Results produced from ‘Search’ should be the next best action
September 12, 2017
Next best action is an algorithm and framework typically associated with marketing programs which determine the next probable action in a sequence of events to affect lead conversion. Within the marketing community, it has been used to determine what, when and how you engage with your customer or prospect. However, this is fundamentally is the purpose of enterprise search technology (now also known as insight engines and cognitive search).
You see, users are on a mission and your application or experience is simply a part of that to assist in that. They care little of why it’s difficult for them to complete that journey and they care less about the underlying technology for that. Every page view should optimally be producing the next step in their journey purposefully…sometimes even considered micro moments .
Keywords used to determine the most probable outcomes/results. However, Profiling and Signals are much more able to predict the likely event of a next action. The difference outlined here is that keywords that are supplied by the end user often are not sufficient or the best indicator of what they are looking for. Is a search for “Transformer” an electrical device or the movie characters? Using signals that are gathered from the navigation path or other sources (like past viewing habits ) is much more useful to predict similar behavior.
Insights in Action
Given that, what are some techniques to product next best action?
Learn To Rank
Learn to rank has been adopted by many of the leading commercial vendors. It’s now also in Solr and Elastic via plugins and extensions. What this does is re-rank the results based on previous outcomes which were determined to be successful. It doesn’t require keywords. Many times these are:
- Previous searches which the user clicked a non-top 3 search results
- Previous searches which when clicked through produced a positive outcome (case deflected, product purchased)
Topic Extractor
From parts of an email or a web page, you can scan information produce relevant knowledge articles, community posts that have been used to successfully close similar cases.
In Action
Given signals from which articles helped close cases (something that is typically gathered in KCS processes) and a topic extractor, you can develop a case detail page that incorporates these insights.
Google uses signals to suggest next actions related to your GSuite content. This include: Next Event, Recommended Reading and Pick up where you left off. (link )