You Too Can Save Loads of Money: The Details of the Bayesian Model

Does their cost/benefit analysis incorporate the cost of paying data scientists to build these models?

I'm really confused as to how the weight of a SKU is not information that's readily available to them from the manufacturer, or why a missing weight can't be determined as part of the fulfillment process. After all, you only need to weigh a SKU once, so it's only a problem the first time that item is packed. You'd think knowing the weights of products in your warehouse would be a solved problem by 3PLs and not one requiring the services of people with graduate degrees in statistics.

Clothes and shoes have a high turnover (fashion) and multiple variants per SKU. I can see a manufacturer not providing perfect or accurate weights for every shoe size for every style.

If the authors see this, I'd be interested to learn what MCMC package they use.

we used our own C++ multi-threaded Gibbs sampler, which turned out to mix fast enough for the estimation - even given the large number of parcels/items we have at Zalando

Since the graph only has variables and the explanations are mentioned in the text, you have to keep cross referencing.

-1 for disabling zoom