- About this Guide
- Types of Keyword Groups
- Keyword Tuning and Management Strategies
- General Keyword Tips
About this Guide
Keyword Groups and the Keywords within each group are a significant part of Conversation AI. In addition to the contextual applications described above, Keyword Groups/Keywords can be applied as a segmentation/filter within several insight reports, allowing you to identify trends and correlations surrounding the use of specific Keyword Groups/Keywords.
This guide will provide some strategies and tips for creating and maintaining Keywords and Keyword Groups. For step-by-step instructions on creating Keywords and Keyword Groups within Conversation AI, check out this support article.
Styles of Keyword Groups
The three primary styles of Keyword Group categories are Intents, Entities, and Behavioral.
Like their name, intents can be thought of as a person’s aim or purpose behind something. Frequently in business contexts, the intent of what is said can be categorized into the ‘underlying why’ something is said or asked rather than the explicit meaning of a phrase.
For example, if a sales rep asks a contact, “How much time do you waste dealing with this issue?” the underlying intent would be along the lines of “Problem Discovery.”
Examples of Common Intent-Based Keyword Groups:
- Problem Discovery
- Words/phrases used to surface problems
- Determine Budget
- Words/phrases used to provide insight into how much lead/contact can spend
- Build Rapport
- Words/phrases used to build rapport and strengthen a relationship with a lead/contact
- Identify Decision Criteria
- Words/phrases used to surface what leads/contacts value the most when evaluating solutions
- Identify Stakeholders and Decision Makers
- Words/phrases to understand everyone that would benefit from a product/service or involved in making a decision
- Show Me You Know Me
- Words/phrases that demonstrate a rep has done some research and understands me or my needs before reaching out
Entity-style Keyword Groups are groupings of significant/important nouns. They are particularly useful for measuring how frequently something (the entity) is mentioned over time. A prime example is measuring which competitor names are mentioned in conversations where a lead says they are not interested.
Examples of Typical Entity-Style Keyword Groups:
- Names of competitors
- Names of features or functions
- Names of Products
- Common titles or role names
Behavioral keyword groups are used to track words/phrases associated with demonstrating positive or negative behavior.
Examples of Common Behavioral-Style Keyword Groups:
- Filler Words (*Risky in terms of creating unnecessary keyword noise)
- Words/phrases that do not add value or could distract/detract from the message
- Words/phrases that are considered offensive and unprofessional
- Listening Language
- Words/phrases that demonstrate someone is listening and convey interest/understanding into what is being said
- Words/phrases that could be considered aggressive, pushy, or forceful
- Demo No-Nos
- Words/phrases that should never be used during a product demonstration
- Words/phrases that convey dissatisfaction or opposition
Keyword Tuning and Management Strategies
The most important strategy to consider related to Keyword Groups and ensuring you get the most value out of Conversation AI is that Keyword Groups should be tuned and refined over time. Three primary Keyword tuning and refinement activities are: Reducing Noise, Capturing Value, and Custom Vocabularies.
Keyword noise is when vague words/phrases are added as Keywords. When Keywords are too ambiguous and appear in most conversations, even if they are combined with additional search context elements, they lack real value or conversational insights. A great example of this is conversations with filler words.
An easy way to identify noisy keywords that could be detracting from your ability to find relevant conversations is by using the ‘Over Time’ insights report. The report shows you the most frequently triggered Keyword Groups and the most frequent keywords triggered in each group.
From the ‘Over Time’ insights report, if you see a Keyword Group that stands out as unusually high, add the name of that Keyword Group to the ‘Keyword Group’ filter to get insights on what specific keywords might be driving that volume. In this example, a Keyword Group called ‘Jargon’ seems to stand out since it appears in a large percentage of conversations but does not seem to provide profound insights/value. Refreshing the report after selecting ‘Jargon’ from the Keyword Group filter shows that the keyword ‘like’ is very noisy. Unless there is a strong need to track this, it is worth removing it as a keyword so that it does not create unnecessary noise.
As the name suggests, capturing value tuning involves identifying new keywords and keyword groups that may be valuable and worth tracking. This is the most difficult of all the tuning activities as there is no report or automatic way to surface these items other than having sales reps/coaches notate interesting patterns or phrases they see when reviewing calls. However, the good news is that there is an easy way to validate the value and specificity of newly proposed keywords. To validate, search for the proposed keywords within the transcript search (make sure to place the keywords/phrases within quotes to find exact matches) and see how many pages of results are returned based on that search. If the number of pages appears higher than expected, it may not be specific enough for a keyword. If the number is lower than expected or does not show any results, then there’s a good chance that would not be a valuable keyword either.
So, if a sales rep/coach recommended we add the phrase “peanut butter cookies” as a Keyword, entering it within the transcript search shows that no conversations contained this phrase, indicating this phrase would not be a valuable keyword.
Anytime a new keyword/keyword group is added, it is good practice to periodically monitor and observe Keyword additions after-the-fact to ensure only the conversations you expect are returned and to ensure keyword-related data insights are not getting drowned out since the change was made.
The final Keyword/Keyword Group tuning activity is making updates to the custom vocabulary. If there is a keyword that often is misunderstood by the speech-to-text transcription, adding the keyword to your customer vocabulary list can help increase the likelihood that custom text is recognized correctly. However, as the Custom Vocabulary feature works based on phonetic similarities, short custom vocabulary entries can have unintended consequences as it identifies any phonetic similarities. For example, a custom Vocabulary of “AI” is too short as it is too phonetically similar to common short words such as ‘my’,’ hi’, ’eye’, ‘lie’, etc., causing the transcript to show AI instead of the actual word.
General Keyword Tips
Here are a few final Keyword/Keyword Group related tips that are worth keeping in mind while using Conversation AI:
- Keywords reflect conversations that occur after the keyword was added, so if you want to search for conversations containing a specific word/phrase before it was added as a keyword, use the transcript/text search feature instead.
- Keyword Group names can not be duplicated, but the same keyword can be added to multiple keyword groups. Therefore, if someone has already created a keyword group that has both keywords you find important and keywords you do not find important, you can just create a new keyword group with only keywords you find important. The drawback to this method is that, if not managed accordingly, adding too many similar keyword groups can quickly get out of hand and make it challenging to find the correct keyword groups.
- Create Keywords based on the underlying intent, rather than specific to a single team's use case or application, and use other context elements to narrow down calls relevant to your search. For example, both Sales and Customer Success Teams may be interested in tracking keywords related to ‘pain.’ Instead of creating a keyword group for each team, create a single Pain keyword group. Use the Team level filters and the Pain keyword group to find the conversations relevant to your specific use case.