Last mile analysis and modern B.I
Last mile analysis, modern B.I
Last mile analysis : The analysis that is performed by a business user with an intent to test a hypothesis or explore an anomaly.
Enterprise data warehouses and B.I tools have existed for decades and yet Excel continues to be the tool of choice for “last mile analytics”.
What is last mile analytics?
The analysis that is performed by a business user with an intent to test a hypothesis or explore an anomaly.
– Business users don’t have time to waste, browsing dashboards, without intent. Most of the time they are trying to answer a question leading to insight about a business process anomaly or inconsistency. This is the last most important mile to derive true business insights.
– Users iterate on data asking questions. Simple pre-canned dashboards are just not enough. Hence users use the dashboards as a starting point, export the data to Excel( Spreadsheets) and do their own analysis. (Even a metrics driven tool like Google Analytics requires taking the data to a spreadsheet for last mile analytics)
– It is well known that humans cannot translate the knowledge in their head to other humans completely. One of the issues with B.I as it exists today is that the “authoring” of B.I content in the form of dashboards or insights is often done with “inputs” from the business, often by IT, and not by the business user herself. This knowledge gap grows wider and wider and soon the business is disconnected from what was built. And they go back to their Excel.
Excel and Spreadsheet hell have been discussed at length and in my opinion, it has nothing to do with collaboration. Google sheets and Excel have done a fabulous job enabling collaboration.
Iterative analysis imposes a burden on the B.I tools and many analysts simply get booted off the system because they were running “rogue queries”. An iterative analysis is by nature intensive and requires experiments, changing cells in a multi-dimensional model and running formulas propagating new values to several cells across dimensions. All this slows down performance and users eventually have to resort to building their own sandbox( Tableau extracts is one such sandbox) so they can do real B.I without waiting eternally for a single query to come back.
This creates a new problem. Now every user is performing the most important part of the analysis on their own Excel or Extract sandbox( many data scientists do the same taking a csv extract to R or Python)
Hence the larger issue with spreadsheets and extracts is that they don’t retain corporate memory of questions asked and parameters provided so they can be reused again and again without reinventing the wheel.
This “corporate memory” of analysis should be captured and maintained and one of the first steps is a consistent metadata at a centralized data access layer, capturing the trail of analysis starting with how certain metrics and calculations are defined. This trail of analysis should also capture the train of thought of an analyst/user and the process to arrive at an insight( as in Jupyter notebooks), not just what the B.I world calls KPIs.
Excel and Extracts can hide such a trail behind cells and formulas on the “client side” isolated from the view of other users.
A few things have to happen to fix this isolation.
- A modern B.I platform should provide HD level detail access to data at the speed of Excel navigation.
- Business analysis and the trail of analysis should be captured in a Notebook form or Tableau Story like tool rather than Dashboards so everyone can see how an analyst thought through a problem. Users should be able to publish their analysis and stories so they can be reused.
- Metadata about analysis( calculations, assumptions etc) should be captured at a centralized layer so it can be reused across all analysis – B.I /ML feature development/engineering / A.I etc.
- Democratize access to data – Everyone should be able to access detail data to build their data stories. Dashboards are built by a few and serve a purpose. But that is not B.I – its just a starting point.
- Executives should demand more than mobile access to simple dashboards. They should demand a how and why of a change in a metric in a story form. There is not much insight in your mobile app showing a fancy graphic that your revenue or ARPU decreased by 10% in the last month unless someone is already working on figuring out the why and how of it and present a story on what happened and how it can be addressed.