A growing number of companies are using sophisticated predictive models powered with artificial intelligence and/or machine learning to assist in critical decisions.
However, even the most advanced models were unable to predict the coming of the COVID-19 Pandemic or Russia’s invasion in Ukraine. This serves as a constant reminder of the limitations inherent in trying to predict the future.
Predictive Modelling Limits Complex Environments
Predictive models are built on past events and are used to predict future outcomes. However, they must adapt to complex situations and environments.
Events that are unpredictable by nature aren’t preceded or correlated with any other data points. This creates a bias in model output towards predictable and safe outcomes.
Companies that rely on these outcomes for their operations are often caught unaware by unforeseen circumstances, which can lead to disastrous results.
Supply Chain Disruptions
The current supply-chain crisis is a clear example of the dangers facing companies who place too much trust in predictive modelling.
Supply chains, in particular are designed toward predictability and are not resilient.
Two of the most recent disruptions to a global supply chain are the product supply shortages that panicked consumers early in the pandemic, and the current microchip shortage that is threatening manufacturers. They won’t be the first.
Complex Problems abound
The unpredictable is defining the business landscape and the world.
Complex challenges that were once complex are now complex.
Complex challenges are increasing, and the business environment is becoming chaotic.
Companies that rely on predictive modeling to drive decision-making within a chaotic environment need to develop organizational resilience. Models are always reflective of the organizations that they are being used.
A distributed supply chain model can help to increase resilience when it comes to supply chain crises. Distributed supply chains can help reduce the negative effects of disruption in any single place.
To Plan for Complexity, Build Contingency Plans
Although historical data may not be sufficient to predict disruption before it happens, enterprises can still conduct experiments to predict the effects of disruptions and provide better experiences.
These experiments can be used by businesses to create contingency plans that will ensure success and better preparedness for the unexpected.
As complex systems become more complex, so will the volume of data generated daily by modern enterprises.
Organizations will eventually have so many data points that it will be almost impossible to extract actionable value from them without a platform and method that can manage all of them.
Some companies still benefit from predictive modeling. However, it should serve as a stepping stone towards a prescriptive approach to modeling that not only projects but also pinpoints the most appropriate responses.
Humans are able to function in a complex, but linear world because of their evolution. Therefore, when faced with complex problems, we tend not to approach them in a linear manner.
Data can be used to develop new business and technology solutions. We can test our environment with data.
Preparing your model for complexity
Experimentation, rather than preconceived notions allows us to get creative and take on probabilities that are not expected. The more data that we have to draw upon, the more information we can gain about the complexity of our environment and the best way to take action.
Prescriptive models that don’t rely on historical data allow us to see the environment as it is, rather than as it was.
Complexity is revealed when business leaders take the time and test models based on inputs that reflect chaos.
New relationships with data
This approach is not easy. It will usually require organizations to reevaluate how they relate with their data. These are three ways they could do it:
1. Accept the possibility of a failed projection.
When modeling, I am reminded often of the saying, “All models are wrong; some models are useful.” Even the most sophisticated models won’t be able to make accurate predictions due to the fact that they are constantly fed data from the past.
Model outputs should not be considered as concrete evidence of the future, but as indicators of the future.
2. To focus on specific goals, conduct A/B testing
Leaders should test a variety of inputs when using models to inform critical decision making.
A business that is on one path might keep it that way and then test inputs to reflect possible changes in internal processes or any other metric to discover different routes.
The more inputs an organisation can collect, the more information they can gain about their operations’ strengths and weaknesses.
3. Take the modeling results into consideration
Sometimes even testing companies fail to make the right decisions. They often act on preconceived notions and not the data.
The Case of seriously flawed data
Leaders who have invested in a plan that is based on one outcome might be more inclined than others to trust their intuitions. This could indicate that their plan has serious flaws.
They can be patient and continue to gather more data to help their models get a better understanding of the environment and create more innovative plans for navigating it.
Image Credit: Provided By the Author; We are grateful!
Pariveda Solutions Vice President
Pariveda Solutions vice president Sean Beard is a consulting firm that creates innovative, growth-oriented and people-first solutions. Sean is responsible for evaluating the potential uses of emerging technology.