- More detail can be found within our Conversation Analytics Playbook > Deploying Conversation Analytics > Creating and managing Insight Topics
To help get you started on your Analytic journey, Evaluagent has provided a core set of topics for you to use immediately, revealing insights into all conversations handled by your agents every day.
You can easily customise any of our topics by applying your own specific terminology, branding, products and service.
To do this, simply select the topic you want to customise, click > and from the ‘Active’ topic, click on the burger menu and select Duplicate > Create a brand-new.
A copy will now be saved in your work tab named ‘Our topics’, where you can start customising the terminology using the topic editor. By selecting edit from the topic’s burger menu, you will be presented with the topic editor, which will slide into view (see an example of a logic-based insight topic below).
Alternatively, you can create your own insight topic from scratch to automatically surface those service-critical moments, finding the most important conversations for you and your business.
When creating a logic-based insight topic, you will need a list of keywords and phrases you would expect to see within conversations between your agents and customers. To ensure you are targeting the right conversations in the right context, you should try to explore every possible word or phrase whilst being conscious of how words are grouped, how agents structure their conversations, and who said what.
By the end of this process, your lists or prompts can become quite extensive, but don’t worry; you can continue to build and enhance these topics over time.
Once this is done, you can open the topic builder by clicking on the +New topic button.
From here, options for technology are presented, select Logic.
Then, options for the type of topic are presented, select Insight topic.
The topic builder will then slide into view.
The above is an example of the logic-based insight topic builder.
Here, you will name your topic, provide a description, and assign a theme to help users fully understand what the topic is designed to do.
You can now start to add the specific words/phrases you want to identify, either individually or by uploading your list from a CSV file.
In the example below, you can see a new topic starting to be formed around Identification and Verification.
Once you’ve added all your terms, you can start to configure each query ‘pill’ (word/phrase) individually. Click on the pill to view and set conditions for your query. These can include:
- Speaker – defines who said the word or phrase: the agent, customer or either
- Slop – the maximum number of words that can exist between the start and end of a phrase or sentence (this is not presented when singular words are used)
- Searched in order – defines the definition for a ‘found’ word or phrase either in the order they are set out using the THEN operator or in any order using the AND operator throughout the whole conversation (this is not presented when singular words are used)
- Search across utterances – groups utterances into one text string either spoken by the agent, customer or both, depending on the speaker conditions set
- Search entities – whether or not entities are to be included in the search. These are explained in Entity & Redaction help guides
- Query profiler – This adjusts how the search is executed and what results are returned. We recommend using the ‘unified’ or ‘default’ drivers in most instances
The example above shows the query conditions for identifying where the agent asks the customer for their name.
It searches against all utterances from the agent, the words must appear within 2 words of each other and be in the order you have defined. This might come in a statement like this…
Agent: “To complete DPA can I please take your name, first line of address and postcode”
Pass conditions for the topic can now be set, for example you might need 3 pieces of information to be heard before the overall topic is considered a pass, for example "your name", "date of birth" and "first line of address".
Additional functionality that can be applied to a single word (term) includes:
- Wildcard searches -These are special characters that can search for variable characters within a word. The asterisk (*) can be used anywhere within a character string: beginning / middle / ends of a root word. For example, educat* would return results including educate, educated, education, educational or educator
- Fuzzy searches - Look for words that match closely but not exactly. This type of search will return matches even when there are grammar / spelling mistakes, for example access and excess can be misused / mis-transcribed (in calls) but if these were turned into a fuzzy search with a Levenshtein distance of 2 (access / excess) then both words would be returned.
- Synonyms - Help to expand queries to look for similar words with the same meaning. For example, if you search the word ‘Dog’ the query will also returns results of ‘pooch’, ‘canine’, ‘pup’, ‘hound’ etc.
- Phonetic searches - Returns phonetically equivalent results based on your desired search for example, if you search the name 'John' you would get hits for 'Jon' also.
- Term search- will match the word / term exactly
Finally, once you save your topic, it will appear in the manage topic table under the tab ‘Our Topics’.
- Before conversations are tagged with insight topics they have to be made active from the burger menu drop down in 'Manage Topic' table. This will ensure all active topics can be seen across the relevant Auto-QA reports.
- A topic can be deactivated from the same place only if they have not been attached / linked to a scorecard.
Related Help Guides:
- How do I test the accuracy of an Insight Topic?
- How do I edit a Topic?
- How do I create filters based on Conversation Insights and Auto-QA scorecards?
- How do I monitor conversations that are not matched to an Insight Topic or Auto-QA scorecard?