Best Practices for Parameter Discovery
Updated
Clear and Specific Guidelines for Parameters
Define guidelines clearly and objectively to eliminate any ambiguity or subjectivity. Use specific criteria and concrete examples to ensure consistent interpretation and improve the AI model's accuracy. Avoid vague language that can lead to multiple interpretations.
Clearly defined parameters are also critical for assessing feasibility and aligning expectations. Collaborate with the client during the discovery phase to establish comprehensive guidelines, ensuring all stakeholders share a mutual understanding of the parameters and their definitions.
Example
S.No.
Parameter Name
Incorrect Guideline
Correct Guideline
1
Creative Response
Check if the agent has sent a creative response.
Check if the agent personalized the conversation by mentioning the customer's name, asking about their day, wishing them a happy birthday if mentioned, or engaging in non-work-related talk.
2
Proactive Agent Response
Check if the agent reduced customer efforts.
Check if the agent replied AI model's accuracy and did not make the customer wait for a response.
Information Should Lie in the Case Details
Information should reside within the conversation transcript, including the audio recording, case properties, voice call attributes, and other standard fields such as channel and account. AQM cannot access or evaluate any information outside these sources.
If external systems are used, such as call recordings or third-party platforms, they must be integrated with Sprinklr to enable effective AQM detection. When importing calls, ensure that all relevant metadata is included along with the recordings to support accurate analysis of AQM parameters.
Example
S.No.
Parameter Name
Guidelines
Feasible?
1
Correct Logs
Check if an agent correctly logs the contact driver in an external CRM system outside of Sprinklr.
Infeasible, because the information is stored outside Sprinklr.
2
Correct Logs
Check if an agent correctly updates the contact driver in the custom field within Sprinklr.
Feasible, because the information is stored within the Sprinklr Case Custom field.
Image Text Analysis Is Not Supported
Image Text Analysis is not supported. The capability to analyze text present within images is not currently available.
Example
S.No.
Parameter Name
Guidelines
Feasible?
1
Confidential Info Shared
Check if an agent shared an image containing "acme care" written on it, as it constitutes a privacy violation for the company.
Infeasible, because it requires image analysis.
2
Confidential Info Shared
Check if an agent mentioned "acme care" in the messages sent to the customer, as it constitutes a privacy violation for the company.
Feasible, because it requires text analysis.
Parameters Should Not Be Overlapping
To ensure clarity and avoid confusion, each parameter captures a distinct aspect of customer feedback. Overlapping parameters can result in inconsistent evaluations and reduce the effectiveness of quality analysis.
When significant overlap exists between parameters, either combine them into a single, comprehensive category or establish clear distinctions that define their individual scopes. For example, if two parameters, such as “Negative Customer Experience” (focused on product or service issues) and “Customer Not Happy” (related to team performance), are too similar, consider merging them. If you choose to keep them separate, define their boundaries clearly to prevent misinterpretation.
The example below highlights two parameters with similar guidelines. Because of their overlap, they should not be created as separate parameters; instead, only one unified parameter should be defined.
Example
S.No.
Parameter Name
Guidelines
1
Negative Customer Experience
Check if the customer mentioned having a negative experience with the customer care team, expressing frustrations such as "I’m frustrated with this" or "I am unhappy with you guys".
2
Customer Not Happy
Check if the customer expressed dissatisfaction with the customer care service, using phrases like "this is terrible customer service" or "what a headache to deal with you guys".
SOP-Based Parameters and Flow Charts
For SOP-based parameters, create detailed flow charts to ensure a clear and comprehensive understanding of the process. These parameters typically involve multiple AI models working together to evaluate a single parameter.
Each flow chart maps the complete process, clearly identifies the different AI models involved and specifies the role of each process. The flow chart, defines the guidelines for each AI model to eliminate subjectivity and ensures consistent interpretation and execution.
Example:
(i) Parameter Name: Sales Offer Made.
(ii) Parameter Guideline: Check if the agent provided a sales offer to the customer when the customer was not dissatisfied. If the agent provided a sales offer when the customer was dissatisfied, score them zero.
(iii) Flowchart: Based on the above details, create a flowchart first and then the checklist rule. The flowchart is created taking the example parameters.