Embrace Uncertainty

Supply chains are rife with uncertainty. From demand to supply, and everything in between. Anything that has lead times, durations, rates, quantities, yields, grades or specs, it has variability. When we plan an exact amount at an exact time, it never comes out exactly as planned. The simple remedy is to add buffers. These buffers can be of three kinds: time, inventory, or capacity. Unfortunately, most companies do not map exactly where and how large the uncertainties are. They just add large buffers across the board, at best differentiated in a few groups, such as ABC classification. Lead times get buffered with extra time, everything else is covered by large amounts of safety stock. And still expediting and stockouts are an everyday fact. This is extremely inefficient, it hurts margins and competitiveness. Customers need to accept long lead times and low service
levels, and we still need to hold excessive levels of inventory to achieve that.

As simple as possible, and then much simpler still

The above issues are all caused by ignoring or fighting uncertainty. The few ways
we attempt to tackle it are ridicously simplistic. In terms of lead times we either
negotiate them or set arbitrarily high values. Neither is very scientific. In terms of
inventory we use simplistic formulas that ignore most significant factors and
approximate the factors that we do consider with incorrect assumptions. Safety stock for example needs to not only buffer for demand uncertainty, but also for supply uncertainty, or you will never achieve the service levels you are targeting. Factors such as lot sizes, order intervals, MRP frequency are all important. But by far the biggest problem is assuming we can use a predicted forecast error and a Z-factor in our inventory formula. The former is known to horribly underestimate the true error, typically by a factor of 2 or 3. The latter assumes a normal distribution which is only true for the fastest, smoothest demand patterns; the patterns which require the lowest safety stocks. As demand patterns get slower, more intermittent, or lumpier the normal distribution breaks down. The more we need safety stocks, the less accurate our trusted formulas get. Simultaneous stockouts for
some items or locations and overstocks for others are a direct result.

It does not need to be that way

Rather than fight uncertainty, your planning needs to embrace it! Instead of
guessing, use scientifically sound means to quantify it, then work with it instead
of against it. It all starts with a proper understanding of demand uncertainty.
Next month you will not sell exactly 100 units of item A; instead demand could
come within a whole range of possible values, each of which has its own
probability of occurring. Instead of predicting that single value of 100 units, a probabilistic forecasts predicts the entire distribution across the whole range of possible demand values. The 100 units will often be somewhat near the middle of that distribution, but it is not very relevant, unless you are aiming for a 50% customer service level. Forecasting the tails of the distribution is much more important. The upper tail (high demand values) will directly impact service levels, whilst the lower tail (low demand values) will impact waste and obsolesence, which is especially important when products have low shelf life or short life time. Traditional forecasting estimates the size of the error (a proxy for uncertainty) after the forecast is generated. It aims to get the best possible average and then the tails are just what they are. Probabilistic forecasting turns this around. It aims to get the best possible forecast of the distribution, including the tails. If we really need an average demand value - for example because an external planning system needs exact numbers - we can simply take the expectancy of the distribution.

Supply chain performance erosion by assuming normal variability

If we isolate just the effect of not forecasting the variability correctly, the deterioration in the graphs below occurs. In other words the graphs assume the forecast has zero bias and the size of the variability is predicted correctly in terms of standard deviation. Only the shape of the uncertainty is predicted wrong. In practice this will not be true and the results are typically much worse still.

Red lines are the probability density functions (pdf) of residual errors of traditional forecasts. Blue lines are what really happens. An unbiased forecast may have zero error in the middle, it is the tails of the distribution that really matter, and where the errors of the traditional forecasts get ever worse as demand patterns become slower and lumpier. 

Don't stop there

It's great to have a forecast with an accuracy that is better - consistently much better - than the theoretical best achievable using traditional methods. But the real benefits are in how you then use that forecast. Instead of plugging values into a safety stock number and using average predicted demand values for all your planning, use the rich information contained in the full probability distributions. First, determine an accurate trade-off curve between inventory and service levels, incorporating all factors that matter, including the full demand uncertainty. Then balance item against item, location against location, including constraints such as capacity and raw material availability, to simultaneously lower targeted overall inventory levels and raise overall service levels. Finally, take the same richness of information into account when creating plans and schedules to actually achieve those targets. Visualize and avoid high risk situations, and grab opportunities to gain market share that you would otherwise only would have known in hindsight when it is too late.

Stabilize your supply chain

When you ignore or fight uncertainty, it fights back, and it always wins. Uncertainty is like the waves on the ocean: they keep coming and you can choose to ride them in exhiliration, or you can fight them and eventually be washed away or be drowned by them. While you fight your supply chain it will be in chaos. You are constantly expediting and transfering product at premium cost. Your customers will continue to complain about delayed or missed deliveries, and some may just go to your competitor. Your planners are working overtime just to keep up with all the fires that need fighting. As soon as you stop fighting and start working with your supply chain it stabilizes. Expediting and inter-warehouse transfers slow to a trickle - typically 10% to 30% of earlier levels. Importantly, planners' time frees up allowing them to spend it on real value-add work, making the total value much greater than the immediate savings.