James Pozzi speaks to Paul Myler, supply chain strategy director for Mars North America, about its adoption of Witness software and how it was employed in its quest for continuous improvement.
Mars stated it required more dynamic software than ones used previously across the company. Which specific areas did you look for in Witness for meeting production demands?
Traditional planning software like JDA or SAP don’t perform dynamic modelling well. They are really linear programs that optimise to constraints, which is adequate for situations where you can buffer with stock. Witness actually simulates how your system will run under different scenarios and provides probabilities of upset conditions. This is very important for chocolate because you have very little buffer. The implication is that an upset condition, instead of eating into your safety stocks, can shut your operations down and cost you a lot of money.
What were some of the key implementation challenges on this particular project?
Our key challenges were in determining the inputs to the model. We had to simulate how our planning software would create a production schedule week-to-week in order to provide realistic inputs to the simulation. It sounds simple, but it’s very difficult. We had to consider questions like: “How often do you schedule X product family?” and “How would your planning software re-sequence and pre-build when hitting a capacity constraint?” We had to think through several constraints, then run a schedule simulation, compare it to real data, and go back to the constraints. Once we obtained a schedule that looked realistic, the back end of the model was easier. Running the model once all that was sorted out is fairly easy, and we could quickly reach recommendations for necessary capacity.
The company runs the model every six months as part of its long-term planning and budgeting process, as a five-year view. Are significant changes commonplace with every half yearly check?
Since we did our homework the first time, and we’ve validated the model over 3-4 cycles, the only changes come when category growth or Mars growth changes over the five-year horizon. The further up the supply chain you go into, say, base stock for films or chocolate, the less overall volatility exists as compared to the volatility on a single item. As a result, total chocolate demands are relatively stable, you just really need to know how much loading you can put on those assets given how your finished goods lines cycle through different product families. A recent example of the model’s success arose when one of our plants reported being very tight on chocolate, although the model had predicted the supply was adequate. Our first reaction was that we made a mistake in the model. So we went back one more time, validated the model and then went to the site and dug into the operational details. We discovered the reason they were short was they were having issues hitting stated rates in the model. Once we focused on that and got the rates back up to standard, we found the output mimicked the model’s predictions.
The Manufacturer is hosting the Future Factory: High Performance Supply Chain event in London on April 29th.
This event will not only explore transparency on a global scale but also how to bring a supply chain back to the UK whilst still ensuring best practices in terms of lean efficiencies; financial imperatives; and environmental benefits. Subscribers to The Manufacturer can attend this event for free. Sign up now for 12 months for just £95.
Due to the positive results of the model, Mars has been able to locate other supply chain opportunities that need a similar modelling approach. What were these?
Similar to other CPG companies, simulation is used when you need probabilities of occurrences given curveballs that are thrown your way. We’ve found simulation to be a good finished goods line level tool to accurately scope the size and amount of buffers, packaging equipment, and the like. These are typically simpler than the chocolate model, as they are contained pieces of the operation. Most supply chain questions can be answered in a linear programming application with some professional judgement applied to the outputs. Where the complexity of probable inputs becomes too complex, or volatility makes it hard to predict how a system will behave, I believe simulation is more suitable.
Do you foresee predictive software becoming increasingly integral to the food industry, particularly in light of automation investment?
Absolutely. Predictive solutions will be more important as senior executives will want to know in real-time the impact of variables proposed within the S&OP cycles. I see real-time predictive analytics in this environment being used in the future for calculating how shorter term changes to promotional events will affect the rest of the supply chain and whether or not desired service levels can be achieved when a decision is made to seize short-term, unplanned demand opportunities. Like any of these solutions, the limiter will not be – and isn’t currently – the software to do it. The limiter is having enough qualified people that can feed these tools relevant inputs to obtain a predictive result that mimics reality. These tools can get very complex, very quickly. I believe that more traditional simulation – like in Witness – is better for medium-to-longer-term large decisions like capital investment and network optimisation.