Create a Custom Model with a Report

4 mins read - Updated on Oct 28, 2026

You can create a custom model from a Report you have already analyzed, so that the labels Kimola gave its feedback become a model you can use again. Kimola turns the Report's classified content into a training set, with each piece of feedback and the label it got on a row, and trains a classifier on it. You start it from MODELLING in the Report's Settings, and a four-step wizard takes you through it: Training Set, Mapping, Labels and Review. The model then classifies new feedback with the same labels in your Reports and Funnels.

This is the shortest way to a custom model when a Report's themes already work well. You might read a first dataset with the Automatic Classifier, check the themes it found, and turn the Report into a model so that every later analysis uses the same labels. If you have no Report to start from, you can create a custom model with a prompt or with a training dataset.

Getting Ready

Custom models are available in the paid packages, not in Starter. You need a Report whose analysis is complete. Sign in to Kimola, open Reports from the main menu and open the Report.

What the Training Set of a Report Holds

The training set Kimola builds from a Report has two columns: Content, the text of a piece of feedback, and Tag, a label it got. A piece of feedback appears once for every label it got, so a multi-label Report gives a training set with more rows than the Report has feedback. The rows are kept in a sheet named Records.

The Tag column holds the labels of every classifier the Report used. If the Report was classified with a content classifier and the Sentiment Classifier, the themes and the sentiment labels Positive and Negative end up in the same column, and a model trained on it learns them as labels of one classifier.

Example

A software company analyzes 500 support tickets with the Automatic Classifier and the Sentiment Classifier, and each ticket gets a sentiment label and two themes, such as Billing and Account Access. Its training set has 1,500 rows: every ticket three times, once with each of its labels.

A classifier works best when its labels describe one kind of thing: topics or sentiment, not both. If the training set mixes them and you want a topic model, download it as an Excel file with Download Training Set, delete the rows labeled Positive and Negative, and create the model with the edited file by the training dataset method. The same file is also the place to rename, merge or remove labels before training.

Tip

Download Training Set, next to Create a Custom AI Model, lets you check the labels of the training set before you create the model.

Step 1: Open Modelling in the Report's Settings

Open the Report and click Settings in its menu. The MODELLING section says "You can create a custom AI model based on the classified content in this report." Click Create a Custom AI Model.

[Screenshot: The Settings page of a Report with the MODELLING section and the Create a Custom AI Model link]

The wizard opens with its first step, Training Set, already done: Kimola has built the training set from the Report, so there is no file to upload. If your package doesn't include custom models, a window says so and offers to upgrade your package or contact sales. If you have reached the custom model limit of your package, it offers to upgrade or to delete a model you don't actively use. See kimola.com/pricing for the packages.

Step 2: Check the Text and Label Columns

The Sheets & Columns screen shows a preview of the training set, with the number of rows in total and the file's name and size under it. The two columns are already chosen:

Text Column is Content, the feedback itself.

Label Column is Tag, the label of each row.

Leave them as they are and click Next.

[Screenshot: The Sheets & Columns step with Content as the Text Column and Tag as the Label Column above the preview]

Step 3: Check the Label Distribution

The Labels screen lists every label in the training set with its Count (the number of rows that carry it), its Percentage of all rows and its Status. A sentence above the list sums up the training set, for example "Your 16-label training set has some balancing issues but suitable for model building."

The Status column marks the labels that need attention. Above the overrepresented threshold means the label has far more rows than the others, so the model leans toward it. Below the underrepresented threshold means it has too few rows for the model to learn it well. A training set with flagged labels can still be used, and Next takes you on. A training set needs at least 2 and at most 25 labels: outside that range the screen says "Your training set is invalid." and you can't continue.

[Screenshot: The Labels step with each label's count, percentage and status]

The wizard doesn't let you change the rows of a Report's training set. To balance the labels, download the training set, add or remove rows, and create the model with the edited file by the training dataset method.

Example

In the software company's training set, Positive and Negative together have a row for every ticket, so Negative, the more common of the two in support tickets, might be marked above the overrepresented threshold, while a theme mentioned in a handful of tickets, such as Data Export, might be marked below the underrepresented threshold.

Step 4: Name and Review the Model

The Review screen sums up the model before it is trained. Fill in the fields and check the rest:

Model Name is the name the model gets in Kimola (e.g., Support Ticket Topics). It must be different from the names of your other models.

Category sets the kind of classifier: Content Classification sorts feedback into topics, and Sentiment Classification detects how customers feel. Choose the one that matches the labels of the training set.

Dataset shows how many rows the model is trained on, and from which sheet and file.

Required Query shows the queries training will use, next to your package and its usage.

Click Complete! to start training.

[Screenshot: The Review step with Model Name, Category, Dataset and Required Query]

Training uses queries by the rule queries = rows × models × Dimensions. A training set is one model with no Dimensions, so every row of the training set is one query. The rows are the label rows, not the pieces of feedback: the software company's 500 tickets make 1,500 rows, so training uses 1,500 queries. See How to Calculate Queries.

Note

Training uses one query per row of the training set, and a piece of feedback has a row for every label it got.

What Happens Next

After you click Complete!, Kimola uploads the training set, then the window says Getting Ready while it optimizes the model for the best accuracy. This may take a little while, and you don't need to keep the window open: on the Models list, the model shows Training... until it is ready.

When training is done, the window says All Set! and Go to Model! opens the model's page. On the Models list, the model shows its number of records and its consistency, the percentage of semantically consistent pairs in its records. See What is Model Consistency?.

[Screenshot: The All Set! screen with the Go to Model! button]

Before you use the model in an analysis, test it on your own text or on a file, and explore its records to add, delete or download them. If the results aren't what you expected, see Optimize Dataset-Based Custom Models.

Once the model works as you want, choose it when you create a Report using a custom model, so that a new dataset is classified with the same labels as the Report it came from. In the Enterprise package you can also add it to a Funnel, and every Space using the Funnel classifies its files with it.

Was this article helpful?

Tell us about your thoughts and experiences regarding the article.