Skip to main content

šŸ§  AI Summaries & Sentiment Analysis

Updated over a month ago

Overview of available AI features for single review analysis

Important note:

This article applies only to the analysis of individual reviews.

It does not describe the behavior or logic of the AI dashboard, where responses are analyzed in aggregate and on a statistical level.

šŸš€ Quick Overview

flowit provides many different AI-powered analysis. The review data is for example used by the dashboards and connected-insights (CI) to provide you the best possible insights into what your employees are sayinh. There are two AI-powered features on an individual review Level:

  • AI summaries condense questions, sections, and responses from all employee, reviewers and additional reviewers responses into clear, actionable insights helping you with your preparation for the review meetings and goal setting for the next period.

  • Sentiment analysis classifies the emotional tone of individual responses.

Both features operate independently and serve different analytical purposes. The sentiment analysis uses a simple and fast sentiment AI model, while our summaries use a slower more computationally intensive and smarter large language model with many customizations and guardrails around it.

šŸ¤– What AI features are available?

1. AI Summary

AI summaries automatically synthesize review contents (i.e. sections, questions and responses to:

  • identify key themes faster

  • provide a clear overview of discussed themes

  • see the red thread through different types of questions and sections

  • reduce large volumes of text important key points

  • support clearer decision-making

Important:

AI summaries are always based on all submitted responses within a review, regardless of sentiment classification. Drafts are not included in the summaries only complete submissions are included.

2. Sentiment Analysis

Sentiment analysis evaluates the emotional tone of individual responses as:

  • positive

  • neutral

  • negative

  • unclear

It serves as an additional analysis feature to highlight emotional patterns.

ā€‹The sentiment analysis looks at the sentiment score of individual words and then classifies the entire sentence according to the score of single words. The scores are machine generated and do not always match the human intuition of what we perceive the sentiment of a response to be. Therefore, we do not use sentiments to generate summaries.

šŸ” Which responses are included in AI summaries?

  • All submitted responses in a review are fully included.

  • Drafts are not included.

  • Sentiments are not included. They are a separate analysis. This means that even if a response has an unclear sentiment score it is still included in the summary.

  • All responses are visible verbatim in the Responses tab.

ā“ Does sentiment analysis affect AI summaries?

No.

The two features are fully independent:

  • Sentiment analysis = emotional indicator

  • AI summary = content-based synthesis

Sentiment has no impact on summaries.

šŸ¤” Why is a response sometimes labeled ā€œunclearā€?

A response may be marked as ā€œunclearā€ if the AI cannot reliably detect emotional intent, for example when:

  • the response is very short or very long

  • emotional language is missing

  • context is limited

  • wording is neutral or ambiguous

Without clear linguistic cues, sentiment cannot be assigned confidently.

šŸ“Š Why is sentiment analysis more reliable in the dashboard?

In the AI dashboard, responses are analyzed in aggregate, resulting in:

  • higher statistical reliability and more answers give a clearer picture of the true sentiment across the entire survey.

Single responses are more prone to ambiguity, especially at longer text lengths. The sentiments are merely an indication of what the machine thinks the sentiment of this answer is. They are merely an aid for you to focus your attention. Your opinion is ultimately what counts. AI can make mistakes. The sentiment-AI is a different AI than the summary AI.

āš™ļø Can sentiment analysis be disabled?

Yes. But only for individual reviews. Not for dashboards.

Sentiment analysis can be disabled at any time with one click.

šŸ”Ž Note:

Changes to the current sentiment analysis logic are not planned.

šŸ§© How are AI summaries generated at flowit?

Is there a ā€œcommandā€ or prompt?

Yes.

flowit uses optimized prompts, tailored to each use case and continuously refined by the data team.

What context is used?

AI summaries are generated using:

  • the respective question

  • the participantā€™s responses

  • the participantā€™s role (e.g., employee or manager)

Currently not included:

  • custom user profile attributes (e.g., job title)
    ā€‹

How is a prompt structured?

Each prompt consists of several coordinated components:

  • System Primer

  • User Primer

  • Content String

  • User Query

This ensures the AI behaves in a:

  • consistent

  • robust

  • context-aware

way for every use case.

Did this answer your question?