Maybe the approach needs to be tried in reverse?

Start where there's too much data for a human to keep up, and then work backwards.

Always possible that what's going on is backwards already, and what's needed is more forward thinking ;)

I still believe Amazon is uniquely qualified to dominate and capitalize in this "mass data gathering" space. They have trucks, packing material, and streamlined logistics. They have a ton of robots that are mobile and/or warehouse-task-optimized. They store and work with data on a massive scale. They sell a ton of products leveraging the statistical analysis of customers and suppliers. What other company has this mix?

Amazon should move families.

In the beginning, do it for free or at a reduced cost.

The catch is that they are then allowed to take a full, detailed record of every single item in the home plus a complete data capture of the moving procedure and then use all this data without restriction. (Thus, this scheme would not work in Europe without a great deal of additional thought and lawyering.)

I admit: I know neither how they model amazon.com customers nor the accuracy of their existing customer model. Regardless of how it already works, a handful of complete data points inserted into their model should help sell more products overall.

Suppose Amazon learns that 100 percent of cat owners have a Petmate carrier yet none of those were bought via Amazon. Or that 60 percent of Eames chair owners and 2 percent of non-owners have bidets. Most people would buy one but not the other at the exact same time and place, right? You'd target the chair purchaser with well-placed bidet ads and likely sell a few more of those.

Maybe Amazon can find these fine-grained customer details out via other channels, but nothing will beat self-gathered ground truth data, especially if it's a byproduct of some other service being offered. And sure, the examples are extreme, but---as with stock market arbitrage---any differences found from the existing model (and there SHOULD be many) should be exploitable.

Each moving job can be accepted/rejected/scheduled based on availability (plus whatever they assess as risks), so Amazon makes sure robotics/logistics capital isn't sitting around unused while also not hurting itself by needing to order a new fleet on a business gamble.

A ton of training data for robots is harvested, which can be used and sold.

There's probably a side-hustle with offering moving insurance, and having a video capture of the entire moving process for motion training would also ensure there's very little speculation of events after a damage claim is filed. This puts real monetary numbers into the training cost function, which should speed up improvements in on-the-fly vision-based value appraisal, price-aware object packing, choice of handling algorithm, etc. which are weird little subdomains but all worthy of advancement.

In the end, if Amazon finds it can relocate people more efficiently than existing firms after a few years in operation (mindful that human labor is expensive and eventually replaceable; plus small companies will never have Amazon's economies of scale) raise prices to somewhere less expensive than the competition while also profitable, and capture the entire rational home moving market.

(I also have no idea if this is actually a good idea and what the costs/profits of each factor of this business should be---more amazon.com sales via better ad targeting, the inventory data itself, the robot data, the relocation, the insurance---but my gut says they have the capital to lightly pivot and outmaneuver competitors until they become profitable.)