Use of AI/ML to Decide Cycle-times is a Futile Exercise
Many companies are dealing with variability of cycle times and changes in the mix of products that make it hard for them to commit to their customers with a more accurate delivery time. To this end, they are depending on AI and ML techniques to estimate what the cycle time should be based on past experience. This is a simplified approach, offered by many supply chain planning vendors, that is not a good strategy for several reasons expressed below. It is really an attempt to provide ATP/CTP capability, that can be readily calculated using deterministic techniques that are more accurate and more reliable.
The way this works is that the ML algorithm learns from previous experience how long it would take to deliver given a customer, quantity, mix pf products and a few other factors. The major shortcoming with this approach is that it relies on past practices to project the future. How would you then improve your cycle times if you are always relying on the past? Furthermore, such an approach does not consider recent disruptions or changes in the mix or certain events such as a large order, arrival of a high priority order, or a drop in volume. The algorithm keeps producing the same cycle times not knowing that a major change has taken place unknown to the algorithm. It will take quite some time to catch up.
Imagining the arrival of a high priority and large order, the algorithm is likely to produce a cycle time that it has learned; not knowing that by re-allocating the orders a much better cycle time and faster delivery date can be given to the end customer. In other words, the approach fails to intelligently perform re-allocation.
Optimization techniques or AI search techniques are much more effective in finding more realistic cycle times and commit dates by re-adjusting the plans based on many different factors and not just relying on “the way we have been doing it.” The optimized cycle times keep improving the performance of the operations, keep finding a better solution and can easily calculate precise cycle time and delivery performance at sub-second speeds for thousands of orders at a time. For more information on how AI can help to perform intelligent allocation and more accurate ATP/CTP, click Here.
