SatisFactory takes a new step! Discover the new semantic analysis model, which relies on Gemini artificial intelligence technology.
With its new internal semantic engine, the result of several months of development, SatisFactory opens a new era in the analysis of the voice of the customer. Thanks to Google's generative artificial intelligence, your analyses gain in depth and reveal nuances, context, and weak signals that traditional approaches left in the dark.
Table of contents
The security of the new semantic engine
The calculation of the semantic tone
Activate the new semantic analysis engine
Configure the classification plan
Make the classification plan evolve
Presentation
Semantic analysis (also called "verbatim analysis") is an advanced natural language processing (NLP) method that can be applied in a feedback management context to comments collected from various sources. Its value lies in the ability to measure, group, and prioritize the subjects mentioned by the respondents, whether they come from satisfaction surveys, online review sites, social networks, audio recording, or other external data sources.
At the center of this approach is the semantic classification plan, which structures the analysis and is defined upstream. It establishes the themes, sub-themes, and concepts expected in the verbatims, offering a coherent framework to transform a large volume of raw comments into clear and directly actionable insights. This finesse of analysis allows in particular to categorize the same comment into several themes, depending on the richness of the subjects it addresses.
The semantic analysis proposed by SatisFactory therefore allows you to:
- Categorize each comment into one or more parent concepts and child concepts, based on the identified themes
- Prioritize the identified customer expectations, according to their importance and their impact
- Detect the polarity (positive, negative, neutral) associated with each concept addressed in the comments
- Highlight persistent irritants, weak signals, and emerging trends all at once
- Provide strategic and operational teams with a clear, structured, and directly actionable vision to guide their decisions
You will find below a comparison between the old and the new semantic analysis model of SatisFactory:
| Analysis model | Old sector-specific Semantic Model by keywords |
New specific Semantic Model by AI prompt |
| Type of analysis |
|
|
| Operation |
|
|
| Activation |
|
|
| Configuration |
|
|
| Processing |
|
|
| Additional features |
|
Sentiment analysis, a key component of our semantic analysis, automatically attributes a polarity (or tone) to each subject addressed in a comment:
- Positive: The comment expresses a favorable opinion, a satisfaction, or a positive emotion on the subject
- Negative: The comment expresses a dissatisfaction, a criticism, or a pain point on the subject
- Neutral: The comment is factual, incomplete, or does not express any strong emotion
The security of the new semantic engine
In terms of security, the new AI semantic analysis model relies on Google Gemini.
When using the artificial intelligence features of the SatisFactory platform, we guarantee maximum confidentiality of your data. Our commitments are based on 3 pillars:
- Systematic anonymization of verbatims: Before any submission of a comment to the artificial intelligence, the prompts are automatically cleaned and anonymized
- Local data sovereignty: Data processing by generative artificial intelligence is exclusively carried out on European servers hosted in Paris
- Limited use of data by our partner: Google formally commits to never using your data for the training of its models and to not sharing them with any third party
To learn more about data security on the platform, we invite you to consult the SatisFactory Trust Page.
The calculation of the semantic tone
Presentation of the semantic tone
Semantic analysis determines the sentiment of the comments by categorizing them into 3 tones:
- Positive tone
- Negative tone
- Neutral tone
This classification brings a double level of sentiment analysis.
- First, the comment is referenced in its entirety as positive, negative, or neutral.
- Then, in addition to your semantic classification plan and the global referencing of the comment's tone, each sub-theme identified in the comment is assigned a specific tone (positive, negative, or neutral).
This complete process makes it possible to cover all analysis needs, by offering both a global evaluation of the comment and a precise understanding of the sentiment associated with each subject mentioned in the respondent's feedback.
Explanation of the calculation of the semantic tone
At the moment the semantic tone analysis is triggered for a comment, the following steps take place:
- Retrieval of the complete feedback of the respondent
- Analysis of each rating left by the respondent (key indicators and satisfaction items/attributes).
- Identification and categorization of themes and sub-themes in the comment (according to the classification plan defined on the platform)
- Assignment of a tone (positive, negative, or neutral) to each sub-theme identified in the comment
- Global classification of the comment's tone (positive, negative, or neutral) depending on the balance of the tones of each identified sub-theme
⚠️ Comments that are too short (less than 3 words) are excluded from the tone calculation and are classified by default as neutral.
💭 To determine the global tone of a comment:
- If the number of negative sub-themes is greater than or equal to the number of positive sub-themes, the global tone is classified as negative.
- Likewise, in the event of a tie between the sentiments in the identified sub-themes (as much negative or positive as neutral), the strong tone (positive or negative) takes precedence over the neutral tone.
Activate the new semantic analysis engine
The activation of the new specific semantic model by AI (or the migration from the old one by keywords) is a project managed by your Customer Success Manager (CSM). Consequently, the first step consists of sending them your need to initiate the co-construction phase.
The steps for implementing the new semantic analysis model are the following:
| Steps | Actions | Responsibilities |
| Step 1 | Constitution of a corpus of at least 300 representative comments for the training of the AI model | You |
| Step 2 |
Reflection on an initial structure for the semantic classification, including the detail of the considered categories and their content Provision of a standard classification plan (subject to availability for your sector in the SatisFactory database) |
You |
| Step 3 | Co-construction of a semantic classification plan including parent concepts, child concepts, and their associated definitions, formalized within a dedicated tab of the project repository | SatisFactory and You |
| Step 4 | Review and validation of the semantic classification plan | You |
| Step 5 | Configuration of semantic themes and sub-themes in the settings of the SatisFactory platform | You or SatisFactory |
| Step 6 | Activation of semantic analysis on the account by the SatisFactory technical team and deployment to production | SatisFactory |
| Step 7 | Optional history recovery for processing by the new semantic engine based on all comments accessible on the platform | SatisFactory |
A few hours after the activation of the semantic model on the platform, you will find the semantic analysis from the "Themes" tab. This advanced analysis, personalized to your own context, is also integrated transversally in many other features of SatisFactory, in order to enrich your analyses and facilitate decision-making.
Configure the classification plan
Once the semantic analysis model by generative AI is activated on your account by the SatisFactory teams, you can configure it very simply.
Follow the procedure described in our dedicated article: Configure the semantic classification plan.
For your multilingual needs, you also have the possibility to translate the name of the parent concepts and child concepts configured on your account (learn more).
Make the classification plan evolve
Once the semantic analysis model by generative AI is activated on your SatisFactory account, its scalable nature allows you to refine it over time.
You have total autonomy to modify your semantic classification plan. These changes will apply only to comments collected subsequently.
The application of the updated classification plan to your history is not automatic. A history recovery service is necessary; we invite you to contact your Customer Success Manager regarding this matter.
During periodic reviews of your system (monitoring or steering committees), an item on the agenda can be dedicated to the review of your semantic classification plan.
Several types of modifications are possible to refine your classification plan:
| Need | Possible actions |
| Categorize unclassified comments |
|
| Identify and track emerging weak signals |
|
| Correct the classification errors of a target of comments |
|
| Add exclusion rules to comments |
|
| Refine the classification of comments |
|
FAQ
- What is the difference between the old semantic model by keywords and the new model powered by artificial intelligence?
The old model relied on a rigid list of keywords defined and added manually in the platform. The new model uses Google's generative artificial intelligence, Gemini, to understand the context and concepts, even without exact keywords; an AI prompt defining the concept is sufficient. It is faster to configure, more flexible, and allows you to detect weak signals or nuances that the keyword approach ignored.
- Is it possible to simultaneously activate semantic analysis by keywords and analysis powered by artificial intelligence on the same account?
No, it is not possible to combine the two analysis methods. In order to ensure global data consistency on the platform and to feed the AI features (such as the flash report, the comments summary, and the insights), only one type of semantic engine can be active per account.
The addition of an AI prompt on an already existing concept automatically activates this engine: it will take over from the keyword processing for future received comments.
- Are my data used to train Google's artificial intelligence?
No, Google contractually commits to never using your data for the training of its AI models, nor to sharing them with third parties. The privileged partnership between SatisFactory (Konecta Group) and Google allows us to guarantee you a maximum level of security and confidentiality.
- Can the personal data and sensitive information contained in the comments be anonymized?
Yes, in order to guarantee security, all personal data is automatically purged from the comments before they are processed by artificial intelligence.
In parallel, we optionally offer an anonymization API to filter personal data as well as insulting remarks right from the collection phase in your satisfaction surveys. This process guarantees total confidentiality: sensitive data is at no time visible or stored in the data flow, because it is automatically replaced by generic terms ("(CREDIT_CARD)", "(PHONE_NUMBER)", etc.).
Consult our article on this subject and get in touch with your Customer Success Manager (CSM) if this option interests you.
- Do I have to define specific themes for each source (reviews, surveys, social networks) or group everything in a global semantic classification plan?
It is imperative to build a unique and global semantic plan. Indeed, during the configuration of the platform, you can define to which comments the processing of the semantic analysis applies, allowing you to have a custom configuration.
Without specific configuration, the engine applies uniformly to all future comments that arrive on your account. Your classification plan must therefore be transversal and cover all the subjects that can be addressed, whether they come from your surveys, online review platforms, social networks, audio recording or other external sources integrated into the platform.
Strategically, this approach guarantees consistency in the analysis of the Voice of the Customer, regardless of the collection channel, and allows the platform's AI features to have a holistic vision of the data.
This does not in any way affect the precision of your analyses. Thanks to filters, you can freely segment the semantic analysis on the platform to target only a specific channel, survey or comment.
- What are the steps and requirements to build my classification plan and activate the new semantic engine?
The first step is to contact your Customer Success Manager (CSM), because the approach can vary depending on whether it is an initial activation or a migration from the old keyword engine.
Concretely, you will have to provide a corpus of at least 300 comments if you do not already have a sufficient volume of comments on the platform, and you must also reflect on the tree structure of your themes.
If you already have a classification plan, we can use it as a working basis for the workshops, just as we can offer you generic templates (depending on the business sector).
The crucial point for the success of the project lies in the drafting of exhaustive definitions for each concept, ideally enriched with examples of customer verbatims, in order to ensure the precision of the analysis by artificial intelligence with your specific context.
- Can the same comment be classified into several parent concepts and child concepts?
Yes, the AI is capable of analyzing the richness of a verbatim. If a customer addresses several subjects (for example, delivery and product quality) in the same message, the engine will classify the comment into all the concerned themes (parent concepts) and sub-themes (child concepts), with the appropriate tone.
- What happens to my old comments classified by the keyword engine if I activate the new artificial intelligence model?
The new AI model applies to all new comments collected following the activation.
To apply it retroactively to the entire history of comments kept on the platform, it is necessary to make a specific request to your Customer Success Manager. Thanks to a finer detection of nuances, your historical data is re-evaluated. The same comment could thus reveal new concepts hitherto undetected, considerably enriching your past analyses.
- With the new AI engine, is it necessary to associate a satisfaction item with a semantic concept?
This is possible but not mandatory. However, if you wish to be able to cross-reference data between these two elements, you always have the possibility to associate them on the semantic analysis configuration page.
- Can we provide a global context to the AI engine to facilitate the design of the semantic plan?
For the moment, the AI semantic engine does not manage a global context. Indeed, it relies exclusively on the descriptions of the themes and sub-themes provided by you.
As the platform is destined to evolve, this contextualization of the specificities, challenges, and jargon of your organization could be developed in the future.
We invite you to provide us with a glossary or a document presenting the different products and services that you can offer and that are likely to appear in the comments.
- Can we import the semantic classification plan by file import?
This feature is not yet available but is scheduled for the second half of the year 2026. You will thus be able to complete a CSV file template and import it into the administration center of the platform to automatically create your semantic classification plan.
- I have the semantic analysis system by keywords, can I still benefit from the qualitative analysis of competitors' reviews?
No, this is unfortunately not possible. The semantic analysis system applied to social networks and competing external data sources (learn more) works exclusively with the semantic analysis by AI prompt.
If you wish to keep your keyword analysis system, this competitive intelligence feature will not be available on your account. To be able to analyze your competitors' reviews, it will be essential to switch to the semantic analysis model by AI prompt.
For further assistance or to report a specific issue, please contact our Support team.
Comments
0 comments
Article is closed for comments.