.png)
Analysis - ISAAC Analysis
Our AI system ISAAC analyses your open text feedback and divides this into categories and sentiments so you can quickly see what your customers are talking about.
NOTE:
Please keep in mind that only the text questions related metric score questions will be analyzed by ISAAC. Custom questions are not analyzed by ISAAC.
IN THIS ARTICLE
1. How to get there
Option 1
On the homepage, click directly on the analysis icon of the touchpoint for which you want to see the analysis.
Option 2
Step 1: Click on Analysis in the left navigation menu
Step 2: Choose the touchpoint for which you want to see the analysis and go to the NPS-ISAAC, CES-ISAAC or CSAT-ISAAC tab, depending on which metrics are available in your touchpoint.
2 General terminology
To understand the text analysis, also known as ISAAC, it is important to get to know some general terminology we use in the knowledge base and in the platform.
Main category: the main topic people are talking about.
examples: personnel, product, price etc.
Subcategory: more specific topic people are talking about.
examples: friendliness, assistance, quality, availability etc.
Category path: a specific combination of 1 main category and 1 or more subcategories.
examples:
- personnel > assistance > friendliness
- product > quality
Sentiment: this can range between very negative and very positive. It indicates the feeling people transfer in their comment. All options are: very negative, negative, unknown (which includes neutral statements and statements of which our AI doesn't know if it's positive or negative), positive and very positive.
Mention: the combination of a category path and a sentiment. One open text feedback can have more than 1 mention.
example: "The person who helped me was very friendly, but the products are of poor quality"
- Mention 1: Personnel > Friendliness - very positive
- Mention 2: Product > Quality - negative
You can do several things to ensure that the text analysis is tailored as much as possible to your personal situation:
- Start with optimizing your ISAAC analysis by choosing the correct sector and channel for your touchpoint and making suggestions in the platform.
- In case you want to take it a step further, you can group existing category paths to meet your specific needs.
3. ISAAC visualizations
3a. Top positive and negative categories
The charts display an overview of the top 10 category paths for both most positive and negative mentions. In this graph, we group together all mentions for very positive and positive and for very negative and negative. The default state of the graph will show you the top 10 most (very) positive mentioned category paths:
- The number of mentions in the graph (between brackets) is always based on the number of mentions for the whole category path.
- The percentage at the end of the bar is calculated as follows: amount of positive and very positive mentions for this category path divided by the total amount of mentions, minus the unknown mentions for this category path.
- On the left side of the graph, the gray bar indicates how often this category was mentioned with negative or very negative sentiment.
You can use the switcher on the top right corner to change the graph to top 10 negative categories. The same logic applies to this graph, but for the category paths with the most (very) negative mentions.
3b. Mentions per main category
The mentions per main category shows you the overview of main categories that are mentioned in the feedback. The main category is the first part of a category path. These group the feedback in more general categories. Take into account that only the 10 main categories with most mentions are shown.
3c. Sentiment spread
The sentiment spread chart breaks down the sentiment behind the feedback, providing an overall insight on whether the feedback on this touchpoint is mostly very positive, positive, negative, very negative or unknown.
In the example below the total of mentions is 491 of which 7,33% is very positive, 66,8% is positive, 22,2% is negative and 1,43% is very negative. Overall we can say that customers are positive about the interaction discussed in this touchpoint.
The remaining mentions have an unknown sentiment. This can have two causes:
- The customer mentioned something in a neutral way, so no sentiment can be linked to the comment
- Our AI was not able to detect a sentiment. This can happen for example when you use leading questions, for example "What can we improve?". If people answer "Price", our AI can not detect that price is here mentioned in a negative way.
3d. Average sentiment by category
In this graph, you can follow-up on the evolution of the average sentiment of your different category paths. The evolution is shown on a weekly or monthly basis - depending on what you choose in the dropdown. The color of the blocks ranges between red (very negative) and green (very positive). By looking at the colors, you can see which categories evolve during time. In the example below, you see that:
- Personnel: starts strong in the week of 13 Jan with a nice green color, and even gets more green during time, but in the week of 10 Feb goes down a little bit, which indicates people are becoming less happy about the personnel.
- Clothes: in the first week, people are rather negative about this topic, while it gets better in the middle, but drops again a bit by the end.
To calculate the average, we give the 5 different ISAAC sentiment categories a number:
- Very Negative: -2
- Negative: -1
- Neutral: 0
- Positive: 1
- Very Positive: 2
We calculate the Bayesian average for this graph to avoid evolutions based on too little data. Let's explain this by using an example:
- Imagine for a given touchpoint you receive 2 items of feedback which are very positive (2) for the Month of January. In February, you receive 2 items of feedback which are very negative (-2).
- If we would monitor this given on the Average Sentiment heatmap it will show a green roster for the month of January and a red roster for the month of February. This would give a distorted picture as the population of respondents is too small to talk about a negative evolution.
- To prevent this behavior, we use the Bayesian average. Calculating the Bayesian average uses the prior mean (m) , and a constant (c). We add 30 items with an unknown sentiment (Score = 0) as a constant (c).
By calculating the average sentiment like this, each positive mention moves the overall average up a bit and each bad mention moves it down, but a category with only a few scores will not rank extremely high or extremely low. You can show the exact average in the graph by clicking "values" (see above).
2e. Category explorer
The last graph you see on the ISAAC page is the Category explorer. This provides an overview of all the categories ISAAC has recognized in your feedback.
- In the first column, you will see the Main Category (1). This is ordered from the category with the most mentions to the one with the least mentions.
- The second column are any subcategories (2) that are linked to the main category.
- You can have an entire string of subcategories, each linked to the one in the column before (3). In this example, we see a total of 4 different levels.
It is possible that you will see the word Regarding next to your subcategories. This adds another level to the analysis, to provide you even more detail. If you click on this, the graph will expand to the right and you will see a linked combination of a main category and subcategories.
In the example below, we see that there is more information linked to the categories Personnel/Helpdesk/Information.
By clicking this, the graph expands and you see the category combination Channel/Phone appear. This means that the customer is talking about the information they received over the phone.
In front of each main or subcategory, you see a number. This refers to the number of mentions, or the amount of times your customers talk about this topic.
In this case, one customer might say something about the Friendliness of the staff and about their Expertise. This will account for one mention in Personnel - Helpdesk - Friendliness and one mention in Personnel - Helpdesk- Expertise, but a total of 2 mentions for Personnel - Helpdesk and Personnel as a main category.
By clicking on the bar with the number of mentions, the specific path will be highlighted, so you can clearly see all the categories that are linked.
When you hover over one of the blocks with your cursor, a pop-up will appear, showing you the Sentiment Spread for this category. In total we differentiate 5 sentiments: Very positive, positive, unknown, negative and very negative.
Finally, you can use the slider to filter your data on the number of mentions of the main category. This way, you can remove any main categories with only a few mentions to make the graph less cluttered. You can also filter out (multiple) sentiments.
3. The side panel
If you want to see more details, you can also click the name of one of the categories. This will open up a side panel where you can see a number of things.
INFO
The side panel can only be opened by users that have access to conversations.
The side panel is available for all graphs. It shows all open text feedback that is linked to the specific category path and will be pre-filtered on sentiment if you click on a graph that only shows a specific sentiment.
- When hovering over the question mark next to the category path, a description of the category appears
- For each open text feedback, you see which respondent said it (in case the touchpoint is not anonymous) and which other mentions are linked to this open text feedback. If there is a filter on sentiment active, you have a button "show more" to show the mentions that do not fit your filter.
- You can translate feedback that is not in the platform language by clicking the translate button. Once translated, you can go back to the original feedback.
- With the eye icon you can go to this specific conversation in the conversations section.
- The show metadata button will show you all specific metadata for that customer.
- The pencil next to the mentions can be used to make suggestions to improve the ISAAC analysis.
- In the side panel of the category explorer, you can see the sentiment spread for that category and all category paths that start with the selected category.
4. ISAAC analysis for specific customer groups
Filtering on participant attributes (metadata) can be very beneficial in determining how satisfied (CSAT), eager to recommend your company (NPS), or how simple a procedure was for a certain customer segment, as well as how they feel about your staff, products, etc. Comparing different groups to one another can also be interesting which is why we added smart filtering to the platform to simplify this process. To learn everything about smart filtering and comparing groups, read this article.