Turning raw data into insightful information can be a massive undertaking. Once data is collected and ready for processing, deploying machine learning algorithms to recognize mathematical similarities in the dataset makes the data analysis-ready, but is still only the beginning. A machine learning algorithm will only tell us that the data follows certain patterns; the real analysis begins when we start determining how and why it follows such patterns. In order to do so, an analyst must have strong statistical skills to determine the significance of the patterns observed by the algorithms, be well versed in the business model and industry specific to the client, and also be able to present the findings in a manner that is quickly understandable to the client.
Our Advanced Analytics team recognizes that visualizing data can be one of the most effective forms of exploratory data analysis, as well as an excellent medium by which to present findings to the client. If done correctly this process need not be linear, but should rather be a cyclical engagement where the observer can view the results but is also guided by them to further question and analyze the information that is being presented. With our expertise in visualization methodologies and visualization tools like Tableau, our team offers a cyclical process of continuous engagement between our clients and their data, where they are constantly gaining insights from the information being presented and are further exploring the answers to new questions every step of the way. This blog post will further explore a scenario where these practices came into play in a real client situation.
West Monroe Partners’ Advanced Analytics team was engaged by a large insurance company, hereby referred to as Insurance Company, to gain better insight into their own sales agents’ behaviors in order to be able to better allocate service offerings to the agents for growth and expansion. Our Advanced Analytics team used the K-Means Clustering algorithm to break out their agencies into various clusters, grouping them by mathematically similar and statistically accurate characteristics. This analysis preparation was performed on over 20,000 agents across 150+ statistical attributes. Once the agents were grouped into their respective clusters, we were able to begin analyzing the defining characteristics of these clusters that separated their agents from the agents belonging to other clusters.
As a deliverable of the engagement West Monroe Partners created drill-through visual representations of how the agencies in their respective segments differed from the others insurance agency’s in order to drive growth conversations among the insurance agency leads.
The charts above show a visual representation of the segments. These charts allow for drill-though analysis as well as contextual analysis (as one chart is drilled into, the remaining charts are relatively affected).
- (Left) – A treemap of the segments showing their responsiveness to different service offering groups; allowing Insurance Company to analyze whether their responsiveness to the service offerings affects their bottom-line statistics.
- (Top-Right) – A map of each insurance agency sized by total auto premiums and colored by segment. This allows us to quickly determine correlations between the geographical location of an insurance agency and its segment assignment, allowing for instant visual geographical analysis of the data.
- (Bottom-Right) – A scattered box plot of home premiums for individual agencies grouped by segment. This allows us to visualize the difference in count of agencies within a segment and quickly compare the minimum, maximums, and medians for each segment.
The next request from Insurance Company was for a comparison to data that was generated by a simple qualitative flow chart to break down the agencies. With only 3 levels of data this lead to a massive gap in data integrity (3 attributes vs the 150+ from our analysis).
The chart below is a Sankey diagram representation which shows the overlap of a segments from a segmentation model created by a prior firm (right) to the ones created by West Monroe’s segmentation model (left). The prior firm’s efforts were a purely qualitative approach, with no mathematical understandings. The comparison of the two groupings showcases how the true breakdown (and subsequent sub clustering within each cluster) that we performed gets away from the subjective analysis and drives firmly into the quantitative statistical modeling yielding hard facts and action items. The client was able to take these clusters, define and drill into multiple levels, and generate actionable and track able insights. As the agents progress based on the offerings, the historical trending is reviewed and visualized to help showcase direction.
Below is a snapshot of how the diagram looks when an individual segment is hovered over. This snapshot shows us that what the prior firm referred to as the Elite segment (based on only 3 levels of data) is actually made up mostly by what West Monroe’s model refers to as the Passive segment (with the true statistical representation across the 150+ levels of data).
By presenting the client with our analysis in a visualization tool like Tableau, they were enabled to quickly see how the agents within their respective segments differed from one another. They noticed that agents within the Passive Agency cluster had a much larger percentage of agents that did not respond to some of their offerings compared to agents within other clusters. They also found correlations between agent participation in the offerings and agent performance in terms of attributes like total home and auto-insurance premiums written. With this type of information, the client was able to determine which agencies they should invest more in, how they should differentiate their offering incentives for certain agents vs. others, and ultimately put them in a position to create well-informed decisions and strategy around how they wanted to continue doing business with the agents on an individual level and on a segmented level This quick analysis capability allows clients to be much more engaged with their data and draw meaningful insights in ways that would not otherwise have been possible without this level of interactivity.