Forecasting for Aviation Spares Inventory – Part 1
Frequently when there are shortages of parts for aircraft maintenance we hear: “We have been doing this for years, nothing has changed, why don’t we have the parts? We need to get better at our forecasting.” Inadequate forecasting techniques are wrongly seen as the root of the problem. The apparent solution is investing in the latest and greatest forecasting models, and when this inevitably fails to fix the issue, we resign ourselves to the idea that our business is unforecastable. But this is not true.The problem is not with the forecast; it is with our application of the forecast when deciding how many parts to hold. Rather than provisioning against the forecast, we must provision against the forecast uncertainty. So, the problem is a multi-dimensional one, and needs to be addressed from at least 2 dimensions mentioned so far, the Forecasting Dimension and the Uncertainty Dimension, to come up with the optimal provisioning quantity. There are other Dimensions to be considered also: Risk and Cost, which are addressed later in this article.
The Forecasting Dimension
When a forecasting project is launched, we hope to see a neat set of trend lines showing the past stretching into the future, with consistent patterns of demand against a range of parts required to complete planned maintenance activity. In reality, we discover the actual demand trend lines can be very erratic for most of the parts, and that the resultant poor forecast accuracy is simply unacceptable. We again find ourselves faced with the belief that our business is unforecastable, or that we are trying to ‘predict the unpredictable,’ so we are dependent on experienced guesswork, bolstered by our reserves of cash, to get by. Throwing the towel in and accepting this unacceptable level of inaccuracy is not the answer, but neither is persisting in resorting again and again to the same types of solutions that we have already seen fail. As Albert Einstein said, “doing the same thing over and over again and expecting different results is the definition of insanity.”
The Uncertainty Dimension
There needs to be a shift in the way we view the problem. We must accept that our business is not like Manufacturing or FMCG (Fast Moving Consumer Goods) planning and cannot be treated as such. The real problem here stems not from flawed forecasting, or even from unpredictable demand, but from refusal to examine, understand and address the true issue with aircraft MRO spares provisioning. The fact is our inventory demand is not chaotic, it is highly stochastic, i.e. “having a random probability distribution or pattern that may be analysed statistically but may not be predicted precisely” (Oxford dictionary). This means that while there is stochastic uncertainty, we can still quantify demand probability statistically and plan to mitigate this uncertainty in an economically efficient manner.
To illustrate this, consider a simple example.
If in the last 12 months we had 10 C-Checks, and one part had the following usage profile:
|# Checks||Quantity Used|
- What would the forecast be for the next check?
Without any sophisticated analysis, we can see that our usage in the 12-month planning horizon is under one per month, and the average usage is approx. 1 per check. When answering question 1 a time-series forecast would predict something in the region of 1 in the next month.
- If we have 2 checks occurring in the next month what quantity are we likely to use?
Again, a back-of-a-napkin analysis would tell us that 60% of the time we do not use the part. It is very likely that we will use 0 in at least one of the checks. It is also likely that the second check will use 0, however, as quantity 2 is used 30% of the time this is reasonably likely in one of the two checks. So, it is most likely that 0 or 2 will be used across the two checks. *
Now let’s introduce a third dimension mentioned above, Risk.
The Risk Dimension
- How many would we hold to deliver a high availability service level for the two checks?
Assuming that we require a relatively low risk service level in the 90% – 100% range, we can immediately see that 1 in 10 checks required quantity 4. With quantity 2 used 30 % of the time it is intuitive that we will hold 6 or perhaps 8 to deliver a low risk, high availability service level.
In this example, we see clearly the complications caused by highly stochastic demand patterns, and the difficulty of forecasting demand while grappling with these complications. We see how easily this difficulty could cause us to provision higher levels than we will probably need. However, the stochastic nature of the demand is an immutable feature of our business; it falls to use to find better methods to work with this difficulty and optimize our inventory costs. The final dimension to be considered, the Cost Dimension, is addressed in the second part of this article.
 *Many forecasting techniques apply time series. As we know, the drivers of demand also include scheduled and unscheduled maintenance, aircraft utilization etc. These combinational demands need to be considered – so a single forecasting model is frequently insufficient.