The holy grail of inventory control is the ability to forecast what customers are going to buy and when, order those parts from the supplier, and sell them to the customer before you pay the invoice to your supplier. And to never ever carry excess stock.

For 10 years or so now c8/c9 has offered some basic tooling that provides dealers an estimate of what stock re-order levels should be; i.e. max checker. It generates reasonable numbers based on analysing the sales figures for up to 12 months, but there are a number of things that cause it to generate poor results under some circumstances; such as sporadic/slow moving parts or seasonal stock.

With advent of c9 and some very powerful enabling technologies available to us, we are now exploring and testing more sophisticated statistical approaches to the problem of providing more material support to spare parts managers in their task of tuning restocking levels.  We are testing statistical models that encompass a dozen different variables and considerations, and we are tuning these models against 'flight simulators', that test how variation in different pre-set biases affect business profitability when it comes to stock control.

Flight Simulator approach

A novel approach we are using is to create a model for restocking and to then test that model by comparing how it performs against a historical period of business operation. We model, or simulate 6 months of sales and restocking activity, exactly as it occurred based on historical data, but we compare how the period of operation would perform with varying models for restocking; where those models can only use information available to them prior to period of simulation; i.e. we test the models ability to accurately forecast sales activity for a period of 6 months.

A flight simulator needs to tell us how successful a restock model is. Specifically things such as:

• How much dead/excess stock results (stock that does not move in 6 months)
• Unrealised profit because of difference between stock and daily discount offerings. i.e. part sold out of stock generates more profit than a part put on overnight order
• Lost sales due to inability to fulfil a sale immediately out of stock (i.e. assume one in 5 customers will not purchase if cannot fulfil invoice immediately)
• Number of daily orders raised
• Overheads with freighting on daily orders

Restocking from a computer aided perspective

Consider the ideal spare part.  A restocking order is placed every 2 weeks. In those two weeks, each and every time, exactly 10 customers come in during that week and buy that part. Every single time, for years and years. In this scenario restocking is easy: you only need to order 10 parts every 2 weeks.

But the reality is more complex.  The amount of sales varies. Sometimes you'll sell 10, others 13, others only 7. Over time the part may become less and less popular.  There is a component of variation in a part and this variation is what makes restocking hard.  Here are two example sales reports.

The first shows a highly variable part that is steadily increasing in demand, the second shows a seasonal part that was possibly affected by the 2000->2009 drought.

Consider a different part for 12 months broken into 2 week intervals. We display what is known as a probability distribution. It shows the likelyhood that a certain qty of the part will sell in a given interval.

Now for this part, there is a good chance for any given 2 weeks somewhere between 10 and 16 parts will be sold. But there is a slight, but unlikely possibility that upto 70 parts may be sold in a single restocking period. In fact in past 12 months this is exactly what happened. It is unusual, but it can happen.

For restocking,  when handling variability the base option is to select a quantity that represent roughly what quantity of parts is more likely to sell and we have sufficient stock half of the time. This is the average. But you may wish to stock more - safety stock, or even stock less, depending on a range of considerations. Such as price of the item, how many you move, the profit gain for performing stocking orders vs overnight orders etc. Illustrating:

C9 Revised Restocking model

New restocking model under development that will, if successful, replace the maxcheck algorithm takes following into consideration:

• Sales history for upto 18 months
• More recent sales are considered more significant than older sales. i.e. if in past tend to sell 5 but now tend to sell 3, the system will bias towards 3
• Sales that occurred this time last year are also considered more significant : i.e. seasonal stock
• Average delay between stock orders. Different shops will send stock orders at different rates, i.e. a major franchise may restock every week, whereas an accessory may be only once a month.  New model automatically compensates for this.

For dealing with variation, the restocking model will bias towards aggressive or conservative based on following:

• List price of the item. Cheap items are stocked aggressively, expensive items which are highly variable are treated conservatively
• The difference between daily discount and stock discount. The bigger the difference the more aggressive restocking is
• In the period of analysis, i.e. typically 18 months, the part is sporadically sold. i.e. sold some periods, but not others. The more sporadic, the more conservative
• Volume of parts moved on average overall. More volume more aggressive restocking

Tuning the model

Model tuning is currently underway. For example what list price represents an ideal price.  Following results based on one dealers data.