We now have the sources of data that we may use to quantify the magnitude of random (unpredictable timing) variation for safety stock:
- Recent (current) historical daily actual time-series demand
- Recent (current) historical daily actual time-series lead times; or an informed estimate of average, 5% minimum and 95% maximum lead-time ranges
To determine correct safety-stock levels, we must next address two key issues:
- Isolating, quantifying and statistically defining the random-variation component of the demand and lead time data outlined above, and
- Quantifying and defining the other factors that have a significant effect.
Note – though we will use historical time-series demand data for our discussion, the logic applies equally to lead-time data.
Our goal in analyzing the historical actual time-series data is to define and quantify the underlying structure of its random variation. However, historical actual time-series demand data contains more than just the random variation we need for safety stock. It is also made up of:
- Mean demand, which is critical to forecasting and to reorder point (should we choose to use this replenishment technique)
- Trend, which we will forecast, but which we must exclude from optimal safety stock. Trend may reflect new product introductions, product phase-outs, market growth, etc.
- Seasonality, which – just like trend – will be in the forecast but must be excluded from safety stock
- Foreseeable events such as promotions, sales, store or distribution-point openings, etc. – like trend and seasonality, these will be in the forecast, but must not be in safety stock
- Unforeseeable, rarely-repeating special causes such as natural disasters
This means we may need to decompose, or “clean,” the raw data in order to isolate the random-variation component. Frankly, this can be a real challenge that requires a combination of statistical expertise and familiarity with the inventory items themselves. The good news is that not every data set requires this cleaning – proper statistical pre-screening can identify the exceptions requiring further scrutiny due to significant non-random variation.
Once the data sets that require cleaning have been identified, the “cleaning” itself begins. This is the topic of our next blog
Hopelessly Unabashed Sales Pitch: TopDown Lean Systems can provide this pre-screening, to avoid “garbage in – garbage out.” More information available on our web site!
