Production Planning Software and Machine Learning Integration
The tech sector and its smaller cousin the Software-as-a-Service (SaaS) sector are full of buzzwords. More often than not these buzzwords promise improvements across the board and just as often leave investors equally disappointed as what was promised never materializes. Machine learning and Artificial Intelligence (AI) are often seen as the new buzzwords but at least have seen several positive applications found in supply chain planning and demand planning software. What then of production planning?
More difficult? Perhaps…
Looking to integrate machine learning principles and technologies into production planning software may be far more difficult than with demand planning and supply chain planning. This problem stems from both architectural and cultural factors involved in the development of software for production planning. Any great piece of production planning software not only needs to plan daily production in conjunction with a factory and staff but also divvy production tasks across the staff and sometimes across multiple factories or departments. This adds a level of complexity to any successful integration of machine learning technology that is not always easily grasped.
One of the mantras of machine learning is continuous improvement and with demand planning, that mantra can be implemented. This is due to the machine learning engine first looking if the forecast was better than the one previously and then how to improve forecasts moving forward. Through this oversimplification, the forecast will see continuous improvement over the number of iterative forecasts.
With supply chain planning, machine learning can serve to monitor scheduling parameters that humans are naturally not very good at. By monitoring several parameters related to scheduling the technology can in time learn when the parameters need to be changed to reduce lead times, for example. However, with production planning, the data needed for continuous improvement can be scattered across a multitude of other systems and incompatible formats. This would then mean that people would have to be responsible for entering clean and usable data, which in turn can result in the system falling to pieces if employees don’t have the right motivation in ensuring the data is clean and usable.
Rather than relying on employees then a piece of middleware would need to be developed that allows the machine learning engine access to clean and importantly current data to develop systems that continuously improve. This does not mean that the task is impossible and can be done. Such pieces of software are available and have been implemented.
Rather, it means that for those developing production planning solutions with an eye on implementing machine learning engines, all the buzzwords marketers love to use need to be ignored. Replacing those buzz words with an honest approach as to the current technologies limitations will help lead to a more robust system where continuous improvement is achieved.