Also, purchased data can be low quality, of unknown correctness and completeness, and not include everything you need in its schema. For one startup, which add

Also, purchased data can be low quality, of unknown correctness and completeness, and not include everything you need in its schema. For one startup, which addressed problems for brands like counterfeiting and gray market diversion, we needed to monitor online marketplaces over time. Off-the-shelf scraping products we tried, which supposedly targeted one of the biggest marketplaces, were very poor, even when operated in their manually-invoked modes (nevermind scheduled). I ended up building a scraper that was correct and reliable, and captured all the data that we needed. And cost around $5/month to run. (This superior capability was not only useful in general for our services, but I understand it helped with cold enterprise sales approaches, from a tiny startup, e.g., maybe something like: "Here's an analysis of impact of X to your brand, based on data we monitor with some of our technology. Would love to discuss with you.") Controlling and understanding the software also gave us the flexibility to add capability quickly, as soon we had a need. For example, since the problems we were addressing were global, as soon as we needed to be able to see what different geopolitical instances of that marketplace were showing to shoppers from those same geopolitical areas, I was able to implement scraping of that. This was remarkably stable/resilient over time, as well. The biggest risk I perceived was something I warned the team about: "The first rule of Scrape Club is: Don't Talk About Scrape Club." I didn't want to get on the radar of the marketplace's lawyers, given that they were profiting from counterfeits and gray market, and possibly have public information turn into something that burns up the remainder of our meager Covid-time runway on lawyer billable hours. (Having data get cut off due to lawyers also seems a risk if using a vendor for the data.) Of course, there are often good reasons to buy data dumps/feeds. Just giving an example of when it wasn't the best choice.