Plan topics

The planning process involves brainstorming an initial list of topics based on the themes in customer comments, refining the list and building tentative topics in the system to verify that your topics capture results, and presenting the topic list to company stakeholders for their approval.

Relevance, sample size, and actionability are key characteristics for optimal topic design. Consider these questions as you plan:

• Relevance

  • What are the subjects does the organization wants to track?

  • What are the subjects Medallia sees the organization's customers frequently discussing?

  • What are the topics that Medallia knows are relevant to the organization's industry?

• Sample size

  • Do these topics have traction? This is especially important for the reliability of statistics shown in reporting.

• Actionability

  • Are these topics specific enough that users can understand needed actions from topic names?

  • Are the topics useful to the organization?

Your topic list should cover the main areas of the organization and the customer experience. The goal is to identify the main factors affecting customer loyalty overall. You do not need to create a topic for every single subject that customers discuss, but the topic list should reflect the themes trending in the majority of the comments.

1. Export at least 1000 survey records.

2. Read 600-700 comments and record the topics you see trending, based on the comments you read.

   As you read, take notes about the subject of each comment. Include specific details in your notes. You will need to refer to these notes later when creating your topic list. For example, for comments such as It was easy to pay for my item or It was easy to assemble the furniture, rather than record the theme of these comments as ease, describe the theme in more detail, such as ease of payment and ease of assembly. Note the following tips for brainstorming topics:

   • Look at your company's website. Often, help and FAQ sections are valuable for gaining insight about the subjects customers are discussing.

   • Look at the type of survey program the topic list is meant to address, as well as the questions in the survey. Is the survey about customers’ online experience? If so, add topics relevant to the online experience and brainstorm other themes that customers may be discussing about their online experience that are not already represented on the list.

   • Read through the topic list and continue to hypothesize. For example, if there are topics on shipping, think about things that might be missing from the shipping experience, and add topics about them to fill gaps.

3. Create a list of topics from the themes you recorded.

   Use your notes, along with the additional information you learned from other sources, to create the initial list of topics. (For information about topic types and examples of topics, see Topics.)

4. Organize the topics into a parent-child structure.

   Create broad parent topics and define more specific child topics for each parent topic. For example, you might have Shopping Environment as a parent topic, with Shopping Environment - Layout and Shopping Environment - Navigation as child topics.

   The advantages of the parent-child hierarchy become clear in reports. While child topics help companies identify the aspects of a particular subject that have the greatest impact on customer loyalty, parent topics can help companies view the data at a high level.

   In the following example of a report filter, filtering by the GUEST ROOM parent topic enables users to view the aggregate results from all the child topics associated with GUEST ROOM (Guest Room Overall, Guest Room Amenities, and so on). There are no rules associated with GUEST ROOM. Instead, the GUEST ROOM parent topic inherits the captured comments from its child topics.

   

   Note: Every topic terminates at the lowest level. For example, a two-level hierarchy terminates at level 2. Only the lowest-level topic should contain rules. Also, there should be no inconsistency or mixing of levels. For example, a two-level hierarchy with some topics terminating at level 1, and other topics terminating at level 2.

Refine your topics

1. Read 300-400 more comments to identify additional topics you missed, and validate the topics that are already on your list. Assess and refine the list, while considering the comments each topic is capturing:

   • Look for topic redundancies and edit the topic list as necessary.
   • Remove topics that are impossible to build.
   • Look at the topics from the company’s perspective, and only keep topics that the company will find actionable or useful for tracking purposes.
   • Consider which segments the company will be able to use in filters. Some topics on the list may work better as segments rather than topics, and some topics may already be represented by a segment already present, creating a redundancy.

2. Create basic versions of your topics using your list, and pare down the list as necessary. For information about creating topics and rules, see Topics screen.

   As you add each topic, add 5–10 of the most obvious rules for the topic to verify that comments about the topic exist. The purpose of this step is to validate that comments about the topic exist, and that the observations you recorded earlier extend to the rest of the data. Later, after stakeholders at your company approve your topic list, you can add a complete set of rules for each topic. For now, just verify that your topics capture comments. Pay attention to the number of comments the topics are retrieving, and pare down the topic list as necessary:
   • Merge — For topics that are related in theme but do not capture many comments, merge them together. For instance, you could merge an Attitude topic with a Professionalism topic if one or both of those topics are low in volume.
   • Remove — Remove topics that capture very few comments.
   • Truncate — If the list is supposed to be 50 topics long and your final list is 70 topics, keep the fifty topics that capture the highest number of comments.
   • Defer — Topics that are pulling in some volume but not enough to make it onto the final list can be presented to company stakeholders as potential topics to add at a later date.
   • Sort — Order the list. Ensure that topics are arranged in a way that makes sense for reporting.

As you finalize the topic list, present your potential topic list to company stakeholders before fully building the topics, and provide them some time to review the list.

Create topics

To create a topic:

1. Navigate to Admin Suite > Text Analytics > Topic Builder, or use the sidebar to select the Text Analytics > Topics option. The Topics screen displays.

2. Use the Topic Set dropdown to select a topic set on the Topics screen.

3. To create a new parent topic, click New Topic, and then click either Standard or Compound. To create a new child topic, hover over a parent topic, click the topic menu, click Add subtopic, and then select either Standard or Compound. For example:

   

   A new topic appears in the list.

   1. Additional list entries appear each time you add a new draft. To remove the additional list entry, click the cancel icon or use the Escape key.

4. Enter a name for the topic.

   If you create a hierarchy for one set of topics, make sure the other topics in your topic set have the same number of hierarchy levels so that all topics have a similar level of granularity. Limit the topic hierarchy to two levels (parent-child). When naming parent topics, ensure they communicate the general subject area.

   Note: The name of a topic is its unique identifier. In the legacy Topic Builder this was known as the Topic ID.

   Be aware that the topic name does not automatically become a rule. For example, for a topic called Bathroom, you would still need to create a rule with the bathroom keyword if you wanted the topic to capture phrases containing that word. For more information, see Create standard topic rules and Create compound topic rules.

5. Optionally, to change a topic's name, click the edit icon next to the topic name.

Activate topics

Activate new or inactive topics to make them available for text processing and reports. While you can activate topics at any time, note the following considerations:

• Processing takes longer than usual if a deployment is in progress.

• Experience Cloud might display an error message if you try to activate or publish topics when text processing (parsing) is in progress. If this happens, try activating or publishing after parsing completes.

• Medallia recommends activating or publishing together as many topics from the same topic set as possible. Processing a single topic takes a similar amount of time as processing multiple topics, so bulk processing tends to be faster.

• Medallia recommends a two-part historical processing strategy for new and updated topics. First, activate or publish topics without tagging historical data so incoming data is tagged immediately. Then perform historical processing with all new and updated topics together once topic rules are edited and finalized.

To activate a topic:

1. On the Topics screen, select one or more new or inactive topics.

2. In the toolbar, click Activate.

   The Activate Topics dialog appears.

3. In the Language property, select whether to process topic rules in only the current language or in all languages.

4. Select whether to Tag Historical Data.

   By default, activated topics apply only to new feedback. You can choose whether to process older data for the topics.

5. If you chose to process historical data, in the Time Period property, select the amount of historical data to process.

   Selecting a shorter time period reduces processing time. Additionally, time-based topics might be relevant only for a specific time period. For example, a topic for New Year’s Sale might be relevant only in January for the current year.

   

6. Click Activate.

   The topics are activated. If you chose to process historical data, the processing begins. For more information, see Process historical data, below.

Publish updated topics

You might need to make changes to a topic after it was first activated. Your topic remains in the Active state while you edit, but an orange circle next to the topic name indicates the topic has unpublished changes. When you are ready for your changes to go live, publish your topics.

While you can publish topics at any time, note the following considerations:

To publish a topic:

1. On the Topics screen, select one or more active topics to publish.

2. In the toolbar, click Publish Changes.

   The Publish Changes dialog appears.

   

3. In the Language property, select whether to process topic rules in only the current language or in all languages.

4. Select whether to Tag Historical Data.

   By default, activated topics apply only to new feedback. You can choose whether to process older data for the topics.

5. If you chose to process historical data, in the Time Period property, select the amount of historical data to process.

   Selecting a shorter time period reduces processing time. Additionally, time-based topics might be relevant only for a specific time period. For example, a topic for New Year’s Sale might be relevant only in January for the current year.

6. Click Publish Changes.

   Note: If there are topics referenced in rules, any unpublished changes for those topics are published automatically and those topics are activated automatically, for the rules appropriate to your Language and Segment selection. The Publish Changes dialog includes a list of those topics.

   Changes to your topics are published. If you chose to process historical data, the processing begins. For more information, see Process historical data, below.

Process historical data

After you activate a new topic, text processing for that topic is applied only to new feedback. Similarly, if you publish changes to an active topic, those changes apply only to new feedback. Process historical data to apply your new and updated topics to older data, and to make the results of that processing available in reports.

1. On the Topics screen, select the active topics for which you want to process historical data.

2. In the toolbar, click Update Historical Data.

   The Update Historical Data dialog appears.

   

3. In the Language property, select whether to process historical data using rules in only the current language or in all languages.

   If there are topics referenced in rules, any unpublished changes for those topics are published automatically for the rules appropriate to your Language selection. For example, if you select All for the Language, and a rule in a topic being processed references another rule, any unpublished changes in that referenced rule are published automatically.

4. In the Time Period property, select the amount of historical data to process.

   Selecting a shorter time period reduces processing time, and Medallia recommends choosing shorter time periods for historical processing. Additionally, time-based topics might be relevant only for a specific time period. For example, a topic for New Year’s Sale might be relevant only in January for the current year.

5. Click Save.

   Historical-data processing begins for the selected topics.

Initiating historical-data processing for one or more topics causes the Historical Processing column to appear on the Topics screen. This happens when you initiate historical processing via any method, such as clicking Update Historical Data, processing historical data through topic activation, or publishing topics with historical data. The Historical Processing column shows one of the following processing states for each column:

• In Queue — The topic is in the queue to be processed.

• Progress bar — For topics being processed currently, a progress bar shows the percentage-complete of that processing.

• Cancelled — An administrator has cancelled historical processing for the topic. To cancel processing, hover over the In Queue state or the progress bar for a topic, and then click the X. To restart processing for a topic, hover over the Cancelled state, and then click Refresh.

• Failed — Historical processing failed for the topic. When this happens, contact Medallia Customer Support for assistance.

Text Analytics processes historical data from the most recent data to older data, so newer data is available in reports before older data.