Types of Analytics Handled by Auto Insights
Auto Insights is our AI-powered virtual analyst and it is very good at descriptive and diagnostic analytics. What does that mean? It means that like a human analyst, Auto Insights is great at summarizing historical data to determine what has happened and why.
Auto Insights' analytic capabilities are focused on these main areas:
Trend analysis—identifying patterns in structured and transactional data to help users understand performance over time.
Root cause analysis—determine drivers of change by utilizing several machine learning techniques such as Cramér's V to highlight the most likely causes (see here for more information).
Common cause analysis—looks at two metrics in your data and quickly gives insights into their relationship. Auto Insights uses the Pearson correlation coefficient, lead/lag, ratios, and impact to show you how closely related the two metrics are, whether or not they usually move together, what the ratio of their relationship is, and which underlying drivers have gone up or down for both and which ones have gone up in one metric but down in the other.
Outlier detection—proactively surfacing unexpected insights hidden within the data through various algorithms including STL, S-ARIMA, and PCA.
How does it work?
Like any good analyst, Auto Insights will determine and apply the most appropriate statistical technique based on the scenario presented before it to provide actionable insights. For example, to conduct root causes analysis, Auto Insights utilizes ensemble learning which combines several machine learning techniques such as Random Forest, Named Entity Recognition (NER), and Cramér's V into one analytical model to decrease variance, bias, and improve prediction accuracy.
Data Stories
Not only does Auto Insights use artificial intelligence to automate the insight generation process, but it also packages these insights into intuitive stories and charts for users to consume (examples listed below). To perform this type of analysis, Auto Insights requires structured and transactional data (more information on data requirements can be found here).
Example data stories:
Large increases/decreases—surfacing largest change and percentage change per category.
80/20 principle—identifying top categories that make up a majority of the total.
New, lost, and returning—calculating new, lost, and returning categories within the current period.
Outliers—highlighting unexpected changes within the dataset.