In the era of the digital revolution, data is the new gold, and when it comes to natural language tools like your Asksuite virtual assistant, being able to easily access data about its performance and extract valuable insights can be a significant competitive advantage.
If you have ever considered creating an action plan to improve your robot's understanding and thereby increase the number of quotes and bookings, this brief article is exactly for you! In this text, we share illustrative and straightforward tips on 9 important actions that can (and should) be taken to enhance your robot's understanding performance over time.
Topics
Monitoring the understanding rate of the last 30 days
Comparing the understanding rate with the average of Asksuite customers
Tracking the understanding rate over time
Approving new teachings recommended by the robot's algorithm
Monitoring the occurrence count of already published teachings
Monitoring "unrecognized phrases" by the robot
Avoiding "synonyms" and "phrase variations" in two different teachings
Editing recommended phrases, if necessary
Separating specific intentions into specific teachings
#1 - Monitoring the understanding rate of the last 30 days
The understanding rate of your robot, when analyzed over the past 4 weeks, provides a snapshot of your robot's performance. This is because 30 days is the minimum period required to gather a reasonable volume of conversations with travelers and gain a clearer understanding of their main interests.
Unless there are exceptions, a period shorter than 30 days usually shows less reliable data when viewed by days or weeks, as fluctuations are natural.
In the image above, on the Robot Intelligence tab of your dashboard, there is a leftmost section with a gauge (similar to a speedometer) that provides exactly that information about your hotel or network unit.
Currently, there are 3 possible statuses for the understanding rate of Asksuite's robots:
- Poor (0% to 49%)
- Average (50% to 74%)
- Excellent (75% to 92%)
- Unfeasible (above 93%) *You can find more information on this topic in another post here.
While the speedometer shows the understanding rate of your hotel, there is a ruler at the bottom of the section that provides a comparison with the average of all Asksuite clients.
Another relevant observation is that the speedometer section shows the average of the selected period in the filter, allowing you to choose the last quarter, the entire period, or a custom period.
Recommended actions related to this section
- Set achievable improvement goals.
The truth is that improvements in your robot's understanding won't happen overnight but rather over the course of months. Therefore, aim for monthly improvements based on the current status of your robot:
-
- Poor (improvement of ~10% per month)
- Moderate (improvement of ~5% per month)
- Excellent (improvement of ~2% per month)
#2 - Comparing the understanding rate with the average of Asksuite customers
Alongside the indicator mentioned in the previous topic, there is a scale that shows the comparison with the average understanding rate of over 1,000 Asksuite clients who use the same AI-powered booking assistant as you.
This data is useful as an interesting market benchmark to provide you with more confidence in how competitive your robot's performance is.
Recommended actions related to this section
- If your robot's understanding is below the average, set the initial goal to reach the same level as the entire user base. This way, you will strive to keep your robot minimally competitive with the market.
- If your robot's understanding is equal to or slightly above the average, set a goal to improve it within the Excellent status. This way, you will make your robot's understanding a distinguishing factor.
#3 - Tracking the understanding rate over time
Right next to the Artificial Intelligence indicators screen, you'll find the evolution of your robot's understanding over the months. This information is useful to get an idea of the future trend in understanding performance and to have a clear view of its historical data. It's also worth keeping an eye on the fluctuations, as shown in the image below:
In this illustrative example, there was a slight drop in May because travelers started asking new questions that the robot hadn't been trained on yet.
Through monitoring, the hotel manager was able to quickly notice the downward trend and publish new teachings that will help the robot respond to those questions correctly.
Recommended action related to this section
- Is this change in the context of questions seasonal? Is there any new topic that travelers are asking about?
#4 - Approving new teachings recommended by the robot's algorithm
In a nutshell, the teaching recommendation algorithm recognizes patterns in received messages and suggests new teachings for user approval. In this context, the recommended teachings include a "keyword," "synonyms," and "phrase variations" to aid in deep learning.
This functionality greatly helps hotel managers save time as they no longer need to search for misunderstood phrases within each chat with travelers.
Furthermore, the recommendations provide a stronger foundation as the suggested synonyms are derived from the dictionary based on their frequency of occurrence, and the phrase variations originate from actual conversations with travelers. This results in increased effectiveness for future interactions with your robot.
You can access the new recommended teachings in two different ways:
- Dashboard > 'Robot Intelligence' tab > 'New Recommended Teachings' section.
- Teach Robot Page
________________________________________________________________________________________________________________
🔍 Note: As shown in the image above on the Teach Robot page, the recommended teachings come with the tag Robot Suggestion. Additionally, they have the status Pending because it is necessary for you to provide a response in order to publish the teaching.
To help you quickly notice when there are new pending suggestions, two alerts have been created, as shown in the image below. They are:
- Counter on the left sidebar menu, just above the Teach Robot page;
- Alert banner at the top of your dashboard.
________________________________________________________________________________________________________________
Recommended actions related to this section
- Monitor the alerts for new recommended teachings on a weekly or biweekly basis.
- If there are new teachings, review the suggested parameters such as keywords, synonyms, and phrase variations. Approve those that are accurate and remove those that do not make much sense in the context.
- Register responses that directly address the recorded phrase variations.
- Afterward, train your robot and test to validate if the robot responds correctly.
#5 - Monitoring the occurrence count of already published teachings
Just like it's important to monitor if there are new teaching recommendations made by the robot, it's also worth checking the number of times travelers have activated already published learnings. These data can serve as valuable insights to better understand the travelers' profile and their main doubts over periods of time. With this information, it's possible to create actions outside the Asksuite platform to generate more value for the guests, such as targeted marketing campaigns, making conversions more accurate and optimized.
As shown in the image below, you can check the number of occurrences of the teachings already in operation within the Dashboard, in the Robot Intelligence tab.
Here are some illustrative examples that demonstrate the value of monitoring the number of occurrences of the most requested teachings over periods of time:
- Teaching: "Discount for residents of Foz do Iguaçu"
A resort in Foz noticed that there were a significant number of inquiries in the chat about benefits for local residents. With this insight, the management added a poster in the hotel lobby highlighting this information to facilitate offline visibility for people as well.
- Teaching: "Quad bike tour"
A farm hotel realized that there were many interested guests in the quad bike tour, which provided a strong argument for investing in vehicle maintenance as well as safety equipment.
- Teaching: "Kiss concert"
A guesthouse realized that many travelers wanted to know more information about the upcoming rock concert happening in four weeks at a nearby event venue. As a result, they included "Rock & Roll" interests in their digital marketing campaigns to target the relevant audience.
Recommended actions related to this section
- Monitor the number of occurrences at least every two weeks.
- Filter by specific periods to observe which teachings show an increasing trend in the number of occurrences, and ask yourself:
- What offline actions can I take to generate more value for travelers based on the insights from these teachings?
- How can I optimize my marketing campaigns based on this data?
- Review the most activated teachings and consider adding new synonyms for keywords, as well as new variations of phrases to enhance your robot's understanding.
#6 - Monitoring "unrecognized phrases" by the robot
For hotel managers, there is tremendous value in knowing the list of your robot's misunderstood phrases during chats with travelers.
For hotel managers, there is tremendous value in knowing the list of phrases that the robot does not understand during conversations with travelers, which you can find in the Phrases not understood by the robot section.
Before we provided this visualization in our dashboard, many users reported that they used to manually monitor chat conversations or social media to extract this information, which consumed a lot of time and energy.
If you want to better understand the context of the phrases listed in the dashboard, simply click on 'View Conversation' to be automatically directed to the specific dialogues where the traveler used the misunderstood term.
In addition, there is an orange button with a plus symbol (+) located on the right, where you can activate two action options based on the respective phrase:
- Create a new teaching OR
- Link it to an existing teaching.
In both options, your robot gains improvements in its understanding to respond more accurately in future occurrences.
Here are some illustrative examples that demonstrate how you can use the misunderstood phrases to enhance the Asksuite virtual assistant:
- Phrase: "What is the voltage?"
A resort in Fortaleza received many questions about this and realized they needed to create a new teaching in their robot to respond about the 220V voltage.
- Phrase: "Do you accept student IDs?"
A water park in the Northeast linked the phrase to an existing teaching called "student discount," thereby expanding the chatbot's repertoire.
Recommended actions related to this section
- Monitor the number of occurrences at least every two weeks.
- If a misunderstood phrase is relevant, take action by either creating a new teaching from scratch or linking it to an existing one.
- If a misunderstood phrase doesn't make sense (as sometimes happens), simply delete it to keep your list clean.
#7 - Avoiding "synonyms" and "phrase variations" in 2 different teachings
One of the most common errors when creating new teachings for the chatbot is registering identical synonyms or phrase variations for different teachings.
When this conflict occurs, the robot doesn't know which teaching to prioritize, resulting in a response of non-understanding for the traveler.
To help you visualize when a teaching has synonym conflicts, an alert has been created to highlight this registration error, as shown in the following image.
In this example, the synonym Decoration is already registered in the teaching Romantic Night, while there was an attempt to also register it in the second teaching Events Saloon.
Note that the "synonym" cell has been highlighted in red, along with an alert icon and a message stating:
This synonym has already been registered for the teaching Romantic Night. Choose only one teaching to be able to respond correctly to travelers.
Recommended actions related to this section
- Always choose only one of the teachings to save the synonym;
- If the synonym is truly applicable to the context of two different teachings, check which one has the highest search volume on the topic. From there, save the synonym within the one that makes the most practical sense.
#8 - Editing recommended phrases, if necessary
Our automatic recommendation algorithm for new teachings also allows your robot to include suggestions for phrase variations.
In practice, these phrases are essentially the text questions that travelers ask the robot and serve as an important parameter for your virtual assistant to better understand the context of the question.
If there is any suggestion for a phrase, you will find it in the right column of the list of phrase variations within the Teach Robot screen, as shown in the image below:
In some situations, it may happen that the phrase suggested by the robot is too long. If this happens, we recommend editing the phrase to include only the traveler's central question, as shown in the image below:
In the example, the initial phrase is "Hello, good evening. I'm from São Paulo and would love to visit your hotel. Can I have a romantic dinner there, by any chance?".
Note that only the end of the sentence (the underlined part) is what really matters for the teaching. Registering more concise phrases will facilitate the robot's understanding to identify the intention of a future traveler with the same question and keywords.
Editing the phrase is very simple. Just click on the text and the editing option will be automatically enabled.
Recommended action related to this section
- If a recommendation by the robot for a teaching has a very long suggested phrase, edit the text to focus only on the central question related to the teaching.
#9 - Separating specific intentions into specific teachings
Once you have reviewed the new teaching with relevant synonyms and variations of phrases that make sense in context, it's time to register a response.
This step is particularly important because the response recorded in the platform will be used by your virtual assistant to address the main concerns of travelers regarding a specific topic.
In some cases, a teaching may contain variations of phrases on very different topics related to the same keyword, which makes it challenging to register a specific response for it. For example:
- Teaching keyword: "Baby";
- Variations of phrases:
- Do you have baby strollers?
- Do you provide baby food?
- I need a bathtub for my baby.
In this example, even though there is a common subject, which is "baby," the questions have different intents.
In such cases, try to create separate teachings for each intent, allowing for specific responses to specific questions.
Recommended action related to this section
- If a teaching has questions with different intents, try to separate each intent into a different learning.
Conclusion
We have reached the end of this article, and we hope that this list of tips helps you implement an effective action plan to continuously improve the performance of your virtual assistant.
Remember to monitor your robot's performance regularly because it can (and will) generate more bookings for your hotel or inn if you keep it as up-to-date as possible.
If your team is lean and you need to handle the tasks yourself, allocate 30 to 60 minutes every two weeks, as your time is valuable and these tasks are indeed important.
If your team has someone responsible for marketing or performance, delegate these tasks to that person. They may have more time to execute the improvement action plan.
The Asksuite tool works in favor of your hotel, so make the most of our dashboards! Save this text in case you have any new questions and share this content when a new person joins your team. 😉
Comments
0 comments
Please sign in to leave a comment.