AI-powered Smart Pairing leverages available customer data for better mapping of unassigned cases to appropriate agents. Sprinklr’s Smart Pairing feature is designed to connect your customers with care agents that are best suited for them leading to optimal outcomes with customers' queries getting resolved quickly and overall improved care KPIs performance.

Smart Pairing uses data from past interactions to dynamically match customers to agents using proficiency in solving the incoming intent, perceived service quality, and other strategies leveraging behavioral, demographic, geographic, and psychographic data (whenever available).

Problem Statement

In standard pairing strategies, the decision about which agent will take care of an inbound message, mainly rely on the agent's capacity, availability, priority, and basic skills (associated via Sprinklr Assignment Engine/Unified Routing). However, these strategies are suboptimal because

• Skills are manually assigned and usually not frequently updated.

• Performances of agents on other important customer perceived aspects, e.g., communication proficiency, patience, etc., are not taken into account.

• Demographic, behavioral, psychographic data about agents and customers, if available, is not taken into account.

How To Use Smart Pairing

• Click the New Tab icon. Under Sprinklr Service, click Unified Routing within Route

• On the Queues window, click Add Work Queue in the top right corner.

• On the General Settings window, enable Use Smart Pairing and select Cases to be routed using Smart Pairing, for example, selecting 70 will apply the smart pairing to 70 percent of all cases coming to this queue.

​

​How Does Smart Pairing Work?

Step 1 - Scoring

• For each closed case and involved agent, intent resolution proficiency is automatically calculated on the basis of the detected intent and the most suitable proficiency metric available for the partner’s data. Examples of proficiency metrics are: numerical survey scores assigned by customers to the agent for closed, same-intent cases; predictive CSAT; customer sentiment intensity detected in the portion of closed, same-intent cases handled by that specific agent; any other relevant custom metric available or a combination of the above. The exact metric(s) and combination strategy used for your brand will be decided based on a case-by-case preliminary assessment.

• For example, if for a brand, we measure a high correlation between Predictive CSAT and numerical survey scores assigned by customers (which are an objective measure of agent proficiency), we could decide to use both metrics as a good indicator of agent performance. Predictive CSAT could be, in this case, used to boost numerical survey scores assigned by customers - given that the latter might be missing for some cases.

• The agent is also scored against typical perceived Service Quality facets such as Access, Accountability, Accuracy, Communication, Courtesy, Efficacy/Effectiveness, Efficiency, Flexibility, Knowledge, Patience, Promptness, Interest, General positive experience, General negative experience, on the basis of mentions found in cases messages (“You’re very slow!”) and case reviews (“The agent was very slow”) - if available.

• These intent-based agent scores can be conveniently displayed in reporting. This feature provides a clear overview of agent proficiency in handling various intents, facilitating well-informed decision-making and ultimately improving customer interactions.

  Moreover, this data-driven approach offers a valuable use case for performance tracking. By observing the performance trend of agents in intent resolution, you can identify strengths and areas for improvement, enabling targeted training and fostering continuous growth within your team.

  To get this capability enabled, please raise a support ticket at tickets@sprinklr.com.

  

  In reporting, you also have the ability to explore the primary intent of an interaction to gain insights into the key contact drivers at the interaction level. This can be achieved by utilizing the "Case Detected Primary Intent" dimension.

  It's important to note:

  1. The initial intent detected in the case, determined by the message creation time, will be designated as the primary intent.

  2. If an intent with a higher priority, as set during the intent model configuration, is detected later in the interaction, it will supersede the existing primary intent.

  3. In cases where two intents have equal priority, the primary intent will be determined based on the message creation time.

  ​

• Virtual scorecards are periodically updated.

​​

​

Step 2 - Pairing

• For each new case, an agent is assigned on the basis of how proficient they are at solving the incoming detected intent (see example below) and/or Service Quality scores, instead of relying on standard pairing strategies. This implies that the case will go to the agent that is best matched regardless of whether another agent was available first. If behavioral, demographic, geographic, and psychographic data is available, custom strategies leveraging this information can be evaluated. Anti-bias mechanisms are in place to ensure fair agent treatment.

​​

​

Step 3 - Managing

• Variations in Care KPIs can be assessed through an ad-hoc reporting dashboard.

• Optimization for overloaded work queues: A typical Care scenario consists of several cases that accumulate into a large backlog. Traditionally, such a scenario is handled by simply assigning the first available agent to the oldest waiting case. To ensure the correct behavior of Smart Pairing in such scenarios, the algorithm first considers the subset of cases in which the available agent would perform the best as per the model (score); then, out of this set, the algorithm selects the oldest case. To prevent any case from experiencing unacceptable latency, a maximum wait time can be configured after which a case would be considered with the highest priority (overriding the Smart Pairing logic).