The 3 success factors for enterprise chatbot initiatives

Enterprise search was to be the grease that freed the rusty gears of organizational information exchange. Point a search engine at content and presto!, the miracle of Internet search could be had by all.

Today, chatbots powered by AI free individuals from having to learn arcane search syntax in order to find the information they need, but inherit, rather than solve the central problems of enterprise search.

An enduring fallacy is that enterprise search is essentially Internet search on a smaller scale - but this is not the case. Internet search relies on observed behavior across a massive number of search transactions to identify and return the most relevant content, where relevance is defined as that which is accessed the most often or that which some business has paid to promote. Dollars and clicks are the primary factors that influence what searchers see. This seems a reasonable approach given that searchers often have no way of knowing what to expect, and have limited ability to assess the quality of the information returned. Absent any other means to judge the information provided, we accept the wisdom of the crowd combined with the self interest of those willing to spend money to influence the results.

While outwardly enterprise search appears similar to Internet search, the factors and forces affecting the results are quite different. The number of searches for any single category of information may be quite small, meaning that the search engine cannot collect meaningful statistics to improve relevance. And, while it may be desirable to identify, tag, and promote content to aid colleagues with completing typical tasks (e.g. check requests, expense reports, etc.), the incentives for individuals to create and maintain these may be much weaker than on the Internet, where money spent promoting content can be converted in to increased sales. Lastly, enterprise seekers will likely know precisely what they are looking for; their ability to discern the quality of the results is much greater than those using Internet search. The failure to appreciate these differences on the part of many users and implementers has led to generally poor outcomes for enterprise search. 

Enter the chatbot, a new and engaging way to search for answers. One can ask it questions in natural language and it will seem to understand and provide a plausible answer. The ease with which one can convey intent through a chatbot - as compared to the old way of typing one or more search terms - may lead to the belief that the mechanism for finding the results have changed. But this is not necessarily the case. 

Natural language processing is amazing, that is true! But once that piece of software has mapped my imprecise utterances to an "intent", and that intent is fed to whatever search technology my organization uses, the same three problems of low search volume preventing results optimization, limited information curation and promotion, and highly knowledgeable seekers with correspondingly high expectations are encountered. 

Unless the three problems identified above are addressed, the addition of chatbots to the enterprise search mix will at best ease the burden on searchers of needing to learn a specialized syntax; and might actually make things worse - since mapping natural language queries to structured intents (often referred to as "training" or "teaching" the AI) also requires time, skill, and ongoing commitment on the part of the organization. I do not argue for abandoning this promising new technology, but rather for a full appreciation of the challenges that must be addressed to realize its benefits.