How will technology like machine learning change our approach to supply chain management – in mill products?
I recently read the execellent compilation of 19 machine learning and supply chain experts’ predictions on the future of supply chain management, and naturally wondered how much of that applies to our type of industries. Take for example the steel industry: this is no “simple” assembly and clearly defined SKUs. Instead we deal with a complex multi-step manufacturing process with process-inherent deviations and a highly flexible product definition.
Can we predict demand based on big data and machine learning?
Let’s take a look into the building materials industry for a second (and many metals companies supply into this industry). Macro-economic trends, legislation & subsidies, income and weather data – all of this localized – influence demand. Contract consumption & similar B2B-type forecast can add to the picture. Hailey McKeefry sums it up nicely in the above mentioned ebook: Social media and newsfeeds can give early warning on events that are likely to impact the supply chain: from M&A activities, factory fires, weather.
Machine learning can first of all automate menial and repetitive tasks. It can give real transparency across unstructured data, to give people, supply chain professionals, the context to take better decisions.
Alternative approaches: simulate, sense and decouple
Even with ML the future is uncertain. Machine learning can help you to identify scenarios that you want to be prepared for. But you can very well simulate the supply chain and financial impact of such scenarios in sales & operations planning, and be prepared for eventualities. T
Demand sensing like in consumer products including POS data will not get us far in B2B-networks typical in mill products. Several companies in the paper and packaging industry utilized IoT-sensors at their customers to measure “real” consumption. Which grades, which rolls have been consumed and will likely need to be replenished? Packaging companies running own packaging machines within their customers plant have this information readily available.
The demand-driven approach acknowledges that not all demand can be predicted. Instead it uses the concept to decouple demand and introduce strategic buffers. Machine learning may help to determine the best decoupling points, and grade combinations to buffer. SAP S/4 HANA 1709 and SAP Integrated Business Planning support the demand-driven approach. (Find more information on DDMRP for SAP Integrated Business Planning here.
Predicting supply problems
Many metals and paper companies have used ML and predictive analytic in the manufacturing and plant maintenance domain. Predictive maintenance and predictive quality play a key role to stabilize the supply side.
Predictive maintenance avoids unplanned downtimes, and thus supply delays. Predictive quality avoids quality deviations, and lower than planned yield quantities – again resulting in supply delays or underdelivery. More stable quality directly results in a more stable supply side.
During a recent plant visit, we discussed material shortages of a different kind. Electric arc furnaces require special graphite cathodes. The prices of such cathodes have risen dramatically and there is a shortage due to environmental legislation regarding the produce of the cathodes in China.
Did you predict this?
Similarly, many EAFs require scrap metal as input material. Depending on where you source material, and on the price, the EAF may no longer be profitable to operate. In my opinion, this is a good use case to apply scenario-based simulations.