Using responsible machine learning to drive customer-centric insights

In this era of digital transformation, most organizations put emphasis on data to make business decisions. But data alone isn’t enough to drive positive outcomes for customers and the business. Analytics solutions – including statistical models, machine learning models, and deep learning models – offer a convenient way to turn data into meaningful business insights.

For organizations to lead from a customer-centric position, they need a comprehensive view of the full customer journey, as well as the ability to obtain granular insight on what is driving customer experience.

A well-defined, data-oriented solution strategy can be broken down into four types of analytics:

1. Descriptive Analytics: Identify what’s happening now and what’s happened in the past. For example, what percentage of customers are unhappy and may churn.
2. Diagnostic Analytics: Diagnose why things happened. For example, what pain points made customers unhappy and caused them to churn.
3. Predictive Analytics: Predict what may happen in the future based on the diagnostic analysis. For example, predict the happiness of a customer and if they will stay or churn.
4. Prescriptive Analytics: Prescribe actions to be taken to affect outcomes. For example, trigger actions that might turn an unhappy customer into a happy one.

This strategy requires proactive signals in order to take action “in the moment” and create relevant experiences for each customer.

According to McKinsey & Company global management consulting firm, half of companies that embrace AI in the coming years can increase efficiencies and create business opportunities that could lead to doubling cash flow. Manufacturing leads all industries in this study due to a heavy reliance on data.

Data about customer interactions is key to predicting satisfaction and behaviors that will allow us to take proactive actions to personalize experiences and improve customer outcomes.

According to the 2020 Gartner AI in Organizations Survey, just 53% of machine learning prototypes are eventually deployed to production. One constant reason is companies do not understand that machine learning is an iterative process, not a one-time development. Without a well-defined strategy or metric that determines success, the project goes in loops. In cases where resources are limited, a “proof of concept” project may be a good option.

The general ML practice includes seven steps:

• Collect the data.
• Prepare the data.
• Choose the model.
• Train the ML model.
• Test and evaluate the model.
• Tune the parameter.
• Predict the outcome.

Several questions should be considered at each iteration. Is this model performing well? Is it stable and learning accurately or is the learning biased? Is the model interpretable and explainable? Common failures include unaccountable black-box mechanisms, poor data quality, and fair model quality.

Some of the pitfalls can be avoided by building simpler glass box models using responsible machine learning principles, a framework put together by the Institute for Ethical AI & Machine Learning.

As the name suggests, a glass box model provides more transparency and clarity based on the principles of human augmentation, bias evaluation, explainability with justification, reproducible operations, displacement strategy, practical accuracy, trust by privacy, and data risk awareness. All these are critically important for machine learning models across most industries.

ML models should be developed in collaboration and with a human-centered design approach at each development and deployment stage. This will enable and empower the data scientist to provide deeper insights to make predictions and identify actions that should be taken.

Responsible Machine Learning practices should be given top consideration to create a holistic view of the satisfaction and value potential of every customer. A more responsible ML will drive action-oriented insights that help shape an effective strategic approach in the era of digital transformation.

This blog is provided for informational purposes only and may require additional research and substantiation by the end user. In addition, the information is provided “as is” without any warranty or condition of any kind, either express or implied. Use of this information is at the end user’s own risk. Lumen does not warrant that the information will meet the end user’s requirements or that the implementation or usage of this information will result in the desired outcome of the end user. ©2022 Lumen Technologies. All Rights Reserved.