Precision and recall depends on product/algorithm and your corpus. For Phraser and FAQ we are typically seeing ~88% precision@1, ~95% p@3 (at 100% recall). For slot filling, ~98% per-label precision. Both are the most accurate in the industry.