Predicting and Preventing Unhappy Customers Using AI

Explore a machine learning product that's improving customer experience inside Dell Technologies.

Behind the scenes, teams at Dell have spent the last two years building a powerful ecosystem of tools for management of consumer (CSG) tech support cases. In this ecosystem, agent workflows are supported by data science products, including a machine learning model which predicts the cases most likely to result in dissatisfied customers (DSATs). This “DSAT Predictor” has helped managing the customer experience become a more proactive process, by increasing visibility of open cases that could benefit from additional intervention by an agent, often before an acute problem has occurred.

The “DSAT Predictor” works by using data about the case, agent, product and customer to flag open tech support cases which are most likely to result in dissatisfied customer responses on a customer experience survey. Cases identified by the machine learning model are displayed to agents as a flag in a dashboard for case tracking and management. This allows tech support agents or managers to intervene earlier on potential problems, thereby improving customers’ tech support experiences.

The project’s benefits go beyond early intervention. It has also led to process standardization without sacrificing regional-specific nuances around customer preferences or expectations. The first attempts at DSAT prediction emerged organically from individual business segments. Over time, there grew to be six different predictive tools, each with its own inputs, algorithm, metrics of evaluation, architecture and deployment environment. In those early days, alignment and standardization across the business was less than ideal. As data science capabilities matured at Dell, there was an opportunity to increase alignment and standardization through a single machine learning approach.

The unification of the DSAT Predictor to a singular, stable model enabled a mirrored unification of case management processes. When the entire organization relies on the same metrics and methodologies, leadership can more easily steer the direction of the organization. Standardization can be overemphasized, however, costing the organization flexibility to regional differences. The case management organization avoids this by tailoring the DSAT Predictor to each region, allowing for the standardized predictive model and case management process to mold to regional cultural norms and business expectations.

Standardized processes and data science together allow integration of new insights and automation to continue to improve agent workflow and customer experience. Tim Lee, a process transformation consultant, noticed that the ecosystem of machine learning models has “made the business more agile” by allowing processes to be implemented organization-wide, but maintaining regional nuance through tailoring the models to reflect regional, site, or country-level differences.

Although standardization has been core to the DSAT Predictor, it works not by reducing agent autonomy but by providing an additional facet of information on which to base case management decisions. The services tech support agents provide our customers continue to be core to the business, so the case management ecosystem began with focusing on a human-in-the-loop “augmented intelligence” framework. This ensures that both agent expertise and machine learning are layered into the system to provide a warm-touch but data-driven experience.

Patrick Shaffer, Process Transformation Consultant and a key project stakeholder, describes the DSAT Predictor as “reducing guesswork and the experiential-based decisions and providing a probability of a DSAT based on years’ worth of historic data and understanding of the business.” That reduction in guesswork allows agents to focus on resolving cases more quickly and providing timely intervention when necessary.

Data science and machine learning are often championed as tools for automation but aiming to delight customers means offering both assisted and unassisted options to diagnose and solve tech support problems. Rather than reducing headcount, the objective is to provide agents the tools to facilitate their work. Optimized troubleshooting flows, next-best-action recommendations, and data-driven insights all play a role. Any additional time agents gain through greater efficiency can be spent on cases where they can have a positive impact on the outcome, such as cases where we know the customer is dissatisfied. Lee pointed out that agents are working diligently to manage multiple cases, often with exceptional attention to detail, but the “DSAT Predictor gives them a chance to bring their heads up and points out when historical trends indicate we may have missed something.”

Since the initial unification of the DSAT Predictor, iterative improvements have continued to fine-tune and improve the product. Model inputs have been aligned and standardized across the business, resulting in an 85% consolidation of inputs compared to the six precursor models. The deployment architecture and algorithm evolved to afford over 20% better model performance, with some business segments experiencing even greater improvements. And when cases are appropriately intervened upon, there is an estimated 15% reduction in dissatisfied customers. Today, the DSAT Predictor runs on more than 1.5 million service requests per year and covers all regions and business segments in the Consumer part of Tech Support Operations.

The DSAT Predictor has measurably improved case management, but Patrick Shaffer sees the current system as just the beginning of data science in case management. He points out that the DSAT Predictor “was highly focused on one facet of decision-making: making the customer experience better,” but there are numerous other challenges that are suitable for a data science solution. He aims to focus on “making agents’ lives easier” and “any time we have a decision point that needs to be made in a workflow, that’s where data science could fit in.”

The data-driven appetite for innovation is echoed across Services with a growing portfolio of AI/ML products to support customer experience. To learn more about other recent Dell Technologies AI and data science efforts, please visit this article.

Rachel Meade

About the Author: Rachel Meade

Rachel Meade is a Senior Data Scientist in the Services Operations Applied Science organization. She develops data science and machine learning solutions with a focus on improving Customer Experience and case management in Services. She is a technical innovator across the full data science spectrum including data wrangling, experiment design, building production-ready statistical and machine learning models, and driving adoption by translating data insights into business narratives. Her passion is in leveraging Natural Language Processing and Deep Learning algorithms to unlock the value of the written word for analytics. Rachel’s Data Science career has flourished at Dell, but began with Favor Delivery in Austin, Texas after transitioning from a finance career in commercial banking and commercial real estate development. Her first career, however, was as a professional ballet dancer, working with internationally renowned companies in the United States and Europe. Rachel holds a M.S. in Business Analytics from the University of Texas in Austin and a B.B.A. in Finance from the University of Houston.