Hey! I'm Andrew. Welcome to my Github π
- π Data visualization is my forte. See my Tableau page!
- π Most of what you see here is Tableau, Python, and Figma
- π§ I'm interested in how our social climate and media trends influence BI decisions
- β³ A nice quote: "If it's worth doing, it's worth doing well"
-
The dataset required basic data cleaning and restructuring via Python. RFM modeling was done by manually assigning customer loyalty groups to various combinations of R/F/M values ranked from 1-4 (4 being the highest).
- Use lambda functions to establish Recency/Frequency/Monetary columns
- Split columns into quantiles with df.quantile
- Define two custom functions for scoring through 1-4
- One function for R scoring, and one for F/M scoring
- Apply functions to previously created columns to form new R/F/M scored columns
- Use for loop to segment RFM varieties into customer loyalty groups
-
The dashboard was designed in Figma and built in Tableau. It includes the loyalty groups (next section) as well as a visualization detailing k-means clustering.
-
Insights were derived from various graphs in the dashboard and noted in both the data insights section of this readme, and the tooltips within the dashboard's sidebar.
The dashboard displays a scatter plot of k-means clusters at the level of Customer-ID. K-means aside, we've also used a segmentation model to categorize our customers into various "loyalty" segments. (see Python notebook for code)
Loyalty | Requirement | Score e.g. | Tier | Description |
---|---|---|---|---|
Platinum | All scores = 4 | 444 | High | highest-scoring, most loyal customers |
Gold | All scores >= 3 | 333, 344 | High | very loyal, second only to Platinum |
Silver | Moderate R/F/M | 133, 324 | High | represents the dependable majority |
New | High R; Low M | 411, 431 | Low | made their first purchase recently, or their first purchase in a very long time |
High-Spend | High M; Low R/F | 113, 214 | Low | top 25% of sales, large purchases but very infrequent, not considered loyal |
Churn Risk | Moderate M/F; Low R | 122, 133 | Low | no recent purchases, at risk of defecting to competition |
Churning | Low R/F/M | 111, 112 | Low | lowest RFM values, likely receptive to competitor-focused targeting |
High-Spend Corporate Tech
Initially, it was assumed that many High-Spend customers were one-off bargain-shoppers. After reviewing our data, we arrived at a different conclusion.
-
The average discount of High-Spend customers was well below the mean: 13.88% < 15.75%.
-
It was more than just promotions that attracted these customers.
-
High-Spend customers strongly preferred Corporate and Consumer Tech products over other categories and segments. Corporate customers likely came became one-time purchasers when outfitting their properties with our products. This knowledge will give us the precision to target these customers more successfully.
Grouping and Clustering
These graphs aim to give perspective to the varying loyalty groups as we compare and contrast them. Similar to RFM scores, clusters are graded on an increasing scale from 1-4..
-
Silver and Gold are the most influential groups with regards to total profit. Platinum and High-Spend are the most valuable per customer.
-
K-means clustering displays the variance within groups and where they overlap.
-
Parts of High-Spend, Silver and Churn Risk all belong to Cluster 3 (hover to see tooltip) - our model interprets each group to be statistically similar to a certain degree. With proper targeting, Silver is a group we'd aim to grow and maintain. A future comparison will ideally show the "fringe" customers of High-Spend and Churn Risk to have ultimately moved into the Silver group.
Distribution Comparisons
The ternary graph illustrates customer representation per RFM score. Switch the graph to off to view the distribution as a matrix.
-
The dropdown menu above the ternary graph filters results by loyalty groups. The High and Low options reflect the tiers of each group and correspond to a column value in the RFM Scoring Key.
-
Color saturation in the RFM matrix is directly correlated to monetary (M) values of each cell. While the lowest RFM scores (111) reflect low monetary values, Churning still represents 10% of our customers.
YoY Comparison
In November and December of 2018, Churning customers were responsible for a total of $0.00 in sales.
- This is a sign that we need to conduct our customer retention efforts differently - especially during the holiday season. 2017 performance during these same months was much higher, mostly due to Consumer Furniture purchases.
B2C Prioritization
The spreadsheet below elaborates on specific values per Customer ID. The size table illustrates categories and segments by proportion of sales.
-
A large portion of Gold and Silver sales are derived from Consumer products of all segments. Home Office products are consistently the least successful products across the board.
-
We previously touched on the success of Corporate and Consumer Technology with High-Spend customers. While a new emphasis on B2B may seem enticing, Corporate Technology sales do not account for a large enough portion of our total sales in order for us to pivot our strategy in that direction. Consumer products should still remain our primary focus.