If you run your own restaurant and you're there every day serving food to customers, you have a very rich stream of immediate feedback on how your company is doing. The purpose is pretty clear. If you have a bad idea, people will tell you that your porkchop milkshake is terrible and you'll stop serving it.

As companies grow, the feedback loops get longer and the signals get attenuated. That allows larger companies to get more and more crazy. Especially so for companies selling to the "enterprise" market. There deals happen very rarely, allowing the crazy to metastasize.

Many non-profits have it worse, because they have broad missions. Their goal is to do good, but "good" often ends up being a broad category, so it's easy for them to end up doing a weird variety of things with varying degrees of effectiveness. Worse, even if you can get people to agree on a clear definition of "good", the feedback loops are weird. Since they rarely get money from direct service, they have to engage in a bunch of semi-related activities to fund the mission stuff. Often that's grant money, which means enterprise-like sales cycles and feedback loops that are way too long.

The non-profits I've seen do best have clear missions, specific revenue sources, frequent feedback, and a lot of discipline around picking projects. E.g., I've volunteered for the Long Now Foundation for years; they're smart, small, and scrappy. Or the California Academy of Sciences runs a damned good science museum, and I think it really helps that each and every day things have to work for their thousands of visitors. But the further non-profits get away from that, I think the easier it is for madness to creep in.