COVID-19 has changed our thinking about everything from conducting meetings to optimizing supply chains to predicting revenue. Unpredictable events like this should also change how we think about analytics, and how to use advanced artificial intelligence (AI) to better predict a range of scenarios and adapt to those that come true.
The first article in this series details the concepts of data perishability: the need to assure the currency, accuracy and relevancy of your organization’s data and analytic models in a fast-changing and uncertain world. This article describes agile analytics — the use of “always relevant” data and models to make better and more informed decisions even when much of the information needed is ever-changing, unknown or even unknowable.
With COVID-19, for example, even the best-informed business decision makers struggle to understand:
- True infection and death rates.
- How well lockdowns are working.
- The percentage of the population that will abide by social distancing guidelines.
- Those who are most infectious and why.
- The availability of a vaccine or improved treatments.
- And how all this will impact their business — regionally to globally.
In such an environment, managers should use analytics not as a telescope that, if only it were accurate enough, would let them see the future perfectly. Instead, they should use analytics as a flashlight that, as they scan their surroundings, can help them find their way through a murky, fast-changing environment.
Agile analytics uses advanced AI to continually assess the currency, relevancy and accuracy of data and analytic models, updating both to reflect changing real-world conditions and how useful given data or models were in the past. In agile analytics, advanced AI also continually assesses for model perishability — developing and testing new models, improving them over time, and giving decision makers an estimated confidence level for each prediction.
Achieving agile analytics begins with a deep understanding of the lifecycle of data — which datasets should be used to train models and make business-critical decisions in uncertain times. It also requires using advanced AI to continually chose the most useful data and refine the models used to analyze it to meet ever-changing needs.
Rethinking Analytics
Business analytics has traditionally aimed at reducing uncertainty, avoiding error, and providing a limited set of models of the future that are as accurate as possible but are typically tethered to historical events and interpretations. Armed with these annual or quarterly forecasts, managers would make decisions about everything from building plants to hiring, product development, product pricing and marketing.
Predictive accuracy is still critical, of course, and is one of the areas in which advanced AI can most help. But in adapting analytics for today’s world, it’s important to realize when it will be impossible to create an “accurate enough” forecast, when questions lack a simple answer, and when the best analytics can offer is an indication of a trend rather than a precise forecast.
Achieving agile analytics requires:
Accepting that some forecasts produced by models will be wrong. Rather than wasting time and effort seeking absolutely accurate predictions, businesses must understand the margin of error in each forecast and prepare to respond to such errors. If a hospital is planning added intensive care unit capacity for an expected uptick in COVID-19 cases, for example, knowing whether the surge is 20% or 60% likely helps it decide how much time, effort and money to devote to the additional capacity.
Focusing less on individual forecasts and more on scenarios that produce outcomes with high/middle/ low ranges of impact. Data and models should not only reflect current conditions (what’s up or down) but what's likely to go up or down an hour, two days, or five days from now based on various scenarios. This helps managers prioritize their efforts and get early warnings on needed changes.
Adapting analytics to consider multiple scenarios rather than simple “positive” or “negative” events. Consider, for example, modeling the impact on a business of the availability of a coronavirus vaccine. What are the impacts if the vaccine is only available in one of the markets the organization serves? If it is released but only available to healthcare workers for the first three or six months? If it is only partially effective and governments reimpose lockdowns as a result?
Looking less for firm projections and more for early indications of trends. A sudden spike in social media posts about car-averse urban millennials suddenly buying vehicles for weekend escapes from crowded cities isn’t enough to base a business strategy on. But it does point to a possible change in consumer behavior that businesses ranging from auto dealerships to vacation rentals should seek to understand.