> For all the data driven ethos that execs tout, this is one area where decisions seem to be made at the whim rather than derived from data. Having worked at p

> For all the data driven ethos that execs tout, this is one area where decisions seem to be made at the whim rather than derived from data. Having worked at places that take the data-driven ethos to an extreme, there are very straightforward pitfalls to ignoring other kinds of reasoning. Those other kinds of reasoning have pitfalls too, obviously (hence the popularity of data-driven decision-making). But this is exactly the type of thing where I wouldn't expect data to be especially helpful. People keep trumpeting studies claiming to measure productivity increases from WFH, and I do believe there's some signal there, as I've been a big proponent of hybrid workplaces my entire career. But the idea that productivity can be measured in a meaningful, complete way, and that that applies broadly to every company, particularly creative knowledge work, is facially ludicrous and not (yet) supported by the data. A decade from now, researchers run careful comparative analyses on companies that went full-remote, along with trying to build narrower measures of productivity, and hunting the data for natural experiments. The idea that we already have data-driven certainty around this is laughable. As an example of how this works: I work at an applied research firm, and as much as I enjoy some of the QoL changes from full-remote, there is no question that my and my colleagues' ability to collaborate creatively has significantly suffered. How do you meaningfully measure the change in our "productivity"? I'm sure we're all shipping more code, but since when is more LoC better? How do you immediately measure the cost of a system with less innovative experiments, and how do you incorporate stochastic and time-lag factors? These factors are very opaque to industry-wide data-centric analysis, so we can't be dismissive of other forms of reasoning, particularly those that understand each data point more intimately (if less rigorously).