[ckorhonen]

The main failure mode here I find is nuances in more specialized datasets. Elasticsearch is great, powerful and easy to work with once you get the hang of it, but is very open-ended with lots of room to build a sub-optimal solution.

Fuzzy matching and boosting often trip people up or lead to folks shooting themselves in the foot relevancy-wise.

If you want really great results, you need to spend the time crafting your query, dealing with synonyms ("pop" vs. "soda"), stemming, typos, negative boosts ("non-alcoholic", in the example) etc.

It's not necessarily hard, just often forgotten or not included in the initial scope.