Intelligent Pairing for Efficient Call Handling

In the digital age, most companies have established inbound customer service teams that operate through chat platforms or phone calls. As businesses increasingly shift toward remote interactions, ensuring efficient and high-quality customer support has become a key priority. However, despite advancements in technology, many organizations still rely on traditional call routing methods that do not leverage intelligent pairing systems to optimize service delivery.

Studies indicate that approximately 95% of customers are routed randomly by agents. This approach can lead to operational inefficiencies, longer waiting times, and lower first-contact resolution rates, ultimately resulting in reduced customer satisfaction. Research shows that implementing an intelligent routing system – considering factors such as agent expertise, request complexity, and customer profile – can bring significant benefits. These include better distribution of workload, reduced idle time among employees, increased first-contact resolution rates, and, consequently, an improved customer experience.

Recognizing the growing need for an efficient resource allocation, a leading telco company approached LTPlabs to harness our expertise in AI and data-driven modeling. The objective was to design a solution that would not only streamline the call allocation process, but also ensure a smarter and more data-driven system for pairing customers with the most suitable operators. Through this collaboration, we aimed to create a cutting-edge decision model that drives operational efficiency while enhancing customer experience.

The telecommunications company envisioned implementing this intelligent call routing model across two critical teams: the technical support team, responsible for addressing technical issues, and the customer service team, which handles general inquiries. These teams are tasked with managing a substantial volume of calls – more than 100k monthly calls in each of the teams. For the technical team, the focus was on minimizing technical interventions, by ensuring that call centers operators only scheduled on-site visits when absolutely necessary. In the customer service team, the objective was to streamline the call handling processes, reducing the average call time.

Before developing a tailored methodology, it was crucial to validate two key assumptions essential to the project’s success:

1. Assessment of performance asymmetries between call center operators:

The first step involved verifying whether operators exhibited varying levels of performance, particularly in terms of average call-handling time (within the customer service team) and the scheduling rate of technical interventions at the client site (within the technical support team).

This hypothesis was confirmed at a general level, and further analysis revealed that such performance asymmetries were even more pronounced when dealing with more demanding clients. Consequently, it became evident that assigning high-performing operators to more complex client profiles should be a core principle of the smart pairing strategy.

2. Existence of arbitrage opportunities within the service process:

The second key premise involved assessing whether there was room for tactical decision-making within the service operations of the two teams analyzed. For the proposed model to generate a tangible impact, it was essential to confirm that, on average, more than one operator is frequently available at the time a client starts a call.

This availability creates the necessary conditions for the smart pairing mechanism to function effectively. In contrast, when only one operator is available at the time of contact, there is no room for optimization.

Having validated these two essential premises, a robust approach was developed around four key streams: a clustering approach, a predictive model, an optimization model, and a digital twin for scenario testing and simulation.

Given the company’s portfolio of over 2 million clients, the initial step focused on the implementation of a clustering model to manage complexity effectively. Creating an individualized, ranked list of operators for each client would have added an undesired level of complexity to the problem. Instead, clients were grouped into clusters based on their profiles, contextual variables, and historical call behavior. This algorithm allowed the segmentation of clients into dozens of distinct groups, each composed of individuals with similar characteristics and consumer profiles.

The second stream focused on the development of a predictive model aimed at estimating the expected call duration and the likelihood of a call resulting in a technical intervention, for each combination of client cluster and operator. This model was built using machine learning techniques.

The third stream – the optimization model – was developed with the objective of generating ranked and optimized lists of operators, based on the outcomes of the predictive model. This optimization primarily aims to ensure that top-performing operators are prioritized in

the rankings for more complex client clusters, while also maintaining a balanced distribution of ranking positions among all operators. In other words, if an operators is placed at the top of the ranking for one cluster, he/she should be positioned lower in the ranking for another. Taking these constraints into consideration, the model delivers structured rankings that drive the greatest possible impact for the business.

Finally, a digital twin was developed to simulate the proposed approach, assess its potential, and explore alternative scenarios. This virtual replica of the operation enabled the team to test different strategies and measure their impact without disrupting the real process. The model was initially validated by replicating the current process and comparing key performance metrics. Following validation, it served as a tool to evaluate the proposed solution and its projected outcomes.

The deployment of our smart allocation system yielded concrete and impactful results across both service lines: customer service and technical support.

In the customer service line, we observed a 6% reduction in average call duration, without a relevant increase in customer wait times. This result underscores the effectiveness of our allocation strategy in enhancing operational efficiency, while preserving a high standard of customer experience. By pairing clients with operators better suited to address their needs, we streamlined interactions and reduced the time required to resolve general inquiries.

In the technical support line - focused on resolving complex issues that frequently require on-site interventions - we achieved a 5,5% reduction in dispatch rates. This led to significant cost savings, given the high operational expense of each client visit.

‍