Conversation AI Search Strategy Guide




Context is Key

The fundamental rule to remember regarding search criteria and strategies within Conversation AI is that Context is Key. The context of any conversation can significantly change the underlying definition of what was said.

For example, knowing if call participants mentioned “pain point” keywords within any conversation is not enough on its own to understand what is happening within the conversation. This is where additional context elements, such as the stage of the opportunity at the time of the conversation, can help you differentiate between two completely different conversations.

If the conversation with pain point keywords occurs within an early ‘Discovery’ stage of an opportunity, there is a good chance that the participants are talking about their current problems and pain points for which they are seeking a solution.

On the other hand, if the conversation with pain point keywords occurs within a late ‘Pilot’ stage of an opportunity, there is a chance that the participants are surfacing pain points they incurred when piloting your product/solution.

As such, it is essential to include as much context as possible within your Conversation AI search criteria.

The easiest way to think of context for your search is to think of the Five W’s (and an H) in Journalism: Who, What, When, Where, Why, and How (hereafter referred to as context elements).

Sometimes you may be unable to provide all six of these context elements as your goal is to find the answer to one or several of these context elements. When this happens, adding as much definition to the context elements you do have/can provide is crucial to filter out noise in terms of surfacing valuable insights towards your goal.

Let’s explore each of these context elements to consider within the framework of strategic Conversation AI search criteria. 


The Why - Your Goals, Key Questions, and Desired Outcomes

Conversation AI offers a wealth of data; as such, the ‘Why’ is the most vital of all context elements. It can be easy to get lost or overwhelmed without a clear focus on what you are looking for or trying to answer.

The best place to start is by clearly defining what you are ultimately trying to learn or influence. Then, after mapping out your ultimate goal, outline any additional questions you may need to answer to achieve that goal. Typically, the more specific a question is, the easier it is to answer.

By approaching the definition of your ultimate ‘Why’ through a series of qualitative and quantitative questions, and by breaking down all knowns and unknowns, you will naturally begin to surface all the other contextual elements that construct a well-defined Conversation AI query.

In a healthy RevOps organization, the respective goals and questions naturally build upon one another as you navigate hierarchical reporting structures.mceclip0.png


It is good practice to make sure you are asking the right questions based on the true intent of what you are hoping to achieve. If you start based on an underlying assumption, use Conversation AI to prove or disprove that assumption. Setting a hypothesis and segmenting conversations and insights to prove or disprove it can give you the confidence and validation to operate on that assumption.

Use Conversation AI to ask a wide range of question types, such as comparative questions to understand key differences between different entities (teams, people, products, etc.), relationship questions to understand correlations between variables and outcomes, and questions that surface cause-and-effect patterns. 

Key Takeaway: Having a clear idea of the questions you are trying to answer or what you are trying to learn will help ensure you get the most value out of Conversation AI.


The What - Things Said or Not Said

The most commonly pictured surface-level use case of Conversation AI is finding conversations in which something was said. However, an equally as insightful yet widely overlooked Conversation AI use case is searching for conversations where something was NOT said

Within Conversation Ai, there are two places to define the ‘What’ context element:

  1. Keywords and Keyword Groups
    1. Use the Keywords filter to add entire Keyword Groups or specific Keywords to your search.
    2. After adding a Keyword/Group, change the search logic to only show conversations where a keyword was NOT said by clicking on the “+” icon next to each keyword within the search criteria. This should change it to a “-” 


  1. Transcript Text Search
    • Use the Transcript Text Search to find conversations where something was or was not mentioned
      • Place your searched phrase between quotation marks to only return exact phrase matches. For example:
        •  A search entry of “Do you use Salesforce” will return only conversations which used that exact phrase.
        • A search entry of Do you use Salesforce will return conversations that contain all the words within the search criteria but are not necessarily used together/within the exact spoken phrase.
      • After adding a word or phrase to the transcript search, to change the search logic to only show conversations that do NOT include the searched word/phrase, click on the “+” icon next to each keyword within the search criteria, which changes it to a “-” 


When searching for conversations with (or without) specific keywords, be as selective and focused as possible when specifying the keyword filters. In the interest of selectivity and focus, it is best to either add each keyword individually from the Keywords filter or, if you add an entire keyword group, review and remove any unnecessary keywords from the Search Criteria.


While a clear definition of the “What” context element in itself is powerful, the value of this filter is multiplied when used in conjunction with additional context elements, most notably the “Who” element. This can drastically change the intent and definition of the “What” element.


The Who - Conversation Attendees

Think of the “Who” context element as the conversation attendees. Did you notice it says ‘attendees’ and not ‘speakers’ of a conversation? An often overlooked dynamic of a conversation, especially a video-based conversation that can have numerous attendees (as opposed to a traditional two-party call), is the consideration of all the attendees of a conversation, including attendees who did not speak/contribute to the conversation. 


The Taxonomy of Who

There are two overarching cohorts within the ‘Who’ context element, Internal Attendees (attendees from your company) and External Attendees (attendees from outside your company such as leads, contacts, or customers). Within each of the two cohorts, there are three tiers of ‘who’ to consider for your search.


The Person: Searching for conversations based on specific attendees (Hosted by a specific Rep or attended by a particular participant.)

The Role: Searching for conversations based on attendees' roles. (You can search by participant titles, but another qualitative aspect to keep in mind is the participants' role within the conversation, such as decision maker, champion, stakeholder, gatekeeper, SME, etc.)

The Team/Company: Searching for conversations based on an internal team or external participant company (Particularly useful when evaluating all members of an internal team or trying to find all conversations held with an external company.)
While it is best to define as much context as possible, the pairing of the ‘Who’ and ‘What’ contexts alone creates robust search use cases such as finding conversations…

  • where something was (or was not) said by a specific participant, 
  • with a particular person in attendance, and something was (or was not) said 
  • held by any member of a specific team in which something was (or was not) said.


The Where - Customer Journey & Outcomes

Within the constructs of journalism, the ‘Where’ context is most commonly associated with a geographical or physical location. However, within the realm of RevOps and defining filterable context elements for Conversation AI, taking into account where a conversation occurred within the stages of a Customer Journey is much more insightful. In addition to identifying the Customer Journey stage, filtering conversations with positive or negative outcomes is valuable in identifying patterns and critical differences between conversation outcomes.


Customer Journey

The previously provided scenario of searching for conversations where the participant said words/phrases within the ‘pain point’ is a prime example of how the stage in the customer journey can dramatically alter the significance of what someone says. Keywords related to ‘pain’ said by the lead/customer within early discovery phase conversations are common and expected as they are likely discussing their current pain points/needs for your service. However, suppose the customer says ‘pain’ keywords within the late pilot/decision stage. In that case, the conversation is much more concerning as it may indicate they had problems during the pilot and the deal/opportunity is at risk.

Within Salesforce, the opportunities object is commonly used to track the stages and progression of all potential revenue-generating opportunities. Using the Opportunity Stage filters is an easy way to segment conversations specific to a deal stage.



There are two different variants of outcomes worth considering in regard to searching for conversations with coachable moments. 

The first outcome variant is the outcome of individual conversations, which usually corresponds to the customer journey stage and the desired result of progressing the opportunity and discussions to the next stage of the customer journey.

The second outcome variant is the outcome of the overall opportunity or deal. Searching and reviewing conversations leading up to closed/won opportunities and closed/lost opportunities leads to significant reflective learning/coaching opportunities.


The When - Date, Duration & Conversation Proximity 

There are three primary ‘When’ context elements to consider within your Conversation AI search parameters (if your use case permits), as they can significantly reduce the number of irrelevant conversations returned.



If you are only searching for recent and actionable conversations occurring within the current month or quarter, specifying the timeframe will help reduce any noise generated by presenting conversations outside of the scope you are seeking.




Defining a minimum or maximum conversation duration is another quick way to filter out results that would not be valuable for what you seek. For example, if you want to provide coaching regarding live demos, you may want to increase the minimum conversation duration time filter via the timeframe duration slider. By increasing the minimum duration, conversations significantly shorter than a demo's average duration are not displayed.  


Conversation Proximity

Conversation AI lets you fine-tune the criteria around keyword searches. This is particularly helpful if your search use case is related to discovering conversations with coachable moments related to specific call segments (e.g., feedback specific to conversation introductions or wrap-ups.) The ‘Mentioned at’ filter for Keyword search allows you to refine your search results to only show conversations where something was (or was not) said within a specified percentage of the conversation, in proximity to the conversation’s total duration (i.e., selecting the 0-20% ‘Mentioned at’ range for a five-minute call will limit the results to the first minute of the conversation.)



The How - Conversation Etiquette

A popular quote paraphrased as “People may forget what you said, but people will never forget how you made them feel” (attributed to Carl W. Buehner and Maya Angelou) underscores a significant point regarding the impact that emotion has on communication. 

This concept, in terms of conversational effectiveness within a business/professional setting, signifies that even if the content of a message itself is perfect, the delivery of the message and invoked emotional response can positively or negatively impact/influence the recipient’s perception of that message. In simple terms, it’s not just about what is said but how it is said.

As such, in terms of Conversation AI, Conversational Etiquette metrics are an excellent representation of ‘How’ something was said, making it easy to search for conversations based on coachable behaviors. 



Was this article helpful?