Know the Lingo: 7 BI Buzzwords Explained
Dec 12, 2018
Originally published with BetterBuys.
Finding the right business intelligence solution is hard enough without also having to decrypt industry jargon. When different software vendors use the same word to describe different (or worse, slightly different) things, it can get even trickier to find what you’re looking for.
To help you get your search off on the right foot, we put together a shortlist of seven common BI buzzwords and their definitions. No jargon, just clear and concise explanations, so you can browse the market with confidence.
First up, a flashback to Latin class.
1. Ad Hoc Reporting
Also known as: self-service analytics, easy-to-use BI
Ad hoc means “for this” in Latin and refers to something done for a specific purpose. Ad hoc reporting, then, is the process of creating reports for a specific occasion (as opposed to for general use). Ad hoc reporting enables end users to build reports from scratch without assistance from IT.
The opposite of an ad hoc report is a canned report. Canned reports come premade (usually by the BI application’s administrator), and all end users simply run or export them to view the output. Canned reports are often read-only, but an end user can turn a canned report into an ad hoc report by making a copy of the canned report and modifying the copy.
2. Augmented Analytics
Also known as: smart data discovery, predictive analytics
According to Dataversity, augmented analytics “automates data insight by utilizing Machine Learning and Natural Language Processing to automate data preparation and enable data sharing.” This BI feature will sift through your data looking for patterns, predicting trends and suggesting means of visualizing them, making data insights more readily available to knowledge workers who don’t specialize in data analysis.
A number of technologies fall under the category of augmented analytics: Natural Language Processing (NLP) allows users to request information using natural spoken or written language, and Natural Language Generation (NLG) relays that information to users in natural language as well. Augmented Data Preparation, another facet of augmented analytics, “empowers business users with access to meaningful data to test theories and hypotheses without the assistance of data scientists or IT staff.”
Cumulatively, augmented analytics make the data more convenient and easily accessible.
3. Unstructured Data
Also known as: big data
Structured data can be stored in a relational database management system (RDBMS) and queried using SQL; whereas, unstructured data, by virtue of its not conforming to a set data model or schema, is typically stored in a NoSQL database.
Transactional and operational data are often structured; whereas, things like text documents, social media data, and satellite imagery go unstructured because they are so variable in form. Consider that an address or a price has a predictable structure and clear relationship to other data, but a text-based document can be any length and contain all manner of information.
It is, however, important to distinguish unstructured data from big data. Big data is high volume, high variety and high velocity data. It can be structured, unstructured or semi-structured but is commonly unstructured; this is where the conflation of terms occurs.
4. Data Lake
TechTarget defines data lake as “a storage repository that holds a vast amount of raw data in its native format until it is needed.” Rather than being stored in a hierarchical format involving files or folders, it is stored in a flat format with metadata tags that help direct queries to the right information.
A data lake should not be confused with a data warehouse, which is a repository of data that has undergone a process called ETL (which stands for Extract, Transform, Load). ETL grooms raw data for reporting by making it easier to analyze. There are a number of other specialized terms referring to types of data sets, but these are certainly the most common.
5. Data Model
Also known as: data structure, entity relationship diagram (ERD), join pattern
A data model depicts the relationship between data objects in a relational database. These objects, typically data tables, may be joined to one another in a variety of ways, and how they are joined will determine what data appears in a report referencing that model.
A hospital might have a Patient table and a Doctor table in its database, for example. The model (how these tables are joined) will determine whether the resulting report shows all patients and all doctors, only patients who have assigned doctors, only doctors with assigned patients or some other subset.
Some BI applications allow end users to select predefined data models from which to build their reports while others allow users to build those models from scratch. Still others generate the necessary models dynamically as users add data to their reports.
6. Embedded BI
Also known as: embedded analytics
Embedded BI provides analytics and reporting capabilities within a transactional business application. Rather than being a native part of the transactional application, the BI solution is integrated such that end users can view, manipulate and analyze their transactional data all without leaving the host application.
Embedded BI solutions sometimes differ in what and how they embed. Some allow end users to access the analytics suite as soon as they log in to the host application; others require a second login. Some allow end users to build reports right in the host application while others embed only the report output.
It’s common for embedded BI applications to be white label—that is, unbranded and highly customizable so as to blend in with the transactional application and provide a seamless user experience.
7. Pivot Table
In a traditional table, each column or field is identified by a column header, and each data point in the field exists on its own row. In a pivot table, a field’s data points become column headers, expanding dynamically as new data points are added. Row headers function in much the same manner, and a third field makes up the body of the table. A pivot table must contain a minimum of three fields but can accommodate more.
In sum, caveat emptor — let the buyer beware (or at least well informed). Anyone thinking about purchasing an embedded analytics platform may use these terms to evaluate key features and select the solution best suited to their business context.