The article didn't mention that Sears almost sunk itself with overwhelming logistical challenges in the early mail order days. It was a close call; if they didn't get a handle on those challenges back then, the capital inefficiencies would have erased them. This article alludes to the problems [1], but there was one article I can't find now that gave some specific examples of the astounding inefficiencies they successfully grappled with. When starting, it was far from obvious that mail order would succeed, just like it was far from obvious that general retailing over the Net would be successful in many categories when Amazon started.
> Then however, I start to wonder what shift might that be?
Not possible, but what I would like to see as a shift: Much deeper logistical integration with stakeholders that involve sophisticated, complex, ever-evolving, and hard-to-impossible-to-reproduce trust relationships between manufacturers, the retailer, the transport infrastructure the retailer depends upon, and the end-consumer. Turns the Net retailer into an end-consumer's direct-from-producer supply chain manager, cut out vast swathes of distribution middle layers, and pass savings directly to the end-consumer, in exchange for embedded preferences shared by the end-consumer over a long period of time (decades). There is yet to emerge a retailer I would trust with sufficient monitoring capabilities to deeply embed within my household, auto-manage all the various supplies, lifting the cognitive load of that aspect of household management. Even if one did emerge, they wouldn't have the history established that I would trust they wouldn't change to monetize that trust later on, because existing incentive structures make that trust very difficult to maintain.
On a concrete scale, I'd be happy to establish that kind of relationship to auto-monitor and auto-order direct from producers various groceries and consumables, for example. And let machine learning pick up enough data from my habits to establish my preferred level of BIFL-ness/value/zero-wastage (or other quality axes), combining with others in my same selection criteria strata to crowdsource the selection results, and auto-suggest best-fit matches, with accompanying explanations, reviews, and historical reviews (an area Amazon currently has a gap at systematizing---very difficult to find out how many products do over long periods of time). For the producer and retailer, this shifts constant fighting for consumer attention to a vast steady demand that they can plan around and address in many ways logistically; if a producer knew the consumer trusted the product that came in the package was X brand dishwasher soap for example, then the packaging can be plain, completely recyclable, and perhaps even reusable, and even delivery routes can be optimized. However, the way the market currently structures incentives, many metrics are excessively gamed by the producers or it outright goes-to-the-highest-bidder, or producers and retailers work together to dilute initially-high-quality offerings over time. The Holy Grail of many marketing execs is this kind of locked-in preference that simply exists as a consumer's background radiation; I'd be happy to play along with that for specific strata of quality metrics, for an expressed explicit profit margin granted to that supply chain, if I knew I wouldn't get fleeced over time.
[1] http://www.searsarchives.com/history/history1900s.htm