How Kimola Uses AI for Research

Kimola analyzes customer feedback the way a research team would. Its AI finds themes in the data itself, gives each theme a comment mentions its own sentiment, and asks every comment a set of research questions called Dimensions. Statistical tests then show where feedback differs between groups or changes over time (Signals), and Action Plans turn these findings into actions after every figure and quote is checked against the feedback.

Dynamic Classification

Many text classification tools sort feedback into a fixed list of categories. A review of a water faucet might land under Quality, Pricing, Delivery or Customer Support. A researcher reading the same review would notice more specific concerns, such as installation difficulty, missing parts, a design flaw or a leak. Predefined categories also carry poorly from one industry to another, so they often miss what customers actually talk about.

Kimola's Automatic Classifier finds the themes in the feedback itself. Kimola processes the feedback before and after grouping it. This lets it decide the themes without any predefined labels and give each theme a clear name that describes what customers say.

When your team needs the same themes in every analysis, a custom Model gives your organization a shared language for its insights.

Aspect-Based Sentiment

Customer feedback is rarely about one thing. With multi-label classification, Kimola gives each comment every theme it mentions, and each theme gets its own sentiment. A review that praises the user interface but complains about slow response times gets both themes: the interface as positive and response time as negative.

This is called aspect-based sentiment. It shows what customers like and what holds them back within the same comment, which a single overall sentiment would hide. If you need one theme per comment, you can choose single-label classification when you create a Report.

Scoring: What Deserves Your Attention

Most analysis tools rank themes by volume. The most common topics always come first, and unexpected findings stay buried below them. Kimola scores and ranks findings with deterministic methods instead, so the same data always gives the same scores and the same order. Kimola uses this approach in several places, in both Reports and Spaces.

Signals apply it across the whole analysis. Kimola crosses every theme and every Dimension with the columns of your dataset, such as country, product or customer segment, and with time. This often creates thousands of combinations. Kimola tests each one to see whether a topic is mentioned more, or more negatively, in one group than in the rest, or more often in recent periods than before. Only statistically significant findings become signals, and they are scored and ranked so the ones that deserve your attention come first. An AI model writes each signal card afterwards, from numbers that have already been computed. The wording doesn't change what is found or how it ranks. In a Space, available in the Enterprise package, signals run again after every new file.

Action Plans build on these findings. Kimola forms hypotheses from the analysis and tests each one against the feedback with counts and sample comments. Supported hypotheses are written as actions, and every figure and quote in them is checked against the feedback. Actions the evidence doesn't support are dropped. The remaining actions are scored and ranked, each with a confidence level.

On the Performance screen, Satisfaction and Dissatisfaction Scores rank themes by how positive or negative customers are about them. A plain percentage can mislead here. Take two themes: one has 5 comments, all positive, and the other has 50 comments, 43 of them positive. By percentage, the first theme scores 100% and the second 86%, so the first one looks like your strongest theme. But 5 comments say little, and a few more could change the picture completely. Kimola's score takes the number of comments into account along with the share, so the theme with 50 comments ranks higher. This lets you compare themes of very different sizes fairly.