>The technology team writing and running all that custom software was maybe 200-300 people.
> So even after I hear all the reasons LinkedIn has to be so huge (sales, support, scale, etc), I'm still left scratching my head.
Bear in mind that organizations, by default, scale superlinearly. If you add a team of 5 engineers, you need to add a manager. If you do that 5 times, you need to add a manager-of-managers. So to double the amount of labor to be done, you need to more than double headcount.
Beyond that, most of the big tech companies have a fleet of engineers building a product and a team 10x that for analyzing the customer experience. They're building streaming processing tools to move the data warehouse batch processing jobs into, so that an a/b experiment start and have statistically valid results within hours. Your retail operation likely can't build and deploy a planogram experiment that quickly, so there's no need to make the analysis pipeline faster.
Or they're analyzing page load times to shrink them down -- over and over again I've seen experiments proving that customers love fast UI, and hate waiting. The smaller the perf opportunity, the more effort it takes to find and fix. Where you dial in at depends on part in how big your customer base is; a 10ms perf fix is more valuable when you have 10million customers a day versus 100k.
And LinkedIn does this on like 3 platforms: web, iOS and Android. Then there's the web properties they bought like Slideshare and Lynda.com.
Bear in mind that organizations, by default, scale superlinearly. If you add a team of 5 engineers, you need to add a manager. If you do that 5 times, you need to add a manager-of-managers. So to double the amount of labor to be done, you need to more than double headcount.
Beyond that, most of the big tech companies have a fleet of engineers building a product and a team 10x that for analyzing the customer experience. They're building streaming processing tools to move the data warehouse batch processing jobs into, so that an a/b experiment start and have statistically valid results within hours. Your retail operation likely can't build and deploy a planogram experiment that quickly, so there's no need to make the analysis pipeline faster.
Or they're analyzing page load times to shrink them down -- over and over again I've seen experiments proving that customers love fast UI, and hate waiting. The smaller the perf opportunity, the more effort it takes to find and fix. Where you dial in at depends on part in how big your customer base is; a 10ms perf fix is more valuable when you have 10million customers a day versus 100k.
And LinkedIn does this on like 3 platforms: web, iOS and Android. Then there's the web properties they bought like Slideshare and Lynda.com.