Thanks @Cali-Johnston! Paccurate is actually a class of software called "Narrow AI," which basically just means a set of algorithms oriented toward a specific goal or set of goals. We tend to avoid calling it AI because that conjures images of Skynet for some people, or just a meaningless buzzword for others. For reference, IBM Watson is Narrow AI- Watson is just a bunch of data scientists feeding data into algorithms.
In practice, the material savings are almost always sporadic- many shipments we don't find savings, but then a minority we find huge material savings. For one of our biggest customers who already had some cartonization logic in place, we save ~13%, or one square foot, of just cardboard per order (averaged out across all orders). And we probably actually only find those savings in 1/10 of their orders (EDIT: multi-item orders).
It only seems like magic because there's a vast amount of data the engine is crunching through to find packing solutions, including material and labor costs, type of packing fill, fragility constraints, negotiated rate tables, etc. All those factors make it too difficult to "eyeball" the optimal packing solution, for even an experienced warehouse worker.