Monat: Februar 2020

Because the topic of „pressure selling“ was in the news today,
we’d like to explain one of the things our shop software automatically does:

Because we import many toys from overseas, it does take longer for bulk orders to arrive. In the mean time there are often toys that do get out of stock. We try to prevent this from happening but it is unavoidable that this does happens. Sometimes the manufacturer has a minimum order quantity or only no other toys from this manufacturer need restocking. Then it may even stay that way for a short while. However unless it is stated otherwise, all these toys will be back for sale later.

Because this is happening regularly, we do show a reminder „only N left“. This is done automatically if and only if very few are left in stock.
The number shown comes directly from our inventory management system. At any time, it should precisely reflect the number we have left in stock that are not reserved for recent orders. Due to orders being cancelled or simple human error (e.g. a simple mistake in the last inventory count or counting an L toy as the XL version,…) that number may sometimes go up again. This is not typical behavior but it does happen.

Why is this important for you?

Yes, this topic sounds boring.
However statistics are the difference between a toy being out of stock or in your hands and between stocking too many of one toy or having the funds to add an additional model to our inventory.
We are a very small company and have limited money to work with.

Last year

Last year in February we made a huge leap by going from educated guesses and estimations to actual statistics for predicting how many toys to restock.
This has made visible a large number of overestimations, wrong impressions and wishful thinking.

It worked much better then human intuition but we did notice a few oddities here and there. Especially around the Christmas holidays, around the sales during Eurofurence or when we where away for MMC or H-Con.

This year

This year we went even further.
Since YiffyToys exists long enough to go back many, many years and we have all the additional values collected since last year, we can now scientifically evaluate how well our predictions match reality.

First, we defined a huge collection of test cases.
Then we very carefully defined the actual metric to optimise for.
And finally we went to compare different ways to do predictions.
Again the results where very surprising and sometimes counter intuitive.
However we know for a fact that our new statistics produce results that are easily 3 times more accurate then before.