Using Customer Insights as
business intelligence

1. Business Context & Demographics

The analysis began by evaluating customer demographics to develop an understanding of who's interacting with the Turtle Games. The initial data mapping reveals a highly specific user base.

• Gender: The customer base leans heavily female, recording noticeably higher engagement and transaction volumes than male counterparts.
DATA_VIZ // GENDER_DISTRIBUTION TYPE: HISTOGRAM

Fig 1. Histogram of customer gender distribution.
• Age: Age distribution models indicate that the primary customer demographic consists of adults, establishing a distinct peak forming around the late 30s to early 40s.
DATA_VIZ // AGE_DISTRIBUTION TYPE: HISTOGRAM

Fig 2. Histogram of customer age distribution.
• Education: The audience is demonstrably well-educated; the largest segment of customers holds a "Graduate" level education, closely followed by "PhD" and "Postgraduate" degree holders.

2. Customer Segmentation

Moving forward, I conducted an analysis of the customer behavior as defined by spending habits, renumeration, and loyalty points.

• Scatterplot Metrics: When plotting customer Remuneration (£) against their internal Spending Score (1-100), the dataset begins to fragment into distinct groups.
DATA_VIZ // LOYALTY_PATTERNS TYPE: SCATTERPLOT

Fig 4. Scatterplot of Renumeration by Spending Score and colored by allocated Loyalty Points.
• K-Means Clustering: By applying K-Means Clustering, an unsupervised Machine Learning algorithm, five distinct groups of customers were uncovered according to their spending habits, income, and allocation of loyalty points. (Cluster 0 to Cluster 4).
DATA_VIZ // CUSTOMER_GROUPING TYPE: K-MEANS CLUSTERING

Fig 5. K-Means Clustering designating five distinct customer profiles.
• Targeted Personas: These five clusters allow Turtle Games to clearly identify different customer groups that interact with their products, providing a foundation for targeted marketing.

3. Statistical Modeling

Multiple Linear Regression (MLR) was selected as the analytical approach because it effectively models the simultaneous relationship between a continuous target variable and multiple independent predictors. This method provides a clear framework for understanding how underlying factors like spending scores and renumeration drive the allocation of loyalty points.

Although both models utilize square-root transformations to increase linearity the final was ultimately selected to prioritize ease of explainability. By excluding additional variables like age, this allows stakeholders to better understand the straightforward relationship that spending scores and renumeration effectively explain the allocation of loyalty points.

4. Product Performance & Sentiment

By conducting an NLP sentiment analysis of customer reviews and summaries, several aspects are made clear about the data quality and availed impressions.

• Juxtaposing Reviews & Summaries: The summaries are intended to summarize the reviews, however, they don't accomplish this in a consistent manner. The summaries are used idiosyncratically by each individual data logger. This is problematic when trying to conduct a sentiment analysis. An intriguing finding is that the summaries were also used to record one-to-five-star ratings.
• Extracting Ratings From Summaries: When looking at the one-to-five-star ratings, product satisfaction appears to be exceptionally high. The vast majority of ratings are 4 star (n=57) and 5 star (n=378) ratings.
• Product Averages: Approximately 98.4% of reviewed products are rated at 3.0 or higher, while only 1.6% of rated products hold rating below this threshold.
DATA_VIZ // REVIEW_RATING_DISTRIBUTION TYPE: VECTOR_SVG
Fig 7. Breakdown of user ratings.
• Making Ratings A Requirement: Only a minority of products are assigned a rating. In the future, it would be highly beneficial to clean-up how summaries are used and to obtain one-to-five star ratings for each recorded transaction. This would provide a rigorous way to assess customer satisfaction empirically.
• Core Impressions: A sentiment analysis provides insight as to whether negative, neutral, or positive sentiments can be attributed to a given review or summary. When strictly for 1- and 2-star ratings extracted frequent keywords: customers found these specific games "boring," "difficult to use," "flimsy," "expensive for what you get," and hindered by "complicated" instructions. The products associated with these ratings require an urgent level of attention by Turtle Games.
• The Truth About Sentiment Analysis: Often complex paragraphs are not well-understood from simple sentiment analysis which have a tendency to lock on to singlular pieces of vocabulary like "good" or "bad," however this does little justice to reviews which express more nuanced impressions.

5. Business Recommendations

• Targeting New Customers: With the allocation of loyalty loints being better understood, this presents a definitive opportunity. It is time to encourage new and lower spending customers to engage with Turtle Games.
• The “Carrot on a Stick” Method: My recommendation is to use loyalty points as a proverbial “carrot on a stick” by introducing sign-up rewards, an increased allocation of points during sales events, and points for engaging with Turtle Games on social media platforms to expand the customer-base.
• Data Quality & Data Annotation: In the future, it would be highly beneficial to clean-up how the transactions are recorded. For instance, ensuring that one-to-five star ratings are associated with each purchase would do wonders. In addition, recording the specific item type and amount spent, plus the time and store location for each transaction would permit for a better assessment of trends over time. Each of these aspects would allow for more prescriptive recommendations that can be tied to specified customer groups.

Full Python & R Notebooks available here.