How to Analyze Call Center Conversations

min read - Updated on Nov 15, 2026

You can analyze call center conversations in Kimola by keeping them in one Space and uploading new transcripts as a file every week or month. Kimola classifies every conversation with the same custom models, calculates the channel's NPS from what customers say and runs statistically tested Signals after each upload. You see which topics customers call about, how they feel when they do and what the latest file changed, and you can turn any finding into a Task for the teammate who owns it. This guide follows the whole path, from the first Funnel to the first Task.

Note

Analyzing call center conversations in a Space is available in the Enterprise package. It is also available in Kimola On-Premises.

The Goal

A call center holds more feedback than any team can listen to. A team that handles thousands of conversations a month can review a few dozen, so the topics that grow slowly, the products that quietly cause calls and the effect of last month's fix stay hard to see. The question this guide answers is the one a head of customer operations, a quality team or a CX analyst asks again every month: what are customers calling about now, how do they feel when they call, and what changed since the last upload?

A call center is a channel that never stops producing feedback, so it belongs in a Space rather than in a one-time Report. A Space grows file by file, and every new file is read against the records before it. Kimola can then tell you what the latest file brought, which a single analysis of a single export can't. If you have one batch of conversations to study once, such as the calls of a campaign week, a Report is the better fit. See Reports vs. Spaces.

The conversations can be in any of the 30+ languages Kimola analyzes. The language of the Funnel you choose only sets the language Kimola writes in, so a team can read German or Turkish calls in English summaries, or the other way round.

What You Need

You need the Enterprise package or Kimola On-Premises, and a few things prepared before the first step. None of them takes long to set up, but each is easier to get right at the start than to change later.

Transcripts as files. Kimola reads .xls, .xlsx, .csv and .tsv files, one file per upload, up to 30 MB each. Each row becomes one record, so export one row per conversation, with the conversation's text in one column and the column names in the first row. Export them from the system that holds your call records.

A date for every conversation. The first file needs one column with the date of each call. Every file of the Space is placed on the same timeline by this column, which is what lets Kimola compare the latest file with the earlier ones.

Columns to compare by. Any other column of the file, such as the queue, the product, the agent team, the channel or the region, is kept with every record. Kimola doesn't classify these columns, but you can filter by them, and Signals can compare the groups they form.

A customer key (optional). If your export carries a customer number, an ID or an email, every customer gets a page with their conversations across all your Spaces.

Custom models. A Funnel is built from custom models, so the models for your call center come first. Step 1 explains which ones.

Teammates. The people who will own the findings need to be members of your team, because a Task can only be assigned to a team member.

Step 1: Build a Funnel

A Funnel is the set of custom models every Space of your team classifies feedback with, so it decides what Kimola can tell you about your calls: which themes they fall under, whether customers sound positive or negative, and which research questions are answered for each one. Build it for how your call center is actually talked about. If your team already has a Funnel with a content classifier and a sentiment classifier that fit, you can use it, and the call center is then read in the same themes as your other Spaces.

The content classifier matters most. Its labels become the themes of the Space, so create it around the reasons customers call, in the words your teams already use for products, processes and problems. The sentiment classifier is what the Space's NPS is calculated from: without one, the Space has no NPS. Add Dimensions when you want more than themes. Pain Points asks what goes wrong or creates difficulty, Unmet Needs asks what is missing or desired, and Experience Stages asks when it occurs. A Funnel holds one model of each kind. See Create a Custom Model with a Prompt to create the models.

Example

For a telecom operator, the content classifier of a call center Funnel might have labels such as Roaming Charges, Coverage Problems, Plan Change and Number Porting, with a Pain Points model next to it. Every call is then read through the same themes, and each theme shows what goes wrong in it.

In the Funnel wizard, you choose the models, then the structure, then name the Funnel and choose its language. Leave Enable Multi-label Classification on for call center conversations: one call often covers several topics, such as a billing question followed by a complaint about a dropped call, and with multi-label classification none of them is lost. See Multi-Label vs. Single-Label Classification. Name the Funnel after what it classifies, and choose as its language the one your team reads in. Create a Funnel walks through each step of the wizard.

Note

A Space keeps the Funnel it was created with, and editing a Funnel later applies only to the files uploaded after the edit. Settle the labels of the content classifier before the first file goes in.

Step 2: Create a Call Center Space

A Space holds the feedback of one channel, so the call center gets a Space of its own. Open SPACES in the main menu and click New space. The wizard asks for the Funnel first and then for the Space's details. See Create a Space for every field.

Choose the Funnel you built in Step 1, then name the Space after its channel. The name field suggests Call Center, and a channel name stays right for as long as the Space lives, unlike a name such as "October calls". The optional description holds up to 250 characters and is the place to say where the transcripts come from and how often they are uploaded, for example "Weekly export of call transcripts from the contact center platform". Click Create Space.

[Screenshot: The Details step of the new Space wizard, with the name Call Center and a description filled in]

Tip

If your chat or email support has a Space of its own, give it the same Funnel as the call center. Spaces with the same Funnel give their feedback the same themes, sentiment and Dimension answers, so the channels can be read side by side, and Home shows them together.

The new Space is empty. It appears in the main menu under HOME and on the Spaces list with No upload yet, and it waits for its first file.

Step 3: Upload Transcripts Regularly

The first file you upload sets the Space's schema, and every later file is matched to it, so the first upload is the one to prepare with care. From then on the work is a routine: export the new conversations, upload them and let Kimola classify them. See Add Files to a Space for the whole wizard.

Open the Space and click Upload the first file, then choose your file in the Dataset step. In the Schema step, choose the Date Column, the column with the date of each call, and then the Text Fields. A text field is a column that Kimola classifies, so name it after what it holds, such as Transcript. When one conversation is spread over several columns, click Join another column to classify them as one text. When the export also carries a text that should be read apart, such as the agent's wrap-up note, click Add another text field and give it a field of its own. Every other column of the file is kept with each record and is not classified.

[Screenshot: The Schema step with the Date Column chosen and a Text Field named Transcript]

Tip

Choose the date of the call, not the date of the export. Every upload is placed on the timeline by this column, and the Space's trends follow it.

If the export identifies the customer, choose the Customer Column, the column with a customer number, an ID or an email. It is optional and can also be added later. See Add a Customer Column to a Space.

Note

A column that is not in the first file's schema is not kept from later files. Include every column you may want to filter or compare by later, such as the queue, the product or the agent team.

In the Upload step, check the SUMMARY, which shows the file's rows and the queries it will use, and click Upload. If the Space already has a file with the same name, Kimola warns you before it counts the same export twice. Tick It is a different file, upload it anyway only when it is a new export that happens to have the same name, as a weekly export often does.

Note

Queries are counted as rows × models × Dimensions, so a file of many conversations and a Funnel with several Dimensions use more queries. The summary shows the number before you upload. See How to Calculate Queries, and kimola.com/pricing for the queries in each package.

Kimola then classifies every text field of every row with every model of the Funnel. On the Files page the file shows Waiting to be classified, then Classifying with its progress, then Classified, and you can leave the page in the meantime. A file counts in the Space only once it is classified.

Every upload after the first uses the Columns step in place of the Schema step. Kimola matches the columns of the file to the schema by name and remembers the matches you choose, so a regular export that keeps its column names is matched without asking again.

Choose a rhythm that fits how your team works, such as one export every Monday or one at the end of every month, and keep to it. Each upload is read against the records before it, so a steady rhythm keeps the Space's trends easy to read.

Tip

If a wrong file went in, take it back from the Files page with Take this file back. Its records leave the Space and its numbers are recalculated. See Take Back a File from a Space.

Step 4: Watch NPS and Signals

When a file is classified, Kimola brings the Space up to date: it recalculates the Space's NPS, runs the Signals again over all its records and writes the Insights again where the evidence moved. You don't start any of this. The first Signals and the first Insights appear after the first file, and the Signals item joins the Space's menu once the Space has its first Signals.

Start on Overview, the first page of the Space. At the top, What your feedback says now is a short text about what customers talk about in the Space's latest records. Next to it, With the latest files lists the themes and Dimension answers that are new, rising or falling in the newest files compared with the records before them, and Fading lately lists what customers mention less often now. Read these after every upload, and open a theme to read its own text and its quotes. See Overview in Spaces and What is an Insight?.

[Screenshot: The Overview of a Space with What your feedback says now, With the latest files and Fading lately]

The NPS card shows in one number, from −100 to 100, how customers feel: the share of positive conversations minus the share of negative ones, as the Funnel's sentiment classifier labels them. It is not a survey score. Conversations that are neither positive nor negative, such as plain questions and requests, count in the total but in neither share, so what matters is the direction the NPS takes from one upload to the next and the NPS of each theme next to it. See How NPS Is Calculated in Spaces.

BEST PERFORMING THEMES and WORST PERFORMING THEMES list the three themes customers are most and least satisfied with. They are ranked by Satisfaction and Dissatisfaction Scores, which are adjusted for the size of a theme, so a theme that hundreds of customers complain about ranks above a handful of angry calls. Performance in Spaces shows the scores of every theme.

Then open Signals to read the findings that passed a statistical test. Differences show where a topic is mentioned more, or more negatively, in one group than in the others. In a call center, the groups come from the columns you kept, such as the queue, the product or the agent team, from the answers of the Funnel's Dimensions and from the files themselves. Every file is a group too, so a topic that stands out in the latest upload becomes a Signal even when its share of the whole history has barely moved. Trends show how the mentions of a topic changed over recent periods. Every card gives the numbers, the group it is compared with, a few quotes and Why this is shown.

[Screenshot: The Signals page of a Space with a Signal card about the latest file and Why this is shown open]

Example

For a bank, a file of last month's calls might show card blocking in 11% of its conversations against 4% in the files before it. That difference becomes a Signal about the file, and the quotes on its card show what customers said when their cards were blocked.

Give every Signal a verdict, Known, New & important, Unimportant or Artifact, with an optional note. A verdict follows the same finding into later runs, so each upload's review starts where the last one ended. Open in Content opens the conversations behind a Signal, which is the quickest way to judge it. See Signals in Spaces, What is a Signal? and Review Signals.

Note

A new Space can have no Signals yet. When the tests find no group or period that stands apart with enough evidence, the Signals item stays hidden. More files, or more descriptive columns in them, give the tests more to work with.

Step 5: Make Tasks

A Signal tells you that something stands apart. A Task gives that finding to the teammate who can act on it, together with the feedback and the numbers behind it. You make Tasks where the finding is: click Make a task on a Signal card, or on a theme in ALL THEMES on Overview, or filter the conversations of one queue or product on the Content page and click Make a task of the feedback these filters show. See Create a Task for every place a Task can start.

A window opens with a draft title and description that Kimola writes from the finding, in the language of the Funnel: why it matters, what to do first and how you will know it is done. Edit the draft, choose the Assignee, an optional Due date and the Priority, and add under Keep informed the teammates who should be told when the Task changes. Then click Create the task. If Jira is connected for your team, Also send to Jira sends the Task there as well. See Send a Task to Jira.

[Screenshot: The Make it a task window with a drafted title and description and the Assignee, Due date and Priority fields]

Example

For an insurer's call center, a Signal shows claim status inquiries in 14% of the latest file's conversations against 5% in the files before it. The claims manager gets a Task titled "Tell customers where their claim stands before they call", with the numbers and a few quotes attached.

Note

If a teammate has already made an open Task of the same finding, the window shows it first, so you can follow it instead of starting a second one.

The Task keeps the finding's evidence as it was the day you made it. After the Task is done, Kimola reads the same feedback again two and four weeks later, in the files uploaded since, and tells you on the Task's page whether customers noticed. This is why the regular uploads of Step 3 matter beyond trends: they are the feedback the measurement reads. See How Kimola Measures Task Impact and Manage Tasks.

What You Get

Themes and NPS. After the first upload, the Space shows its records, its NPS and the themes of your content classifier with their shares and their own NPS, so you see what customers call about and how they feel in each topic. The Dimensions of the Funnel answer further questions for every conversation, such as what goes wrong or what customers miss.

What changed. After every later upload, Kimola writes the Space's text and the texts of its themes again where the evidence moved, so you read what the latest file changed instead of searching for it.

Signals. Statistically tested findings show which topics, queues, products or teams stand apart, and which topics rose or fell over time, each with its numbers and quotes.

Tasks. A finding you act on becomes a Task with its evidence kept as it was that day, and once the Task is done, a measurement says whether customers mentioned the problem less.

Customers. When the export carries a customer column, every customer gets a page with all of their conversations across your Spaces, and you can sort Customers by Lowest NPS to find the ones who write most negatively.

Next Steps

Once the Space runs on its own rhythm, Compare in Spaces puts two groups of conversations side by side, for example two queues or two products. Actions in Spaces writes an Action Plan from everything the Space has classified, with up to 12 evidence-checked actions that you can turn into Tasks. If the same customers also write in your other channels, a Customer Column gathers their feedback from every Space on one page. When different teams own different parts of the call center's feedback, the Kimola team can set up Departments with you; talk to your Account Manager.

To see what Kimola does with support and call conversations, visit kimola.com/conversation-analysis. For ticket data from Zendesk or Intercom, see How to Analyze Support Tickets.

Was this article helpful?

Tell us about your thoughts and experiences regarding the article.