During the first containment last March, RN systems in the supply chain convinced a large proportion of managers despite a lack of flexibility in their use.
Supply chain professionals are optimistic about the potential of artificial intelligence in their operations, but they also struggled with this technology during the coronavirus pandemic, according to a survey by Secondmind which develops machine learning applications for companies.
The survey (involving more than 500 managers and planners of supply chains using AI) showed that 90% of respondents believed that AI would make a difference in their supply chains by 2025, although 82% were disappointed with decisions made based on AI during the pandemic.
This discrepancy highlights the potential barriers to AI, while emphasizing that professionals who have faced these challenges still see the future of the technology.
AI in the supply chain: what does it mean?
AI is a general term that can include many statistical or computer technologies.
Gary Brotman, Vice President of Products and Marketing at SecondmindThe site, considers AI to be a term for the processes that allow a computer to do what a person traditionally does, as relayed by the site. supplychaindive.com. In general, this includes methods such as machine learning and deep learning. "Machine learning is the place where the action takes place.", he said during the investigation.
"The way AI is used in the supply chain can vary quite widely"says Larry Snyder, director of the Institute for Data, Intelligent Systems and Informatics at Lehigh University and a senior researcher at Opex Analytics. "When people from Opex or similar companies say "AI in SC", I think they usually mean more about teams of people building R or Python models and then deploying them using arrays or other tools.", he says finally.
Regardless of the artificial intelligence technique used, users see potential. "There is a certain level of usefulness in automation and the promise that it will improve over time."Brotman adds. And Mr. Snyder added: "Machine training can make significant progress in providing companies with inventory recommendations...". Many companies have begun to use clustering techniques to improve their internal operations.
Human Power in AI
Many AI applications rely on the interpretation of historical data. And this is not the first time an industry insider survey has identified problems with this method. In June, a survey showed that 51% of shippers struggle with a lack of clarity regarding consumer demand.
The survey Secondmind has shown that 59% of respondents believe that artificial intelligence is faster than humans, and 72% believe that it is more accurate when it comes to making predictions. But nearly a quarter of respondents (23%) were disappointed that they could not apply their own experience in the field to make a prediction. "There must be a way to take what the system can provide and maintain in terms of forecasting and then let the expert contribute to that final decision as well.", Judge M. Brotman.
The accompanying report notes that people need artificial intelligence, but they should also be part of the decision-making process.
Mr. Snyder makes an analogy with the spreadsheet, which calculates demand forecasts. If something changes in the business model, the person has to adjust the formulas in the spreadsheet. "The same is true for AI tools. They are only tools - non-sensitive beings - and when they cease to be the right tool for the job, they must be corrected or replaced to work better.", he analyzes. But it's not easy to make these changes in real time. Four out of ten respondents said existing processes are too rigid to adapt quickly to market conditions.
Human Intervention Essential to AI
"AI has promising capabilities, but it is only a set of tools.said Mr Snyder. It's not a magical black box that can do everything we want, with little training or human intervention. Most AI tools are fairly narrowly focused, and it will always take intelligent, knowledgeable people to use them, and especially to create them."
A paper released last month examined many forecasting techniques, including time series analysis, machine learning and in-depth study to help understand demand during the pandemic. Researchers called for a flexible forecasting approach.
"In addition to preparing for fluctuations in demand, particularly due to bottlenecks, our results show that the forecasting approach must be continually adapted to changing needs."The researchers have a unified voice in recommending this. "This also implies a change in forecasting models."