I think when your dataset is small (up to a few million records), your querying complexity is low and you don't need super high performance then the database full-text searching should be sufficient. As an application's requirements change as well as the dataset grows, you can start to consider moving the heavier bits over to a full-text engine like elasticsearch. 

I've made the mistake of "prematurely optimizing" my applications and used Elasticsearch/Searchkick from day 1. The growth was never there and the query complexity remained low for the life of that product. We would have been find with some simple querying and/or using full-text within the database. Instead, we added infrastructure complexity and cost without the justification.

Then, you also have to look at cases like Hey email where they are doing massive amounts of records. I'd imagine that the query complexity is low, but their dataset is massive. Performance is also very important in these situations because someone isn't going to wait 10 seconds for a search to complete and be happy with that kind of delay.