The COVID-19 pandemic has accelerated many companies’ use of advanced analytics and AI. These strategies have helped to engage customers through digital channels, manage fragile and complex supply chains, and support workers through disruption to their work and lives. At the same time, leaders have identified a major weakness in their analytics strategy: the reliance on historical data for algorithmic models. From customer behavior to supply and demand patterns, historical patterns and the assumption of continuity are what give predictive models their power. COVID-19’s impact on how we live and work has challenged those patterns—and the models companies use for making business decisions.
Still, these models have tremendous value. In this McKinsey Live session, partners Roger Burkhardt and Jessica Lamb discussed analytics models and strategies on track for the next normal.
Expand the organization’s data sources
Companies should focus on where they can obtain new insights rather than rely on lagging information. These can come from both from new sources of data and using existing data in new ways. For instance, movement data from cell phones can provide a clearer picture of economic activity by location. In another example, banks that traditionally use credit scores to analyze risk can instead turn to customer-account data, where they might identify gaps in deposits.
Improve decision-making processes
Organizations can’t let their models run on autopilot; to get the most value from them, leaders often need to augment models with human judgment. Analytics leaders must also reassess how we learn from history and weight situations differently based on their context. Simpler models, which are more transparent and explainable, also help. The human-machine hybrid can lead to more analytics opportunities not accounted for with AI alone—for example, in the United States, pandemic management at the state level provides more predictive data as states go through different infection waves and restrictions.
Introduce scenario analyses
Organizations now need to plan for a range of scenarios rather than a single prediction or baseline outcome. For example, when it comes to a functional end to the pandemic, organizations must plan for any number of factors, including vaccine development and distribution, acceptable levels of immunity, and the progression of different therapeutics. All these forces will have an impact on how people live and work, and they are likely to unfold at different speeds and in different geographies. Organizations that model a range of scenarios will be better able to prepare and plan.
Advanced analytics has tremendous benefits for rapid decision making, but organizations need to question their existing models, make adjustments, and remain agile.
Questions and answers from the webinar
- In which industries are you seeing most analytics models applied?
We see a broad base of adoption cross industry as reflected in the figures we shared at the beginning of the session and with a recent acceleration in analytics supporting digital engagement which has seen a step function increase due to COVID-19 restrictions on movement. For one Fortune 100 company, it has seen their digital agenda make a decade’s progress in just 90 days.
We also shared an example about mining, and that’s an industry that is in fact seeing a good adoption rate. There are two things that gate the pace of adoption in mining—implementation of end to end sensors and an operations-centric culture. People need to be comfortable with AI as a way to augment their capabilities even as it changes some traditional ways of making decisions.
In healthcare on the other hand, the industry is poised for rapid adoption. The global pandemic caused a bit of a delay as providers and insurance carriers focused on their response and care. At the same time, it increased the use of digital tools such as digital health and telehealth. We expect that these tools will spur an acceleration of digital and analytics in the healthcare space.
When it comes to fixing analytics models, what are the priorities?
Companies should already have a model governance function that includes systematic monitoring so they can see immediately if model performance or data inputs are drifting. Model governance and a thoughtful triage process based on model criticality are key to prioritizing. The next thing is addressing the issues we’re experiencing by expanding data sources, introducing better scenario planning, and improving decision-making processes. Overall, this is why simpler and more transparent models are important—they are more explainable which means human intervention has better results. And from a strategic perspective, incorporating scenario planning is a priority.
To what extent does this approach to analytics models require increased collaboration?
The wider the data net is, the more likely analytics leaders can come up with something that works and that’s why greater collaboration is critical. We’re seeing three important areas of collaboration. In supply chains, it’s getting faster and better intelligence from all the players in the chain so organizations can make better decisions. There is also greater collaboration even with competitors to understand demand; many organizations understand that there’s value in sharing information for better demand planning. And within an organization, collaboration between the data scientists and analytics leaders and the business and industry experts is critical.
Can previous black swan events help build these predictive models?
Many leaders have learned a lot working through previous crises. Each time, the key in predictive models is understanding whether the behavior changes were permanent and what the new normal looked like. For example, when the MERS pandemic struck, organizations found much greater digital engagement, and that stuck. Organizations will have to monitor key markers, and the level of digital engagement and e-commerce will be one of them.
What are potential risks and considerations when applying AI at an early stage?
Companies need to make sure they understand the assumptions and data that are driving their models and choosing intrinsically transparent models or explainability techniques is key to this. They also need to be thoughtful about where to combine models with complementary human judgment. Last, they’ll also need to root out bias and have an end to end modeling process supported by tools to ensure harmful biases aren’t embedded in the models.
From a talent perspective, organizations need to not only have great data scientists and engineers who can build the models in a thoughtful manner, but build a culture of collaboration. Bringing together the people who build the models and the business experts is critical for tackling the underlying business challenges. Many successful organisations have invested in “translator” roles to bridge the two worlds.
For more on this topic, please watch the webinar recording and read the articles “Leadership’s role in fixing the analytics models that COVID-19 broke” and “When will the COVID-19 pandemic end?”