allows us to identify the most common topics mentioned in the reviews. The summary of the reviews allows us to better understand the problems that guests were experiencing. And the answers generated by AI help to build relationships with guests, show them that we listen to them and that we are committed to improving.
What would be the next step?
Logically, such a system cannot be left to its own devices and answer by itself. Human supervision is needed, to choose and modify the answers to be published. But once this starts to be done, AI can be fed back with the answers chosen for each review and start to learn how to answer. Over time, AI will answer with the style and constraints with which we have trained it.
What else can be done?
It is clear that AI understands the context of a text and can give precise answers. But what else could we do?
Another strong point of generative AI is the ability to analyze data and make predictions.
Normally, in a hotel reservation, we collect data such as the date and time of purchase, geolocation by IP, hotel, room type, reservation amount, email, loyalty level, etc.
When guests arrive at the hotel, we complement this data with the data extracted from their passports. And during their stay, if we have the right tools, we can obtain their habits and preferences, types of restaurants and favorite foods, if they drink wine and what type of wine they like, if they go to the SPA, if they go to the gym, if they book excursions, if they buy cigars, if they order Room Service, etc.
In the end, we have a lot of information about our guests.
We also have information coming from marketing campaigns, Google ADS, Facebook, website navigation, behavioral cookies, from the company’s APP, etc.
Now, what would happen if we structured all this information (or not) and made it available to AI?
We can start asking:
I want to do a campaign (with such characteristics) for season x. Can you give me the profiles to which I can send it to be effective?
Can you generate the emails for each of them?
Can you include images (based on the ones it has already been trained on) to make these emails more attractive to their recipients?
Can you create a schedule for sending emails so that they are read at the right time?
In a conversation with one of Google’s engineers, he told me that CRMs are no longer necessary, a database in BigQuery connected to AI is enough. AI learns and understands what each field is and generates the appropriate connections between them to be able to answer the questions asked.
On the other hand, Google’s new tensor processing units (TPUs) allow the resolution of these queries to be practically in real time.
One of the great advances of generative AI is that it understands what we ask in natural language and it is not necessary to learn or handle complex systems for the extraction of results.
If we applied this same technique at the level of hotel operations, with all that information we have, we can generate personalized “journeys” for each guest. Have activities, restaurant reservations, gifts and “amenities” for the guest ready before they ask for it. And we can integrate all this with the hotel’s APP to offer the services and information that we know they will consume.
In the end, it all comes down to integration, data and analysis.
We are going to the world of “hyperpersonalization” of the service.
What if I wanted to try more things?, such as feeding the AI with historical information on occupancy, prices, ebitda, GOP, GOPPAR, etc., adding information from global economic indices, inflation, CPI, debt rates, etc., global data on tourism, regional occupancy, number of flights, number of visitors, etc., and finally, the forecasts of these same values in the future and see how the AI deduces the missing values and makes forecasts.
What would I ask him?
Can you calculate the rates for next month by room type and hotel?
What benefit would I get if I increased the tariff by 5%?
We just have to know what to ask.
Conclusions
- We are facing a new technology that is going to revolutionize (is revolutionizing) all business processes, customer service, and mass data analysis, creative processes, business operation processes, etc., and of which we have not yet seen or glimpsed how far it can go. It is a new world where the door to creativity is opened to us in all areas.
- It is not a technology of the future, we already have it available and the only thing that will happen in the future is that it will improve exponentially.
- It is simple to implement. It is very simple to start a project of AI. In the example I have shown, I think it took me less than 30 minutes to implement the AI processes.
- Training is needed, the AI must be trained to make it an expert in what we want it to do.
- It is a strategic decision to start using AI in business processes and decision-making. We have to bet on it with all the consequences. AI is disruptive and it will change the way we work.
- Companies that do not adopt the internal development of AI in their business processes will lose competitiveness.
P.D.
Part of the text of this article, the English translation, and the cover image, have been generated by AI.
