How to Fail with AI in Your Supply Chain Planning

To prevent failure, leadership needs to consult with supply chain leaders and planners to determine the root causes of the issues that can be solved with technology’s help and choose the right approach for their organization.

Pexels Typing On Computer

Across industries, business leaders are looking to invest in artificial intelligence (AI) and machine learning (ML) technologies to mature their businesses. One lesson of the pandemic has been that digital maturity better positions companies to navigate disruptions. Supply chain leaders realized they cannot delay digital transformation because old ways and legacy systems leave planners mired in mundane tasks without time to focus on more strategic, business-critical functions essential in a disruption.

Where has AI ranked on the digital transformation agenda to date? In the 2020 MHI Annual Industry Report, only 12% of supply chain professionals reported using AI, although 60% responded that their organizations expect to leverage the technology in the next five years. So, while adoption has not yet been high, Gartner has named AI one of its Top 8 supply chain trends for 2020 and predicts that 75% of organizations will move from piloting to operationalizing AI by 2024. As supply chain leaders look to this future, how can they set up their organizations to succeed in this investment?

AI is a broad but powerful set of capabilities with significant promise for a host of common supply chain planning issues. But, to prevent failure, leadership needs to consult with supply chain leaders and planners to determine the root causes of the issues that can be solved with technology’s help and choose the right approach for their organization.

Identify business problems first, then choose an AI or ML solution

For the companies that have already invested in these solutions, many have paved the path to failure. In an MIT Sloan Management Review and BCG survey, 65% of executives reported that they have not seen value from AI investments they have made in recent years, and 40% of companies making “significant investments” in AI do not report business gains from these investments. For the companies that did invest in these solutions, why haven’t they seen the desired results?

Here are four ways to head down a path to failure, along with suggestions to avoid charting that course.

Start with the technology. One reason why business leaders don’t see success right away is because they are not clear on what problem they are trying to solve before choosing the best approach to address the problem. Business leaders may choose AI or ML technologies thinking that they will be the silver-bullet solutions to solving myriad business problems. They may even feel pressure from senior leadership to pursue AI or ML for fear of missing out on “what everyone else” is doing. However, starting with technology without first determining the business problems it needs to solve sets AI or ML projects up to fail. Business leaders should work with supply chain teams to frame the problems first and then choose the right tool and math fit to solve those problems. Advanced analytics is a deep and broad set of tools, so organizations should not limit themselves to AI, just because it’s trendy, if a tried-and-true approach will work even better.

Prioritize accuracy over interpretability. ML models excite data scientists more than traditional models for valid reasons, particularly for their significant improvements in predictive power and model accuracy, but these gains typically come at the cost of interpretability. These powerful models can make it harder for planners to be able to extract the information they need to make business-critical decisions, because they are black boxes that do not explain why the models generated the results they did. Planners will not use what they cannot interpret, so to prevent failure, business leaders should choose solutions that build interpretability into their design, enabling planners to benefit from better accuracy by providing them a way to understand the drivers. Research advancements have made interpretability features a viable option, so organizations don’t have to be forced into this tradeoff.

Turn the lights out and send the planners home. Computational advancements, fancy math and automation can make a fully autonomous supply chain seem like a good idea. Give a computer large volumes of historical data and train it to find the patterns of response. Then, apply those patterns to future data, automate the response and turn the lights out as the door closes. But these models don’t know what to do when they face a disruptively new pattern, and there wasn’t data available that could have fed the models that matches the current state of the world, so the models are now failing.

Fortunately, the planners have not failed, because while advancements have been made in artificial intelligence, computers have not yet achieved artificial general intelligence, which is what allows them to truly think like people. Humans still excel at capabilities like understanding cause and effect and deriving information from context, which are exactly the kinds of skills needed in a disruption. Supply chain planners around the world have shown agility and responsiveness in keeping the lights on and the supply chains running using their very human strengths. A path to failure is to expect too much from the math and too little from the humans, so business leaders should focus on solutions that leverage the best of both. AI can beat a human any day at finding patterns in massive amounts of data, so the best combination is to automate the mundane and augment the human’s intelligence, not replace it.

Invest in data wrangling. Data scientists spend as much as 80% of their time preparing data for analysis, a stage that requires so much effort and transformation that the metaphor from herding recalcitrant livestock came into being. Data is at the heart of machine learning, so insufficient or poor-quality data can lead to failure, but so can investment in external professional services to do the work instead. They won’t know an organization’s data like the planners do, and over-reliance on third-party services can lead to latency in results while businesses wait for them to do their work. Building internal capacity and putting the planners’ needs at the center of choosing a solution is crucial to making the right decision, because unless the company employs data scientists to analyze the supply chain data, matching supply and demand still falls under the domain of the planner. Therefore, they need to be able to easily use tools that don’t require them to know all the math and programming intricacies. Instead of investing in services, organizations must invest in a solution with data fusion techniques to wrangle the data and automate machine learning, so planners can decide on which signals might add value to supply chain forecasts instead of learning to code in Python to ingest the data to do so.

AI and ML provide myriad benefits for supply chain planners and organizations at large, but only if the right tools are deployed for the corresponding issues. By identifying the core business problems, prioritizing interpretability, combining human judgement with AI/ML and investing in solutions with data fusion capabilities, supply chain planners will be able to focus on more business-critical decisions, and business leaders will begin to see positive results from their investment.