Frost & Sullivan’s latest report “Industry Reshaping Caused by COVID-19” released on September 28, 2020, identified nine major global trends under the impact of the new crown epidemic. The analysis report pointed out that the epidemic prevention work will accelerate the deployment of AI solutions and AI innovation, and will also make AI and machine learning tools an important foundation for accelerating comprehensive digital transformation across key business projects.
But McKinsey Digital has a different view. They believe that the sudden attack of COVID-19 has made people see the fatal weakness of artificial intelligence (AI) and analysis technology.
Although the COVID-19 epidemic has indeed led many companies that have not yet achieved digital operations to actively try AI and analytics applications, Tamim Saleh, senior partner of McKinsey Analytics, said that the predictive ability of analytics technology is dying.
Saleh said, “Although analysis technology has indeed helped many companies to tide over the crisis, it also shows the risk of making mistakes. As a technology that relies on past patterns and behavior details to predict future trends and guide corporate policies, COVID The emergence of -19 has undoubtedly brought a lot of shocks to analysis. This wave of epidemic that has changed our life and work in all aspects is clearly beyond the scope of understanding of analytical techniques.”
No pandemic manual
Saleh also pointed out that “because the data used was collected before the outbreak, a considerable part of the forecasting models have lost the ability to judge — or collapsed because of COVID-19.”
Saleh believes that this is because such models can no longer reflect the world we currently live in. “Like ordinary people like us who don’t have a new crown survival manual to guide the details of life under the epidemic, analysis technology can’t find suitable data to predict what will happen next.”
According to Saleh, the retail industry is a typical example.
“In the retail industry, the analysis model mainly focuses on the actual passenger flow. But after the outbreak, the passenger flow instantly returned to zero. And retailers are not prepared, at least not prepared to transfer all sales online. We Research shows that about 60% of new users of digital channels have decided to use online services now and even after the epidemic is over. This represents a fundamental change in consumer behavior. In this way, when the physical store reopens, the original The algorithms used for forecasting, budgeting and revenue must be outdated.”
Saleh also mentioned, “In order to obtain more accurate prediction results again, the only solution is to redesign the model that is truly relevant to the world we are currently in.”
The new business normal: pragmatism is true
Saleh pointed out, “Nowadays, companies need a pragmatic spirit of survival more than ever. Under the new normal, there are a large number of unknown factors, so the decision-making ability of people will once again become the key to determining the fate of the company. In other words, Since most current models cannot obtain perfect data or historical correlations that help predict future events, they must instead rely on humans to make judgments.”
Saleh believes that companies need to follow the 28th principle, which means completely changing the entire analytical method system.
“The most important task now is not to re-establish teams and projects based on analysis and data, but to re-establish the important position of human judgment. At the beginning of 2020, predictive models and AI judgment programs are already complex enough to provide assistance to many companies . But in the face of unprecedented special circumstances, the human factor has become more and more critical in all aspects.”
Based on the long-term, subversive analysis
In Saleh’s view, multiple methods, including model complexity and feature trade-offs, are expected to bring a full-scale disruption to AI and analysis technology.
“The higher the complexity of the model, the lower its transparency, and it will be difficult for us to determine whether it has good predictive capabilities. Taking theaters as an example, although we can grasp a large amount of data related to ticket bookings, performance prices, etc., It is impossible to deduce how it will operate after four years.”
“But there are always two sides to the problem, and the analysis models themselves are divided into different categories. If your AI and analysis models are mainly used for production quality control, then COVID-19 will hardly affect them.”
In contrast, for AI and analysis models used in industries that are susceptible to human behavior (such as the retail industry), the emergence of COVID-19 has clearly changed its application scenarios.
“This is because the epidemic will eventually affect our lives in many ways, including what products we want to buy, how much money we can spend, and how to respond to advertising messages. Therefore, any model that relies on behavioral analysis will soon Outdated. To truly keep up with the times, we need a more pragmatic spirit.”