Search-driven BI has emerged as an important trend in our industry. According to Gartner, “by 2020, 50% of analytic queries will be generated using search, natural-language processing or voice, or will be auto-generated”.
If you are in the market looking for a vendor in this space, what should you be looking for? Here are 7 things that we think are important.
# 1 Search to Author
In many BI tools, search is used to find existing pre-canned reports and dashboards. These tools match search terms with attributes of the pre-canned objects such as name and description.
In modern search-driven BI tools, search is used to author new reports and dashboards. Users interactively ask questions of their data in a simple search bar, and instantly get an answer as a chart or table. Users can save and share this answer with others.
# 2 Guided and Secure
As users search, the system should guide them by automatically helping to complete the search and reach the desired answer. Guidance should include suggestions of column names and data values.
Suggestions must be secure, honoring column-level and row-level security rules. Searches should be on centralized, governed, trusted data sources and not fragmented data extracts to user desktops.
# 3 Smart, Learning System
Smart guidance should incorporate constructs such as string matching (prefix and substring), synonyms, plurals, spell corrections, superlatives, phonetic matches etc. to guide users to insights they might not otherwise reach.
A guidance system must teach, but it should also learn. A system that continuously learns can offer more relevant personalized suggestions for each user as they use the system.
# 4 Speed at Scale
Search-driven ad-hoc analysis is feasible only if the data processing backend is extremely fast. As users type in a search bar, the responses must be instant. Such quick response requires a backend data processing engine capable of sub-second analytic query processing over terabytes of data.
# 5 Complex Schemas
The system should be flexible such that it can search on any arbitrary schema, without the need for any special modeling per business use case. Such a system should easily handle multiple join paths and double counting errors in many-to-many joins.
# 6 Rich Expressibility
A rich search language is key. Rich languages can express complex analytic queries. This allows users to utilize constructs such as top/bottom, growth, pre- and post-aggregated filters, time-series and comparisons in the search.
# 7 System-Generated Searches
Finally, the system should be capable of automatically generating interesting searches for users. By combining its knowledge of user’s previous interaction patterns and the characteristics of data, a system can automatically infer interesting insights. Such AI and machine-learning-based insights should be pushed to users’ desktop and mobile devices.
Use the above criteria to evaluate your options for an enterprise-class search-driven BI solution.