Building topics can require manually crafting a set of specific terms and rules to cover potential variations, such as synonyms, misspellings, and slop distances. For instance, creating a topic like "dissatisfaction" would require inputting terms such as “unhappy,” “frustration,” and “displeased.”
While powerful, this manual approach can be time-consuming.
Introducing 'Build with AI'
To address these challenges, we have recently launched a build with AI feature that simplifies the process with the LLM generating a robust set of query terms, making your topics more comprehensive and less prone to error.
This guide will walk you through how to use this feature to save time, improve accuracy, and ensure that no key customer sentiment is overlooked.
By following the prompt above, you will be directed to the topic manager screen where you can manage and create new topics.
To begin, click on 'new topic' where you will be given the option to create an insight topic or line item. For the purpose of this guide, we will create an insight topic using the text analytics technology type. Following these steps will bring in the modal where we get to create our topic and corresponding queries.
Start by naming your topic, providing a description, and assigning a theme. These details will help users understand what the topic has been designed to do. Next, click on the 'build with AI' text which will prompt you to complete the following fields;
- Interaction format: This covers the format in which the interaction is expected to take place i.e. calls, email, live chat etc.
- Interaction theme: This refers to the recurring idea discussed in the conversations. For example customer dissatisfaction, damaged products, complaints, refund & payments etc.
- Interaction description: This covers what you would like the topic to look out for specifically. For best practices and sample descriptions, please scroll down.
- Examples: Include at least 3 examples that signpost what this topic might look like in a real-life interaction between your agent and the customer.
Once the fields are filled, click on the 'start topic build' button to commence.
Using the 'dissatisfaction' example from above, the LLM has generated a robust list of queries accounting for who said what between the agent and the customer.
The flexibility of this feature allows you to manage the queries by removing or adding more queries and configuring each query ‘pill’ (word/phrase) individually. Simply click on the pill to view and set conditions specific to your query.
Note: The same process applies when building a text analytics line item.
Best practices for building topics with AI
- Be clear and specific: The more specific and clear your initial description and examples are, the better the LLM will perform. Avoid ambiguous phrases.
- Limit complexity to one aspect: It is best to focus on one aspect at a time. For example, if you want to track dissatisfaction, don’t combine it with other aspects like product inquiry or pricing issues in the same topic.
- Include relevant context: To help the LLM generate accurate and useful queries, it’s important to provide relevant context in the interaction description. This context tells the AI what kind of interactions or language to focus on, making the generated query more aligned with your needs.
Sample descriptions to get the best output
| S/N | Type | Description | Rationale |
| 1. | Bad Example | The agent does not set expectations with the customer. | The description is not specific enough. |
| 2. | Good Customer Example | Identify when a customer expresses dissatisfaction by identifying negative language, complaints, or expressions of frustration, such as "this is not acceptable," "I'm unhappy," or "this is the worst service.” Restrict the queries to only customer responses, do not include any agent responses only customer ones. The focus should be on how the customer has expressed dissatisfaction and nothing else. Generate the queries but limit to 5 words per query |
The description is specific and focuses on one theme- customer dissatisfaction. |
| 3. | Good Agent Example | Identify when an agent sets expectations by identifying phrases that provide timeframes, next steps, or specific outcomes. Look for language such as "I will," "you can expect," "we will get back to you by," or "this will be resolved within”. Restrict the queries to only agent responses, do not include any customer responses only agent ones. The focus should be on how the agent sets expectations and nothing else. Generate the queries but limit to 5 words per query | The description is specific and focuses on one theme- agent setting expectations. |
| 4. | Good Example (line item) | The agent asks the customer to verify their identity by stating their full name, their postcode, and their location. Generate the queries but limit to 5 words per query | The description is specific and focuses on verifying the identity of the customer. |
Frequently Asked Questions (FAQs)
- Will I get charged for using this feature?
No, there are no extra charges for using this feature. If you have any further questions relating to cost please contact your account manager. - Can I still build a topic without using the build with AI feature?
Yes, the option to manually build queries still exists, simply create your queries without clicking on the 'build with AI' text. The Gen AI-powered topic creation feature is an option, not a replacement for the manual method. - What if the LLM does not include the terms I want?
After the LLM generates a query, you have the opportunity to review and refine it. If there are missed terms you consider important, you can manually add them. - Is there a limit to how many topics I can create using this feature?
No, there is no limit. Build as many topics as you want and learn as you go!
Related Articles
How do I build an insight topic
Comments
0 comments
Article is closed for comments.