CODE:

# Calculate mean star rating

mean_rating <- mean(reviews$rating, na.rm = TRUE)

print(paste("Average Star Rating:", round(mean_rating, 2)))

# Plot histogram of star ratings

ggplot(reviews, aes(x = rating)) +

  geom_histogram(binwidth = 1, fill = "blue", color = "black") +

  labs(title = "Distribution of Star Ratings", x = "Star Rating", y = "Count")

OUTPUT:

"Average Star Rating: 4.04"

Analysis

1. Average Rating = 4.04

   • The average star rating is 4.04, meaning most customers are generally satisfied with the products.

   • However, it's not a perfect 5.0, indicating some areas for improvement.

2. Star Rating Distribution (Histogram)

   • The majority of reviews are 5-star, showing that most customers had a positive experience.

   • There are fewer 1-, 2-, and 3-star reviews, but they still exist, meaning some customers were dissatisfied.

   • 4-star ratings are present but noticeably lower than 5-star ratings.

Marketing Implication: Sephora should highlight positive reviews in marketing while addressing concerns from lower ratings to improve customer satisfaction.

CODE:

# Calculate review length

reviews$review_length <- nchar(as.character(reviews$review_text))

# Plot histogram of review lengths

ggplot(reviews, aes(x = review_length)) +

  geom_histogram(binwidth = 50, fill = "purple", color = "black") +

  labs(title = "Distribution of Review Lengths", x = "Review Length (characters)", y = "Count")

# Define long and short reviews based on median length

median_length <- median(reviews$review_length, na.rm = TRUE)

reviews <- reviews %>%

  mutate(review_type = ifelse(review_length > median_length, "Long", "Short"))

# Apply VADER sentiment analysis

reviews$sentiment <- vader_df(reviews$review_text)$compound

# Function to compare sentiment scores

compare_sentiment <- function(data, rating_value) {

  subset_data <- data %>% filter(rating == rating_value)

  long_reviews <- subset_data %>% filter(review_type == "Long")

  short_reviews <- subset_data %>% filter(review_type == "Short")

  long_mean <- mean(long_reviews$sentiment, na.rm = TRUE)

  short_mean <- mean(short_reviews$sentiment, na.rm = TRUE)

  t_test_result <- t.test(long_reviews$sentiment, short_reviews$sentiment, na.rm = TRUE)

  list(

    long_mean = long_mean,

    short_mean = short_mean,

    p_value = t_test_result$p.value

  )

}

# Compare sentiments for one-star reviews

one_star_comparison <- compare_sentiment(reviews, 1)

print(one_star_comparison)

# Compare sentiments for five-star reviews

five_star_comparison <- compare_sentiment(reviews, 5)

print(five_star_comparison)

OUTPUT:

$long_mean

[1] 0.2847692

$short_mean

[1] 0.1069118

$p_value

[1] 0.05727717

$long_mean

[1] 0.8346481

$short_mean

[1] 0.6845736

$p_value

[1] 3.180911e-12

Analysis

1. Review Length Distribution

   • The histogram shows that most reviews are short (around 200-500 characters), with fewer very long reviews.

   • There are some extremely long reviews, but they are rare.

2. Sentiment Analysis (VADER Scores)

   • One-Star Reviews

     1. Long reviews have an average sentiment score of 0.28 (slightly positive).

     2. Short reviews have an average sentiment score of 0.11 (closer to neutral).

     3. p-value = 0.057 → This means the difference is almost statistically significant (just above the 0.05 threshold), suggesting longer one-star reviews may have slightly more positive sentiment.

   • Five-Star Reviews

     1. Long reviews have an average sentiment score of 0.83 (highly positive).

     2. Short reviews have an average sentiment score of 0.68 (still positive but lower).

     3. p-value = 3.18e-12 (very close to 0) → This is statistically significant, meaning longer five-star reviews tend to have significantly more positive sentiment than shorter ones.

Marketing Implication: Longer reviews tend to express stronger sentiment, both positive and negative, so Sephora can leverage detailed positive reviews in marketing while analyzing lengthy negative reviews to identify key areas for product improvement.

CODE:

# Load text processing libraries

library(tidytext)

library(tm)

library(wordcloud)

# Create word cloud for positive reviews (4 & 5 stars)

positive_reviews <- reviews_clean %>% filter(rating >= 4)

wordcloud(positive_reviews$review_text, max.words = 100, colors = "blue")

# Create word cloud for negative reviews (1 & 2 stars)

negative_reviews <- reviews_clean %>% filter(rating <= 2)

wordcloud(negative_reviews$review_text, max.words = 100, colors = "red")

Analysis:

Positive Reviews

• Common words include “skin,” “cream,” “moisturizer,” “hydrated,” “love,” and “smooth”.

• Customers highlight the hydration benefits, with words like “moisturizer,” “hydrated,” and “smooth” appearing prominently.

• “Love” and “amazing” show that many customers are highly enthusiastic about their purchases.

• The word “Mer” suggests that the brand La Mer is frequently mentioned in positive reviews, possibly indicating strong brand loyalty.

Managerial Implication: Marketing should lean into the strengths customers love—hydration, smooth application, and overall quality. Leveraging user-generated content (e.g., customer testimonials, influencer collaborations) to emphasize these strengths can help reinforce the brand's reputation. Additionally, Sephora can use “hydration” and “smooth skin” messaging in ad campaigns to align with what customers value most.

Negative Reviews

• Common words include “dry,” “price,” “thick,” “greasy,” “expensive,” and “hype”.

• Pricing is a recurring issue, indicating that some customers feel the product is overpriced for its effectiveness.

• Texture-related complaints (“thick,” “greasy,” “dry”) suggest a mismatch between product feel and customer expectations.

• The word “hype” appearing frequently may indicate that customers expected more based on marketing but were disappointed with the actual results.

Managerial Implication: Sephora should adjust marketing messaging to better set expectations—if a product has a thick consistency, emphasizing “rich hydration” or “luxurious feel” could shift the perception positively. Additionally, offering sample sizes or mini versions could reduce price-related complaints, allowing customers to try the product before committing to a full purchase.

CODE:

# Scatter plot: Review Length vs. Sentiment Score

ggplot(reviews_clean, aes(x = review_length, y = sentiment)) +

  geom_point(alpha = 0.4, color = "purple") +

  geom_smooth(method = "lm", color = "red") +

  labs(title = "Review Length vs. Sentiment Score", x = "Review Length (characters)", y = "Sentiment Score")

CODE:

# Calculate average sentiment per star rating

avg_sentiment <- reviews_clean %>%

  group_by(rating) %>%

  summarize(avg_sentiment = mean(sentiment, na.rm = TRUE))

# Bar chart of Average Sentiment by Rating

ggplot(avg_sentiment, aes(x = factor(rating), y = avg_sentiment, fill = factor(rating))) +

  geom_col() +

  labs(title = "Average Sentiment by Star Rating", x = "Star Rating", y = "Average Sentiment Score")

Managerial Implication: This confirms that not all negative reviews are fully critical—some dissatisfied customers still mention positive aspects of their experience. Highlighting these mixed reviews in marketing (e.g., "Not for me, but still a quality product!") could help build trust and manage expectations for new customers. Sephora can also proactively address common issues in FAQs or product descriptions to prevent misunderstandings.

CODE:

# Load necessary libraries

library(ggplot2)

library(dplyr)

# Histogram of Star Ratings

ggplot(reviews_clean, aes(x = rating)) +

  geom_bar(fill = "blue", color = "black") +

  labs(title = "Distribution of Star Ratings", x = "Star Rating", y = "Count")

Managerial Implication: Sephora is positively perceived overall, but the sharp contrast between 5-star and 1-star reviews suggests a polarizing experience for some customers. This could indicate that expectations are not always met, leading to some extreme negative feedback. A closer look at 1-star reviews (via word clouds and sentiment analysis) can help identify common complaints, allowing Sephora to adjust product messaging or improve formulations where necessary.

Campaign Theme:

• “Real Results, Real Reviews” – Showcasing authentic customer experiences.

Content Plan:

• UGC (User-Generated Content): Encourage customers to share their before-and-after skincare results.

• Influencer Partnerships: Partner with beauty influencers to create tutorial videos.

• Social Proof: Feature long, positive reviews in Instagram Reels & TikTok videos.

Platforms & Timing:

• Instagram, TikTok, and YouTube Shorts

• Best Posting Time: Evenings (6-9 PM, Tues & Thurs)

Metrics to Track:

• Engagement Rate (likes, shares, comments)

• Sentiment Analysis (before vs. after campaign)

• Increase in 4-5 Star Reviews

CODE:

# Use sentimentr

vader_sentiments <- full_data_filtered %>%

  mutate(sentiment = sentiment_by(review_text)$ave_sentiment)

# Compare with rating

ggplot(vader_sentiments, aes(x = factor(rating), y = sentiment)) +

  geom_boxplot() +

  labs(title = "Sentiment Score by Star Rating", x = "Star Rating", y = "VADER Sentiment")

CODE:

# Bar chart showing average ratio per brand

brand_ratios <- ratio_data %>%

  group_by(brand_name) %>%

  summarize(mean_ratio = mean(pos_neg_ratio, na.rm = TRUE))

ggplot(brand_ratios, aes(x = brand_name, y = mean_ratio, fill = brand_name)) +

  geom_col(width = 0.5) +

  labs(

    title = "Average Positive/Negative Review Ratio by Brand",

    x = "Brand",

    y = "Mean Positive/Negative Ratio"

  ) +

  theme_minimal()

CODE:

# Plot wordclouds by rating group and brand

plot_wordcloud <- function(data, brand, group) {

  data %>%

    filter(brand_name == brand, rating_group == group) %>%

    count(word, sort = TRUE) %>%

    with(wordcloud(word, n, max.words = 100, colors = brewer.pal(8, "Dark2")))

}

# Examples:

plot_wordcloud(review_words, "L'Occitane", "High")

plot_wordcloud(review_words, "L'Occitane", "Low")

plot_wordcloud(review_words, "La Mer", "High")

plot_wordcloud(review_words, "La Mer", "Low")

CODE:

# Average rating by skin type, eye color, etc.

full_data_filtered %>%

  group_by(skin_type) %>%

  summarize(avg_rating = mean(rating, na.rm = TRUE), n = n()) %>%

  ggplot(aes(x = reorder(skin_type, avg_rating), y = avg_rating)) +

  geom_bar(stat = "identity", fill = "steelblue") +

  coord_flip() +

  labs(title = "Average Rating by Skin Type", x = "Skin Type", y = "Average Rating")

Both the VADER sentiment scores and the positive-to-negative ratio method yielded consistent brand comparisons. However, VADER provided a more nuanced understanding of emotional tone, while the ratio method was better for quick, interpretable benchmarking.

Managerial Implications: This analysis highlights the importance of looking beyond star ratings to understand the language customers use when describing their experiences. While both L’Occitane and La Mer receive praise for hydration and texture, customers who give lower ratings often mention issues like greasiness, breakouts, or skin irritation, especially among those with oily or sensitive skin. These patterns suggest that product formulas or marketing strategies should be adjusted to directly address these concerns. Additionally, the demographic insights show that customers with certain skin types consistently give higher ratings, which can guide more targeted campaigns or inspire new product lines. The comparison between L’Occitane and La Mer shows that even small differences in review sentiment and satisfaction ratios can impact how brands are perceived in the market. By combining sentiment analysis with customer attributes, brands can make more strategic decisions about product development, positioning, and competitive differentiation.

Applying the Customer-Centric Framework

1. Customer Acquisition:

   1. This box offers a way to reach both Starbucks loyalists and beauty enthusiasts through joint email campaigns, co-branded packaging, and digital ad targeting. Launching pop-up stations in select Starbucks stores with QR codes for sign-ups could boost conversions and create real-world buzz.

2. Customer Retention:

   1. By releasing a new box each season tied to a Starbucks drink theme (e.g., Pumpkin Spice in Fall, Peppermint Mocha in Winter), the partnership creates a subscription cadence tied to customers' natural habits. Incentives like early access, loyalty rewards, or personalized item selection could encourage renewals.

3. Customer Profitability:

   1. Data from repeat purchases, seasonal engagement, and social sharing behavior would help identify high-value segments. Targeting these customers with personalized upsells—like full-size versions or exclusive seasonal drops—would increase lifetime value.

4. Customer Experience:

   1. The packaging and unboxing should evoke the Starbucks drink experience through color, scent, and product selection. Calming lavender aromas, pastel tones, and a short, relaxing product routine build emotional connection and delight.

Additional Analytics and Data Collection

• To strengthen customer insights and refine offerings, the following data should be gathered:

• Search trends for lavender-related skincare across platforms like Google Trends and TikTok

• Customer reviews from past calming or spring-themed products for language analysis

• Survey data on scent preferences and skincare habits by season

• Engagement data from Starbucks and L’Occitane’s social media and loyalty programs

Social listening could reveal rising seasonal ingredients or wellness trends that inform future boxes.

Calculations to Perform

• Customer Acquisition Cost (CAC) vs. Customer Lifetime Value (CLV) by channel

• Churn rate of seasonal subscribers

• Net Promoter Score (NPS) by box

• A/B test results for product pairings or scent intensity

• Sentiment analysis on user reviews and social comments related to the lavender theme

These metrics would guide what to keep, change, or expand in future boxes.