One can make very precise measurements (using seconds, millimeters, and micrograms) for operating a machine tool making engine parts. And one can make precise measurements (using minutes, meters, and grams) for operating a facility such as a factory or a warehouse. But what level of precision is realistic when measuring operations across an entire supply chain with many different kinds of vehicles moving scores of products between facilities spread out across a region or around the world?
Is it possible to measure the exact weight of every product in a supply chain? What if the products are delivered in shipping containers loaded with a similar mix of items, but not the exact same mix of items in every container? Is it possible to measure the precise length of a delivery route? When people talk about the distance between two facilities do they mean the distance from one facility’s loading dock to the other facility’s loading dock, or is it the distance from the front door of one facility to the other? Is it realistic to attempt to measure supply chain operations in minutes, meters and grams, or is it better to go with hours, kilometers, and kilograms (or maybe just days and tons)?
Useful Levels of Precision in Supply Chain Models
SCM Globe measures route distances based on how closely the calculated route lines actually follow real roads. The screenshot below shows blue route lines following the real roads. They follow closely — but not exactly (we do the best we can). Even two actual vehicles traveling on these roads would record slightly different route distances due to things such as swerving within their lanes or the turning radius of the turns they make. Potential distance errors from close, but not exact, route lines can add up to a kilometer or two on a short route, and much more on a longer route, especially if there are lots of twists and turns on the route. So route distances are always approximations.
Vehicle speeds are also approximations. When a truck departs, will it always travel at exactly 90 kilometers per hour? What if there is a traffic jam or the truck has engine problems? What about the time it takes to drop off or pick up products at facilities on the route? Specifying exact speeds down to the last kilometer per hour is not possible in the real world. So within a supply chain simulation truck speeds represent an estimated average speed over the route that includes the factors above. And for a ship on a journey across the ocean it’s speed too can only be an estimated average that includes many factors from wind and waves to storms and mechanical problems.
When setting the delay between departures for vehicles what does it mean for a truck to leave a factory at precisely 20.05 hours after it returned from its last delivery? Is that realistic? Will there be a manager with a stopwatch on the loading dock dispatching trucks at precisely the right moment? Setting the delay at 20 hours is much more realistic. Supply chain models should be defined at levels of detail that correspond to what is possible in the real world.
The pace of operations within individual facilities in a supply chain such as factories and warehouses are largely under the control of people operating those facilities. Managers can set speeds of assembly lines and define tight sequences of robotic operations, and maintain those speeds and sequences hour after hour, day after day. In contrast to this, the pace of operations in supply chains that exist out in the world, beyond the confines of individual facilities, are subject to impacts from hundreds of sources, many of which are hard to predict and often impossible to control. Realistic supply chain models must take this into account.
Define products at shipping case or pallet load level, not at the individual item level. Attempting to track millions of individual items through global supply chains is theoretically possible, yet the apparent accuracy that results is neither realistic nor useful. Unless you are modeling a small part of a supply chain at a detailed level there is no need to track individual items such as individual bottles of water, or individual shirts, or laptops or candy bars etc. These items ship from factories to retailers in larger cases – quantities of 100, 250, 500 or more. So define and track products at that level. For instance, if Product A is typically shipped in cases of 100 items, then define Product A using the price, weight and volume data for a 100 item case, not an individual item. Set demand, on-hand, and delivery amounts at the case level.
In real supply chains if a store forecasts selling 268 items of Product A, the inventory manager will round that number to the nearest case sized quantity. The same is also true when calculating the Economic Order Quantity (EOQ). If the EOQ equation says the best re-order size for a product is 173 but the product ships in cases of 100 the store will round that number up or down to equal the nearest number of cases.
Use measurements to define your supply chain entities in a realistic manner. Round your numbers to one or two decimal places. There is no need to use measurements that go beyond two decimal places, and often one decimal place or whole numbers with no decimal places will be quite adequate.
Margins of Error in the Simulations
Small differences in measurements one way or another are not significant because they are within the margin of error of the supply chain model. Good opinion polls and research surveys operate with acceptable margins of error defined as 2 – 4 percent with a 95 percent confidence interval. This is also true of good supply chain models and simulations. Results shown by simulations are accurate within a margin of error. That margin of error is mostly dependent on the time span of a simulation. The longer the time span of the simulation the greater the margin of error will be.
Simulations are predictions of future performance; they forecast what will happen in a given situation under certain conditions. And regardless of how much data you use, or how much analysis you do, predicting the future is always an estimate!
Simulation results can vary 2- 4 percent and sometimes more depending on differences in the data used to define the products, facilities, vehicles and routes of a given supply chain. It is important to remember that no simulation result is ever more than a snapshot in time because over time, as the world unfolds and the data changes, so too will the simulation results. So supply chain simulations should really be run continuously as conditions change and new data becomes available. Just as we continuously run weather forecasts as conditions change.
Weather forecasts are created by running simulations based on weather models. If a 5-day forecast made on Sunday predicts 2 inches of rain on Wednesday afternoon, but it actually rains 2.4 inches on Wednesday evening, does this mean the weather model is wrong? No, it does not.
It means that with the data available when the forecast was made (several days in advance), that was the best estimate that could be made. As more current data is fed into the model and the days go by, the forecast results change with changes in the input data. In forecasts and simulations this variability in predictions created by small differences in input data is known as the “Butterfly Effect”. This is illustrated in the picture below showing the predicted track of a hurricane. In the first few days the predictions are tightly packed, but as the days increase the individual predictions become more dispersed.
Regardless of the Butterfly Effect, weather forecasts show what the weather is most likely to be over a period of time; and short-term forecasts are always more accurate than long-term forecasts. A 24-hour forecast is more accurate than a 5-day forecast.
SCM Globe simulations are typically focused on 15, 30, and 60-day periods to achieve the accuracy of shorter term forecasts. Despite slight variations in the input data used to define supply chain models, simulations show which designs work best in any given situation. They show whether trucks or railroads or airplanes deliver the lowest costs. They show the best locations for different facilities, and they show how inventory flows through those facilities. They also identify facilities where problems are most likely to occur, and provide data for deciding how to fix these problems.
If you use the simulation results to keep adjusting your supply chain model so as to minimize cost and inventory while always meeting product demand, then your supply chain model, and all other models in a given situation, will converge on one or two best solutions (known as “attractors” – see more about this in the “Butterfly Effect”). Small differences between the models do not make a significant impact on their overall performance. Small differences in operating costs and inventory levels shown in the simulations are not significant if they fall within the margin of error for the simulation.
Supply chain models producing simulation results which are within the simulation margin of error are equally good. For instance, suppose a simulation based on model A creates a total on-hand inventory level of 350,356, and a simulation based on model B creates a total on-hand level of 350,489. The difference between these simulation results is about one percent and that is within the simulation margin of error. So both supply chain models are equally good in this area. However one supply chain model may achieve this inventory result at a significantly lower operating cost, or have other advantages and for that reason it would be the best model overall.
Realistic Levels of Supply Chain Optimization
Because measurements for operating a machine tool or an individual factory can be done at greater levels of detail and precision, optimizing solutions for machine tools and factories can also be very precise. However, measurements of supply chain operations must necessarily be less precise. So optimizing solutions for supply chains are also less precise. Optimal solutions that exceed what a supply chain model can realistically measure may be mathematically possible, but such solutions produce supply chain designs and operating plans that cannot be implemented in the real world.
For instance, suppose there is a truck in your supply chain that runs on a route from Cincinnati and drops off products at stores in Indianapolis and Chicago. After some experimentation you find the best result comes from a delay between departures for the truck of precisely 13.36 hours, as shown in the screenshot below.
This best result also assumes the truck can run its route to Indianapolis, Chicago and back (calculated at 957.84 km) and then wait precisely 13 hours and 21.6 minutes before departing again. It also assumes the truck will maintain a speed of exactly 90 km/hr over the length of the route. Calculations can be done assuming all these variables will be just as specified. But that does not mean the resulting precision of the calculations is realistic because the assumptions that made them possible are unrealistic.
Small changes to these precise but unrealistic numbers can deliver results that are almost as good. And small changes can create realistic designs that could be implemented in actual supply chains. Having a delay between departure of 12 hours instead of exactly 13.36 hours, and making minor adjustments to the product delivery quantities at the two stores gives results that are somewhat different but are still within the margin of error for the simulation. So they are just as good, and much more realistic.
Strive for Good Supply Chains not Perfect Supply Chains
Unless you are creating a detailed model of a very small part of a supply chain, it is best to avoid the illusion of accuracy that comes from using two or more decimal places of precision in the creation of your supply chain model. It is mathematically possible to use two or six or ten decimal places of accuracy, and create precise optimization solutions for a given period, but this is still just an abstraction, and not something that could actually be built or operated in the real world.
Remember, there is no need to use measurements that go beyond two decimal places, and often one decimal place or no decimal places will be adequate. Be conservative in the numbers you use. Assume products are a bit heavier and bigger, make vehicles travel a bit slower, set rent costs higher, and define demand levels somewhat greater than expected. When you find a supply chain model that works well with those assumptions, then you know it will work even better if the actual numbers turn out to be more advantageous than those you used.
In a world that is volatile, unpredictable, complex, and ambiguous (VUCA), events can be hard to predict and hard to control. So supply chains should be built with an appropriate degree of resiliency. The degree of resiliency is determined by the conservative nature of the measurements and assumptions used to design a supply chain. Resilient supply chains are those that have extra capacity available as needed to deliver and sustain good performance even under conditions that are more demanding than what was expected.
Perfect performance in a predictable world is an illusion. Value lies in designing resilient supply chains that deliver good performance in an uncertain world.
NOTE: As you build more complex supply chains be sure to use advanced modeling techniques presented in “Tips and Techniques for Building Supply Chain Models“