QuantumBlack - A McKinsey Company -

Goal Help a leading APAC Pay TV company transform their approach to churn to counter new threats to its customer base with 3 goals: Drive down churn by setting up a proactive, transformational approach to churn, driven by analytics Build muscle and capability in data and analytics Pilot a new integrated and agile way of working Impact Over 6 months we worked together with the client to setup a proactive, whole of company approach to churn with 5 major impacts: Setup the first high performance cloud environment Created the first integrated dataset for the company, joining 20+ data sources and creating a 33TB asset Developed 2 machine learning models with explainable AI, improving accuracy by 1.4-2.5x Setup an integrated agile squad to execute campaigns from the model Executed over 150 tests to create the largest proactive campaign and experiment in the company’s history

Overview Global PharmaCo aiming to understand from Real World Evidence size and treatment journey of EU patient populations in new indication as well as to predict patients early-on to provide optimal treatment. Context Global PharmaCo seeking to leverage Real World Evidence in EU countries to engage medical stakeholders as well as payers in access discussions and to develop AI capabilities to strengthen future evidence generation practice. Focus was to support a pre-launch asset in new indication where there was limited understanding of the eligible patient population leading to delayed diagnosis and non-optimal treatment pathways for patients Therefore objective was to Size and better characterize patient populations in key European countries Predict patients disease progression and find risk-factors to identify patients earlier on Demonstrate the potential value of optimized treatment Analytic approach and data sources Used anonymized patient–level medical records from different sources to overcome limitations in EU data sets Build consistent research data model to apply same analyses to different data sets and increase replicability of analytics Mapped prevalence and incidence for target patient population on large RWE data sets validating different definitions Developed novel methodology to identify patient events not directly retrievable from medical codes leveraging expert panel Analyze local patients characteristics, journeys, and treatment pathways to quantify overall opportunity towards optimal treatment Applied predictive and explanatory machine learning models to identify patient populations most likely to progress to higher disease severity Performed clustering of patient archetypes and identifies allocated risk-profile. Impact 50% of patients going through sub-optimal treatment journey 5-12 months prediction of patients ahead of time (AUC 0.83) 5-30% opportunity to reduce hospitalizations Insights are expected to support: Market access and payor discussion – with health economic model that can be build based on prevalence estimates Medical – with published results that can be used for medical knowledge dissemination and education of HCPs Commercial – with insights about patient pathways that can be used in messaging to physicians

Focusing on factors with the greatest impact on the manufacturing process helps boost manufacturers’ efficiency while continuing to meet high quality standards. Goal For years, the semiconductor was motivated by Moore’s Law, which posits that the number of transistors in a densely integrated circuit will double every two years. The furious pursuit of semiconductor improvements often presents manufacturers with the dilemma of how to achieve better performance while maintaining product quality. One semiconductor industry company wanted to reduce defects in its manufacturing process—a highly instrumented and complex environment with much room for error—thereby improving its year-on-year production efficiency. Insight and Action QuantumBlack aggregated data from 22 sensors and selected a random forest model to assess the probability of failure based on a number of operating factors available during the live manufacturing process, such as temperature, pressure, voltage, current, and resistance. We identified and quantified the key characteristics of the manufacturing process responsible for thickness-related quality issues, including optimal voltage and temperature controls, lamp power, and throughput. We validated a model that can provide several hours of advanced warning for semiconductor wafers with quality issues, which has helped the company significantly improve its manufacturing operations.