Despite its many advantages, semantic search does not solve every search problem, according to Singh. “Semantic search has its limitations,” he pointed out. “Particularly when the content is badly misaligned with the problem, even the most advanced embedding models cannot bridge the gap,” he acknowledged. In those cases, the client needs to be convinced that the only viable solution is to rephrase or rewrite the content to better align with the questions being asked. “These discussions can be challenging, requiring well-supported datapoints, a clear presentation of data science insights, and a significant overhaul of the knowledge structure,” Singh remarked. “By addressing these gaps, however, we can ensure that the search system meets user expectations and delivers meaningful results.”
Search has proved to be a conundrum for many organizations, but the advent of new techniques and better integration into KM platforms will prove to be a driver for greater adoption. “A hybrid approach is the best solution,” said Stewart. “Lexical is great for precision, and semantic is great for recall. But with the advent of GenAI, which makes it easy to access so much information and get back a credible-sounding natural language reply, users also need to be vigilant about validating the responses.”