A key promise of embedding AI in a Business Intelligence tool is how it can help to surface hidden insights that we often miss using the traditional method of building manual reports and dashboards. However, there are insights and there are ‘insights’.
Automated insights are defined as insights that software can generate on its own — without human assistance. With the rise of automated insights in recent years, many BI tools claim that they have the power to automatically generate actionable insights for the business (e.g. finance, sales, marketing teams).
However, as the technology of automated insights is relatively still in its early stages (as compared to other BI technologies), there is a huge gap in the quality of auto-generated insights created. Namely, the quality of automated insights generated by one BI tool is not the same as another. Some automated insights may not be ‘insights’ after all.
Whilst one could select any number of criteria for a BI bakeoff — e.g. fancy chart options, auto generated text etc — , I want to focus on truly what should make an automated insight an ‘insight’. So for this BI tool comparison, the criteria to assess the quality of automated insights are:
- Aha moments
Arguably the most important criteria. If the software fails to generate accurate insights users will lose trust and adoption will not happen.
Out of all the automated insights generated by Power BI, none were accurate.
The root of Power BI’s problem here is its lack of an understanding of how to treat a measure using business context. Power BI applies a default calculation type (count) on all measures and dimensions without taking into consideration the nature of the measure itself whether that is revenue, expense, employee number or lending balances.
All of the automated insights generated by Hyper Anna are accurate because of the treatment of measure and dimensions. This is because Hyper Anna uses different techniques in machine learning such as random forest and gradient boosting algorithms (gbm) and topic modelling algorithms to identify relationships between measures and dimensions.
‘Aha Moments’ are insights that are meaningful & unknown to the users up until the point of discovery. Typically these are insights that are 2–3 level deep, often not covered in traditional reports and dashboards (which often can only cover high-level KPIs).
Even though Power BI advertises 7 types of automated insights, in the test, it failed to deliver any meaningful or insightful insights. All of the insights generated tended to be either very high level or nonsensical. This is due to the fact that Power BI applies fairly naive and basic algorithms (top items, distribution) in attempt to uncover hidden insights.
Hyper Anna’s utilises complex algorithms such as time series decomposition and clustering to create level 2 & 3 insights. Users can expect to find insights that are focus on changes, proportion and size across both long and short term trends.
Both products are on par in this category. All automated insights for this demo dataset (25MB) were generated in less than 2 minutes across both products.
Whether it’s auto-generated song designed to delight or an auto-generated insight designed to surface hidden patterns in business performance, nobody wants to receive automated nonsense. With the advances in BI technology, new generation BI tools like Hyper Anna are at the forefront of delivering insights that are truly insightful — accurate, fast and truly designed to surface ah-ha moments — so teams can focus on the actions that matter most.