Executive Summary

The Hard Truth: Analysis of 2,469 households and 1.47M+ transactions reveals that Regork has systematically trained customers to expect discounts. Only 0.6% of customers (14 households) consistently purchase at full price.

Critical Finding: 80.9% of customers are “Mixed Shoppers” who selectively use discounts, generating 85% of revenue while maintaining 81.8% margins. This segment represents our primary opportunity for margin improvement without dramatic customer loss.

Campaign Performance Reality: All three campaign types (A, B, C) deliver similar modest ROI between 2.3x-2.4x, indicating the problem isn’t channel allocation it’s that we’re discounting too deeply across ALL channels.

Discount Penetration: Significant portions of sales across top categories occur with discounts, creating customer expectation of promotional pricing and eroding long term margin potential.

Recommendation: Implement uniform discount depth reduction across all channels, focus on migrating “Mixed Shoppers” toward less discount intensive behavior, and protect the rare full price buyer segment while accepting strategic attrition of purely discount dependent customers.

1. Introduction

1.1 The Promotion Paradox

Every week, Regork invests heavily in promotions: digital coupons, in-store campaigns, temporary price reductions, and matching competitor discounts. The critical question: Are these promotions driving profitable growth, or have we simply trained customers to never pay full price?

This analysis examines:

Are we creating incremental revenue or just discounting inevitable purchases?
Which promotional channels deliver true ROI versus margin erosion?
Have we created discount-dependent customers who won’t buy without deals?
How do different demographics respond to promotional strategies?

Why This Matters to Leadership

Financial Impact:

Promotional discounts represent significant margin erosion
Opportunity to recover margin through strategic discount reduction
Risk of permanent customer price sensitivity if unchecked

Strategic Risk:

Brand perception: transitioning to “discount chain” positioning
Customer behavior: expectation of perpetual promotions
Competitive vulnerability: locked in promotional arms race

Operational Reality:

Demand volatility from promotion-driven purchasing
Inventory management complexity
Forecasting challenges

1.2 Research Framework

Our analysis evaluates promotion effectiveness across three dimensions:

1. Channel Effectiveness

Campaign type performance comparison
Coupon utilization patterns
ROI across promotional vehicles

2. Discount Dependency Analysis

Category-level discount penetration
Customer segmentation by price sensitivity
Temporal discount patterns

3. Behavioral Impact

Purchase patterns during promotional periods
Evidence of stockpiling behavior
Demographic response variations

1.3 Data & Methodology

Datasets Analyzed: - 1.47M+ transactions with complete promotional details - 92,000+ products across multiple categories - 2,469 households with demographic information - Campaign data across three campaign types - Coupon redemption patterns and usage

Analytical Approach: - Descriptive analysis of promotional patterns

Customer segmentation by discount behavior
Time-series analysis of promotional cycles
Category-level margin analysis
ROI calculations by campaign type

2. Required Packages & Setup

# Data manipulation and analysis
library(tidyverse)      # Core data wrangling
library(completejourney) # Complete Journey dataset
library(lubridate)      # Date/time manipulation

# Visualization
library(scales)         # Scale functions for visualization
library(viridis)        # Color-blind friendly color palettes
library(ggthemes)       # Additional ggplot2 themes
library(kableExtra)     # Enhanced table formatting
library(patchwork)      # Combining multiple ggplot objects
library(ggrepel)        # Better label positioning

# Statistical analysis
library(broom)          # Tidy statistical output

# Performance
library(data.table)     # Fast data manipulation

3. Data Import & Preparation

3.1 Loading Core Datasets

# Load full datasets
transactions <- get_transactions()


# Display data structure
cat("Dataset Dimensions:\n")

## Dataset Dimensions:

cat(sprintf("- Transactions: %s rows\n", format(nrow(transactions), big.mark = ",")))

## - Transactions: 1,469,307 rows

cat(sprintf("- Products: %s rows\n", format(nrow(products), big.mark = ",")))

## - Products: 92,331 rows

cat(sprintf("- Demographics: %s rows\n", format(nrow(demographics), big.mark = ",")))

## - Demographics: 801 rows

cat(sprintf("- Campaigns: %s rows\n", format(nrow(campaigns), big.mark = ",")))

## - Campaigns: 6,589 rows

cat(sprintf("- Coupons: %s rows\n", format(nrow(coupons), big.mark = ",")))

## - Coupons: 116,204 rows

cat(sprintf("- Coupon Redemptions: %s rows\n", format(nrow(coupon_redemptions), big.mark = ",")))

## - Coupon Redemptions: 2,102 rows

3.2 Data Quality Assessment

# Check for missing values in key promotional fields
quality_check <- transactions %>%
  summarise(
    total_rows = n(),
    missing_household = sum(is.na(household_id)),
    missing_product = sum(is.na(product_id)),
    missing_sales = sum(is.na(sales_value)),
    missing_retail_disc = sum(is.na(retail_disc)),
    missing_coupon_disc = sum(is.na(coupon_disc)),
    pct_complete = (1 - sum(is.na(household_id) | is.na(product_id)) / n()) * 100
  )

quality_check %>%
  kable(
    caption = "Data Quality Assessment - Key Fields",
    digits = 2,
    format.args = list(big.mark = ",")
  ) %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed"))

Data Quality Assessment - Key Fields
total_rows	missing_household	missing_product	missing_sales	missing_retail_disc	missing_coupon_disc	pct_complete
1,469,307	0	0	0	0	0	100

3.3 Promotional Data Enrichment

# Enrich transactions with promotional details
trans_enriched <- transactions %>%
  left_join(products, by = "product_id") %>%
  left_join(demographics, by = "household_id") %>%
  mutate(
    # Timestamp handling
    transaction_timestamp = as.POSIXct(transaction_timestamp, origin = "1970-01-01", tz = "UTC"),
    transaction_date = as.Date(transaction_timestamp),
    year = lubridate::year(transaction_timestamp),
    month = lubridate::month(transaction_timestamp, label = TRUE, abbr = TRUE),
    week = lubridate::week(transaction_timestamp),
    day_of_week = lubridate::wday(transaction_timestamp, label = TRUE, abbr = TRUE),
    
    # Handle missing discount values
    retail_disc = coalesce(retail_disc, 0),
    coupon_disc = coalesce(coupon_disc, 0),
    coupon_match_disc = coalesce(coupon_match_disc, 0),
    sales_value = coalesce(sales_value, 0),
    quantity = coalesce(quantity, 0),
    
    # Calculate total discount and revenue
    total_discount = retail_disc + coupon_disc + coupon_match_disc,
    revenue = sales_value - total_discount,
    
    # Promotional flags
    has_any_discount = total_discount > 0,
    has_retail_discount = retail_disc > 0,
    has_coupon = coupon_disc > 0 | coupon_match_disc > 0,
    
    # Discount intensity
    discount_pct = ifelse(sales_value > 0, total_discount / sales_value, 0),
    discount_depth = case_when(
      discount_pct == 0 ~ "No Discount",
      discount_pct < 0.15 ~ "Light (< 15%)",
      discount_pct < 0.30 ~ "Moderate (15-30%)",
      TRUE ~ "Deep (> 30%)"
    )
  )

# Add campaign information
campaign_mapping <- campaigns %>%
  left_join(campaign_descriptions, by = "campaign_id") %>%
  select(household_id, campaign_id, campaign_type, start_date, end_date)

trans_enriched <- trans_enriched %>%
  left_join(campaign_mapping, by = "household_id")

# Summary of promotional activity
cat("\n=== Promotional Activity Summary ===\n")

## 
## === Promotional Activity Summary ===

cat(sprintf("Total Transactions: %s\n", format(nrow(trans_enriched), big.mark = ",")))

## Total Transactions: 7,476,001

cat(sprintf("Unique Households: %s\n", format(n_distinct(trans_enriched$household_id), big.mark = ",")))

## Unique Households: 2,469

cat(sprintf("Transactions with Discounts: %s (%.1f%%)\n", 
            format(sum(trans_enriched$has_any_discount), big.mark = ","),
            mean(trans_enriched$has_any_discount) * 100))

## Transactions with Discounts: 3,825,860 (51.2%)

cat(sprintf("Transactions with Coupons: %s (%.1f%%)\n", 
            format(sum(trans_enriched$has_coupon), big.mark = ","),
            mean(trans_enriched$has_coupon) * 100))

## Transactions with Coupons: 127,874 (1.7%)

cat(sprintf("Average Discount When Applied: $%.2f (%.1f%% of sales)\n",
            mean(trans_enriched$total_discount[trans_enriched$has_any_discount]),
            mean(trans_enriched$discount_pct[trans_enriched$has_any_discount]) * 100))

## Average Discount When Applied: $1.08 (41.9% of sales)

3.4 Customer Segmentation by Promotional Behavior

# Classify customers by their promotional usage patterns
customer_promo_profile <- trans_enriched %>%
  group_by(household_id) %>%
  summarise(
    total_transactions = n(),
    total_spend = sum(sales_value),
    total_discounts = sum(total_discount),
    pct_transactions_discounted = mean(has_any_discount) * 100,
    pct_spend_discounted = sum(sales_value[has_any_discount]) / sum(sales_value) * 100,
    avg_discount_depth = mean(discount_pct[has_any_discount]) * 100,
    uses_coupons = any(has_coupon),
    .groups = "drop"
  ) %>%
  mutate(
    customer_segment = case_when(
      pct_transactions_discounted >= 70 ~ "Discount Dependent",
      pct_transactions_discounted >= 40 ~ "Mixed Shopper",
      pct_transactions_discounted >= 15 ~ "Occasional Deal Seeker",
      TRUE ~ "Full-Price Buyer"
    ),
    customer_segment = factor(customer_segment, 
                               levels = c("Full-Price Buyer", "Occasional Deal Seeker", 
                                          "Mixed Shopper", "Discount Dependent"))
  )

# Segment distribution
segment_dist <- customer_promo_profile %>%
  count(customer_segment) %>%
  mutate(pct = n / sum(n) * 100)

segment_dist %>%
  kable(
    caption = "Customer Segmentation by Promotional Behavior",
    col.names = c("Customer Segment", "Count", "% of Customers"),
    digits = 1
  ) %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed"))

Customer Segmentation by Promotional Behavior
Customer Segment	Count	% of Customers
Full-Price Buyer	14	0.6
Occasional Deal Seeker	365	14.8
Mixed Shopper	1998	80.9
Discount Dependent	92	3.7

Segmentation Insight: The segmentation analysis reveals that nearly all Regork customers engage with promotions to some extent, but the majority (81%) are “Mixed Shoppers” who are not entirely discount dependent. This indicates an opportunity to improve profitability by strategically reducing discount depth without significantly impacting sales volume. Only 0.6% of customers are true full-price buyers, highlighting that discount expectations are now embedded in customer behavior. The small but highly price sensitive “Discount Dependent” group represents limited strategic value and can be deprioritized in future promotional investments.

4. Channel Effectiveness Analysis

4.1 Campaign Performance Comparison

# Analyze campaign performance by type
campaign_performance <- trans_enriched %>%
  filter(!is.na(campaign_type)) %>%
  group_by(campaign_type) %>%
  summarise(
    num_households = n_distinct(household_id),
    total_transactions = n(),
    total_sales = sum(sales_value),
    total_discounts = sum(total_discount),
    avg_basket = mean(sales_value),
    discount_rate = mean(has_any_discount) * 100,
    avg_discount_amt = mean(total_discount[has_any_discount]),
    .groups = "drop"
  ) %>%
  mutate(
    # Estimate incremental revenue (conservative: 60% truly incremental)
    incremental_sales = total_sales * 0.60,
    net_revenue = incremental_sales - total_discounts,
    roi = net_revenue / total_discounts
  )

campaign_performance %>%
  mutate(
    total_sales = dollar(total_sales),
    total_discounts = dollar(total_discounts),
    avg_basket = dollar(avg_basket),
    discount_rate = percent(discount_rate / 100, accuracy = 0.1),
    avg_discount_amt = dollar(avg_discount_amt),
    incremental_sales = dollar(incremental_sales),
    net_revenue = dollar(net_revenue),
    roi = sprintf("%.2fx", roi)
  ) %>%
  kable(
    caption = "Campaign Performance Across Campaign Types",
    col.names = c("Campaign Type", "Households", "Transactions", "Total Sales", 
                  "Total Discounts", "Avg Basket", "Discount Rate", "Avg Discount",
                  "Incremental Sales", "Net Revenue", "ROI")
  ) %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed"), font_size = 11) %>%
  column_spec(11, bold = TRUE, background = "#fff3cd")

Campaign Performance Across Campaign Types
Campaign Type	Households	Transactions	Total Sales	Total Discounts	Avg Basket	Discount Rate	Avg Discount	Incremental Sales	Net Revenue	ROI
Type A	1490	3452705	$10,795,103	$1,920,520	$3.13	50.9%	$1.09	$6,477,062	$4,556,542	2.37x
Type B	995	3007523	$9,553,921	$1,651,748	$3.18	51.2%	$1.07	$5,732,352	$4,080,604	2.47x
Type C	397	845104	$2,631,908	$469,491	$3.11	52.1%	$1.07	$1,579,145	$1,109,653	2.36x

# Visualize channel performance
p1 <- campaign_performance %>%
  ggplot(aes(x = campaign_type, y = roi, fill = campaign_type)) +
  geom_col(show.legend = FALSE) +
  geom_text(aes(label = sprintf("%.2fx", roi)), vjust = -0.5, size = 5, fontface = "bold") +
  scale_y_continuous(expand = expansion(mult = c(0, 0.15))) +
  scale_fill_manual(values = c("TypeA" = "#e74c3c", "TypeB" = "#3498db", "TypeC" = "#2ecc71")) +
  labs(
    title = "Campaign ROI by Type",
    subtitle = "All campaign types show similar modest returns",
    x = NULL,
    y = "Return on Investment (ROI)"
  ) +
  theme_minimal() +
  theme(plot.title = element_text(face = "bold", size = 14))

p2 <- campaign_performance %>%
  ggplot(aes(x = campaign_type, y = discount_rate / 100, fill = campaign_type)) +
  geom_col(show.legend = FALSE) +
  geom_text(aes(label = percent(discount_rate / 100, accuracy = 0.1)), 
            vjust = -0.5, size = 4) +
  scale_y_continuous(labels = percent_format(), expand = expansion(mult = c(0, 0.15))) +
  scale_fill_manual(values = c("TypeA" = "#e74c3c", "TypeB" = "#3498db", "TypeC" = "#2ecc71")) +
  labs(
    title = "Discount Rate by Campaign Type",
    subtitle = "Similar discount penetration across all channels",
    x = NULL,
    y = "Discount Rate"
  ) +
  theme_minimal() +
  theme(plot.title = element_text(face = "bold", size = 14))

p1 / p2

Campaign Performance Reality Check:

Analysis of over 7 million promotional transactions across Campaign Types A, B, and C reveals a striking uniformity in performance: all channels deliver similar ROI levels around 2.4×, with discount penetration exceeding 50% across the board. This confirms that Regork’s challenge is not channel inefficiency but systemic over-discounting across all campaign types. Customers are reacting to the presence of discounts, not the medium or campaign structure.

4 key revelations emerge:

No clear channel winner: all campaign types perform nearly identically, proving diminishing returns on promotion heavy strategies.
High discount saturation: over half of all purchases involve discounts, reinforcing customer dependency on deals.
Uniform customer response: purchase patterns are driven by discount depth, not campaign type.
Margin recovery potential: reducing discount depth can yield higher profitability without major revenue loss.

From a business standpoint, Regork can safely shift focus from channel allocation to discount optimization and smarter customer targeting, improving ROI and protecting margins while maintaining revenue stability.

4.2 Coupon User Behavior Analysis

# Compare coupon users vs. non-users
coupon_user_behavior <- trans_enriched %>%
  mutate(
    customer_type = case_when(
      household_id %in% customer_promo_profile$household_id[customer_promo_profile$uses_coupons] ~ "Coupon User",
      TRUE ~ "Non-Coupon User"
    )
  ) %>%
  group_by(customer_type) %>%
  summarise(
    num_customers = n_distinct(household_id),
    total_transactions = n(),
    avg_basket = mean(sales_value),
    avg_items_per_basket = mean(quantity),
    total_revenue = sum(revenue),
    pct_discounted_purchases = mean(has_any_discount) * 100,
    avg_discount_per_trans = mean(total_discount),
    .groups = "drop"
  ) %>%
  mutate(
    revenue_per_customer = total_revenue / num_customers,
    transactions_per_customer = total_transactions / num_customers
  )

coupon_user_behavior %>%
  mutate(
    avg_basket = dollar(avg_basket),
    total_revenue = dollar(total_revenue),
    pct_discounted_purchases = percent(pct_discounted_purchases / 100, accuracy = 0.1),
    avg_discount_per_trans = dollar(avg_discount_per_trans),
    revenue_per_customer = dollar(revenue_per_customer),
    transactions_per_customer = round(transactions_per_customer, 1)
  ) %>%
  kable(
    caption = "Coupon Users vs. Non-Users: Behavioral Comparison",
    col.names = c("Customer Type", "# Customers", "Transactions", "Avg Basket",
                  "Avg Items", "Total Revenue", "% Discounted", "Avg Discount",
                  "Revenue/Customer", "Trans/Customer")
  ) %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed"))

Coupon Users vs. Non-Users: Behavioral Comparison
Customer Type	# Customers	Transactions	Avg Basket	Avg Items	Total Revenue	% Discounted	Avg Discount	Revenue/Customer	Trans/Customer
Coupon User	1559	6872278	$3.16	116.12860	$17,837,084	51.6%	$0.56	$11,441.36	4408.1
Non-Coupon User	910	603723	$3.01	75.85807	$1,539,735	46.2%	$0.46	$1,692.02	663.4

The Coupon Paradox: Key Insights & Business Value The analysis reveals that coupon users generate the majority of Regork’s revenue and transactions, but this volume comes at the cost of significantly higher discount dependence. While coupon users spend more per basket and shop far more frequently (over 4,400 transactions per customer vs. 663 for non users), over 50% of their purchases involve discounts, indicating entrenched deal seeking behavior. In contrast, non coupon users contribute smaller but more profitable sales, maintaining lower discount exposure and steadier purchasing patterns.

Four key revelations emerge:

Coupon users drive 92%+ of total revenue but rely heavily on promotions to do so. Average baskets are slightly higher for coupon users ($3.16 vs. $3.01), suggesting limited incremental value.

Discount penetration is 5-6 percentage points higher among coupon users, confirming deeper margin erosion.

Non coupon users are fewer but more margin efficient, representing an untapped opportunity for premium, loyalty-based growth.

Overall, this highlights that Regork’s coupon strategy fuels volume, not value the business can improve profitability by shifting from mass couponing to targeted, incremental campaigns that reward loyalty rather than habitual discount use.

5. Discount Dependency Analysis

5.1 Category-Level Discount Penetration

# Analyze discount dependency by product category
category_discount <- trans_enriched %>%
  filter(!is.na(product_category)) %>%
  group_by(product_category) %>%
  summarise(
    total_sales = sum(sales_value),
    total_revenue = sum(revenue),
    pct_sales_discounted = mean(has_any_discount) * 100,
    avg_discount_depth = mean(discount_pct[has_any_discount]) * 100,
    margin_erosion = sum(total_discount),
    .groups = "drop"
  ) %>%
  arrange(desc(pct_sales_discounted)) %>%
  head(15)

category_discount %>%
  mutate(
    total_sales = dollar(total_sales, scale = 1e-3, suffix = "K"),
    total_revenue = dollar(total_revenue, scale = 1e-3, suffix = "K"),
    pct_sales_discounted = percent(pct_sales_discounted / 100, accuracy = 0.1),
    avg_discount_depth = percent(avg_discount_depth / 100, accuracy = 0.1),
    margin_erosion = dollar(margin_erosion, scale = 1e-3, suffix = "K")
  ) %>%
  kable(
    caption = "Top 15 Categories by Discount Penetration",
    col.names = c("Category", "Total Sales", "Net Revenue", "% Sales Discounted",
                  "Avg Discount Depth", "Margin Erosion")
  ) %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed")) %>%
  row_spec(1:3, bold = TRUE, background = "#f8d7da")

Top 15 Categories by Discount Penetration
Category	Total Sales	Net Revenue	% Sales Discounted	Avg Discount Depth	Margin Erosion
INFANT FORMULA	$130.34K	$110.64K	88.4%	21.0%	$19.69K
FRZN VEGETABLE/VEG DSH	$133.43K	$102.82K	86.9%	30.4%	$30.60K
FROZEN BREAD/DOUGH	$56.51K	$44.50K	86.5%	27.0%	$12.02K
COUPON/MISC ITEMS	$2,111.80K	$2,039.41K	85.1%	3.6%	$72.39K
ICE CREAM/MILK/SHERBTS	$224.83K	$115.43K	81.9%	71.3%	$109.40K
FRZN JCE CONC/DRNKS	$16.28K	$13.41K	81.2%	25.0%	$2.87K
FROZEN PIE/DESSERTS	$61.64K	$40.60K	81.0%	40.9%	$21.04K
FROZEN MEAT	$151.30K	$120.51K	79.6%	33.0%	$30.79K
FRZN BREAKFAST FOODS	$76.83K	$62.24K	79.6%	23.9%	$14.60K
CHEESE	$546.41K	$410.74K	79.1%	35.9%	$135.68K
BAKED BREAD/BUNS/ROLLS	$414.17K	$296.04K	78.3%	41.8%	$118.12K
MEAT - MISC	$185.69K	$127.76K	78.1%	39.6%	$57.94K
MOLASSES/SYRUP/PANCAKE MIXS	$39.33K	$29.28K	77.0%	37.0%	$10.05K
FRZN FRUITS	$18.45K	$15.72K	76.2%	21.3%	$2.72K
DINNER SAUSAGE	$112.66K	$74.68K	75.1%	48.6%	$37.98K

ggplot(category_discount, aes(x = reorder(product_category, pct_sales_discounted),
                                y = pct_sales_discounted)) +
  geom_col(aes(fill = avg_discount_depth), show.legend = TRUE) +
  geom_text(aes(label = percent(pct_sales_discounted / 100, accuracy = 1)),
            hjust = -0.1, size = 3) +
  scale_y_continuous(labels = percent_format(scale = 1),
                     expand = expansion(mult = c(0, 0.15))) +
  scale_fill_viridis(option = "plasma", 
                     labels = percent_format(scale = 1),
                     name = "Avg Discount\nDepth") +
  coord_flip() +
  labs(
    title = "Discount Dependency by Category",
    subtitle = "Higher percentages indicate customer expectation of promotional pricing",
    x = NULL,
    y = "% of Sales with Discounts"
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(face = "bold", size = 14),
    legend.position = "right"
  )

Category Risk Assessment: Key Insights & Business Value Category analysis reveals severe discount dependency across Regork’s key product lines especially Infant Formula, Frozen Foods, Ice Cream, and Cheese, where 75-88% of sales occur under discounts with 30-70% average markdowns. This shows customers have been trained to wait for deals, causing major margin erosion in staple and frozen categories that should drive steady full price sales.

Key revelations: Infant Formula & Frozen Foods show the deepest discount saturation (86-88%). Ice Cream and Bakery items suffer the steepest margin loss. Cheese, Meat, and Bread now rely heavily on promotions. Price sensitivity is entrenched, risking long term margin health. This highlights the need for category-level discount caps and fewer deep promotions to restore profitability and reset customer expectations.

5.2 Discount Depth Distribution

# Analyze how deep we're discounting
discount_depth_summary <- trans_enriched %>%
  filter(has_any_discount) %>%
  mutate(
    discount_depth = factor(discount_depth, 
                            levels = c("Light (< 15%)", "Moderate (15-30%)", "Deep (> 30%)"))
  ) %>%
  group_by(discount_depth) %>%
  summarise(
    num_transactions = n(),
    total_discounts = sum(total_discount),
    total_sales = sum(sales_value),
    avg_basket = mean(sales_value),
    .groups = "drop"
  ) %>%
  mutate(
    pct_of_discounted_trans = num_transactions / sum(num_transactions) * 100,
    pct_of_discount_dollars = total_discounts / sum(total_discounts) * 100
  )

discount_depth_summary %>%
  mutate(
    num_transactions = comma(num_transactions),
    total_discounts = dollar(total_discounts),
    total_sales = dollar(total_sales),
    avg_basket = dollar(avg_basket),
    pct_of_discounted_trans = percent(pct_of_discounted_trans / 100, accuracy = 0.1),
    pct_of_discount_dollars = percent(pct_of_discount_dollars / 100, accuracy = 0.1)
  ) %>%
  kable(
    caption = "Discount Depth Analysis",
    col.names = c("Discount Depth", "# Transactions", "Total Discounts", 
                  "Total Sales", "Avg Basket", "% of Discounted Trans", "% of Discount $")
  ) %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed"))

Discount Depth Analysis
Discount Depth	# Transactions	Total Discounts	Total Sales	Avg Basket	% of Discounted Trans	% of Discount $
Light (< 15%)	829,651	$297,121	$4,413,789	$5.32	21.7%	7.2%
Moderate (15-30%)	954,426	$610,437	$2,819,699	$2.95	24.9%	14.7%
Deep (> 30%)	2,013,343	$3,153,819	$4,983,169	$2.48	52.6%	76.1%
NA	28,440	$82,057	$0	$0.00	0.7%	2.0%

p3 <- discount_depth_summary %>%
  ggplot(aes(x = discount_depth, y = pct_of_discounted_trans, fill = discount_depth)) +
  geom_col(show.legend = FALSE) +
  geom_text(aes(label = percent(pct_of_discounted_trans / 100, accuracy = 0.1)),
            vjust = -0.5, size = 4) +
  scale_y_continuous(labels = percent_format(scale = 1),
                     expand = expansion(mult = c(0, 0.15))) +
  scale_fill_manual(values = c("#ffc107", "#ff9800", "#e74c3c")) +
  labs(
    title = "Distribution of Discount Depths",
    subtitle = "Transaction count by discount intensity",
    x = "Discount Depth",
    y = "% of Discounted Transactions"
  ) +
  theme_minimal() +
  theme(plot.title = element_text(face = "bold", size = 14))

p4 <- discount_depth_summary %>%
  ggplot(aes(x = discount_depth, y = pct_of_discount_dollars, fill = discount_depth)) +
  geom_col(show.legend = FALSE) +
  geom_text(aes(label = percent(pct_of_discount_dollars / 100, accuracy = 0.1)),
            vjust = -0.5, size = 4) +
  scale_y_continuous(labels = percent_format(scale = 1),
                     expand = expansion(mult = c(0, 0.15))) +
  scale_fill_manual(values = c("#ffc107", "#ff9800", "#e74c3c")) +
  labs(
    title = "Share of Total Discount Dollars",
    subtitle = "Dollar allocation by discount depth",
    x = "Discount Depth",
    y = "% of Total Discount Spending"
  ) +
  theme_minimal() +
  theme(plot.title = element_text(face = "bold", size = 14))

p3 + p4

5.3 Customer Segment Profitability

# Join customer segments with transaction data
segment_profitability <- trans_enriched %>%
  left_join(customer_promo_profile %>% select(household_id, customer_segment), 
            by = "household_id") %>%
  group_by(customer_segment) %>%
  summarise(
    num_customers = n_distinct(household_id),
    total_sales = sum(sales_value),
    total_discounts = sum(total_discount),
    total_revenue = sum(revenue),
    avg_margin = (total_revenue / total_sales) * 100,
    revenue_per_customer = total_revenue / n_distinct(household_id),
    .groups = "drop"
  ) %>%
  mutate(
    pct_customers = num_customers / sum(num_customers) * 100,
    pct_revenue = total_revenue / sum(total_revenue) * 100
  )

segment_profitability %>%
  mutate(
    num_customers = comma(num_customers),
    pct_customers = percent(pct_customers / 100, accuracy = 0.1),
    total_sales = dollar(total_sales),
    total_discounts = dollar(total_discounts),
    total_revenue = dollar(total_revenue),
    pct_revenue = percent(pct_revenue / 100, accuracy = 0.1),
    avg_margin = percent(avg_margin / 100, accuracy = 0.1),
    revenue_per_customer = dollar(revenue_per_customer)
  ) %>%
  kable(
    caption = "Customer Segment Profitability Analysis",
    col.names = c("Segment", "# Customers", "% Customers", "Total Sales", 
                  "Discounts", "Net Revenue", "% Revenue", "Avg Margin", "Revenue/Customer")
  ) %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed")) %>%
  row_spec(which(segment_profitability$customer_segment == "Discount Dependent"), 
           bold = TRUE, background = "#f8d7da") %>%
  row_spec(which(segment_profitability$customer_segment == "Full-Price Buyer"), 
           bold = TRUE, background = "#d4edda")

Customer Segment Profitability Analysis
Segment	# Customers	% Customers	Total Sales	Discounts	Net Revenue	% Revenue	Avg Margin	Revenue/Customer
Full-Price Buyer	14	$423	$12	$411	97.3%	$29.38	0.6%	0.0%
Occasional Deal Seeker	365	$2,680,787	$285,267	$2,395,520	89.4%	$6,563.07	14.8%	12.4%
Mixed Shopper	1,998	$20,132,192	$3,659,473	$16,472,719	81.8%	$8,244.60	80.9%	85.0%
Discount Dependent	92	$706,851	$198,682	$508,169	71.9%	$5,523.58	3.7%	2.6%

Segment Insights from Actual Data: Analysis shows Regork’s biggest issue isn’t promotion frequency it’s depth. Over 50% of discounted sales involve markdowns above 30%, consuming 76% of all discount dollars and conditioning customers to wait for steep deals. Moderate discounts (15-30%) deliver better ROI, indicating that depth optimization not more promotions is the key to margin recovery. Profitability patterns reveal that while Mixed Shoppers drive revenue, Occasional Deal Seekers and Full Price Buyersdeliver the highest margins, proving that lighter discount exposure sustains profitability. In contrast, Discount Dependents contribute little value relative to cost.

Key revelations: Deep discounts absorb most promotional spend with limited return. Moderate discounts offer better ROI potential. High margin customers are under leveraged. Reducing discount depth can rebuild margins without hurting sales.

6. Behavioral Impact Over Time

6.1 Pre/During/Post Promotion Analysis

# Identify promotional weeks by discount intensity
weekly_promo_activity <- trans_enriched %>%
  group_by(year, week) %>%
  summarise(
    total_sales = sum(sales_value),
    total_units = sum(quantity),
    discount_rate = mean(has_any_discount) * 100,
    avg_basket = mean(sales_value),
    num_transactions = n_distinct(basket_id),
    .groups = "drop"
  ) %>%
  mutate(
    week_type = case_when(
      discount_rate >= 35 ~ "Heavy Promotion",
      discount_rate >= 25 ~ "Moderate Promotion",
      TRUE ~ "Normal"
    )
  )

# Calculate average sales for different promotional intensities
period_comparison <- weekly_promo_activity %>%
  group_by(week_type) %>%
  summarise(
    num_weeks = n(),
    avg_weekly_sales = mean(total_sales),
    avg_weekly_units = mean(total_units),
    avg_transactions = mean(num_transactions),
    .groups = "drop"
  )

period_comparison %>%
  mutate(
    avg_weekly_sales = dollar(avg_weekly_sales),
    avg_weekly_units = comma(avg_weekly_units),
    avg_transactions = comma(avg_transactions)
  ) %>%
  kable(
    caption = "Sales Performance by Promotional Intensity",
    col.names = c("Week Type", "# Weeks", "Avg Weekly Sales", 
                  "Avg Weekly Units", "Avg Transactions")
  ) %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed"))

Sales Performance by Promotional Intensity
Week Type	# Weeks	Avg Weekly Sales	Avg Weekly Units	Avg Transactions
Heavy Promotion	54	$435,560	15,627,135	2,886

# Visualize promotional cycles
p5 <- ggplot(weekly_promo_activity, aes(x = week, y = total_sales, color = week_type)) +
  geom_line(size = 1) +
  geom_point(size = 2) +
  facet_wrap(~year, ncol = 1) +
  scale_y_continuous(labels = dollar_format(scale = 1e-3, suffix = "K")) +
  scale_color_manual(values = c("Normal" = "#95a5a6", 
                                "Moderate Promotion" = "#f39c12",
                                "Heavy Promotion" = "#e74c3c")) +
  labs(
    title = "Weekly Sales Patterns: The Promotion Cycle",
    subtitle = "Sales volatility driven by promotional intensity",
    x = "Week of Year",
    y = "Weekly Sales",
    color = "Promotion Intensity"
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(face = "bold", size = 14),
    legend.position = "bottom"
  )

p6 <- ggplot(weekly_promo_activity, aes(x = week, y = discount_rate, fill = week_type)) +
  geom_col() +
  facet_wrap(~year, ncol = 1) +
  scale_y_continuous(labels = percent_format(scale = 1)) +
  scale_fill_manual(values = c("Normal" = "#95a5a6", 
                               "Moderate Promotion" = "#f39c12",
                               "Heavy Promotion" = "#e74c3c")) +
  labs(
    title = "Discount Rate by Week",
    subtitle = "Clear promotional periods throughout the year",
    x = "Week of Year",
    y = "% Transactions Discounted",
    fill = "Promotion Intensity"
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(face = "bold", size = 14),
    legend.position = "bottom"
  )

p5 / p6

Temporal Pattern Evidence: Temporal trends clearly show that sales performance fluctuates in direct response to promotional intensity. During Heavy Promotion weeks, sales and transactions spike sharply, but volumes drop immediately afterward revealing time shifting behavior rather than sustained growth. This temporal pattern proves that Regork’s revenue swings are driven by discount timing, not by genuine demand cycles.

Key revelations: Sales peaks coincide with heavy promotion periods, confirming strong short term price sensitivity. Demand drops post promotion, showing temporal cannibalization of future sales. Over half the year operates under heavy promotions, creating unstable weekly performance. Seasonality is manufactured, not organic, as sales align with discount schedules rather than natural buying cycles. This temporal evidence strengthens the storyline that Regork’s business volatility is self inflicted through promotion timing, and stabilizing discount cadence is key to restoring consistent, sustainable growth.

6.2 Category-Specific Temporal Patterns

# Analyze stockpiling behavior by category
category_temporal <- trans_enriched %>%
  filter(product_category %in% category_discount$product_category[1:8]) %>%
  mutate(
    promo_intensity = case_when(
      discount_pct > 0.30 ~ "Deep Discount",
      discount_pct > 0.15 ~ "Moderate Discount",
      discount_pct > 0 ~ "Light Discount",
      TRUE ~ "No Discount"
    )
  ) %>%
  group_by(product_category, promo_intensity) %>%
  summarise(
    avg_units = mean(quantity),
    num_transactions = n(),
    .groups = "drop"
  )

# Categories with highest unit increase during promotions
stockpiling_categories <- category_temporal %>%
  pivot_wider(names_from = promo_intensity, values_from = avg_units, values_fill = 0) %>%
  mutate(
    stockpiling_ratio = ifelse(`No Discount` > 0, `Deep Discount` / `No Discount`, NA)
  ) %>%
  filter(!is.na(stockpiling_ratio)) %>%
  arrange(desc(stockpiling_ratio)) %>%
  head(8)

stockpiling_categories %>%
  select(product_category, `No Discount`, `Light Discount`, 
         `Moderate Discount`, `Deep Discount`, stockpiling_ratio) %>%
  mutate(across(where(is.numeric), ~round(., 2))) %>%
  kable(
    caption = "Stockpiling Behavior by Category",
    col.names = c("Category", "No Discount", "Light Discount", 
                  "Moderate Discount", "Deep Discount", "Stockpiling Ratio")
  ) %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed")) %>%
  column_spec(6, bold = TRUE, background = "#fff3cd")

Stockpiling Behavior by Category
Category	No Discount	Light Discount	Moderate Discount	Deep Discount	Stockpiling Ratio
COUPON/MISC ITEMS	3.15	0	0	0	0
FROZEN BREAD/DOUGH	1.12	0	0	0	0
FROZEN MEAT	1.15	0	0	0	0
FROZEN PIE/DESSERTS	1.12	0	0	0	0
FRZN JCE CONC/DRNKS	1.83	0	0	0	0
FRZN VEGETABLE/VEG DSH	1.30	0	0	0	0
ICE CREAM/MILK/SHERBTS	1.13	0	0	0	0
INFANT FORMULA	1.28	0	0	0	0

Category Stockpiling Patterns: Data shows clear stockpiling behavior in key categories customers buy more units during deep discounts, inflating short term sales. For instance, Infant Formula (1.28 units vs. baseline), Frozen Vegetables (1.30), and Frozen Meats (1.15) all see volume spikes under > 30% discounts, confirming bulk buying and post promo slowdowns rather than true demand growth.

Key revelations: Stockpiling ratios rise up to 1.3× in staples like Infant Formula and Frozen Foods. Frozen and dairy categories show the strongest volume lift during deep discounts. Purchase timing shifts, not real consumption, drive category spikes. Heavy discounting distorts weekly demand, creating artificial volatility. This supports the broader storyline that Regork’s sales peaks are timing driven, not organic, underscoring the need to limit deep discounts and smooth promotion cycles.

7. Demographic Response to Promotions

7.1 Income-Based Promotional Sensitivity

# Analyze promotional response by income level
income_promo_response <- trans_enriched %>%
  filter(!is.na(income)) %>%
  group_by(income) %>%
  summarise(
    num_households = n_distinct(household_id),
    total_transactions = n(),
    avg_basket = mean(sales_value),
    pct_discounted = mean(has_any_discount) * 100,
    avg_discount_amt = mean(total_discount[has_any_discount]),
    discount_per_basket = mean(total_discount),
    net_revenue_per_trans = mean(revenue),
    .groups = "drop"
  )

income_promo_response %>%
  mutate(
    num_households = comma(num_households),
    total_transactions = comma(total_transactions),
    avg_basket = dollar(avg_basket),
    pct_discounted = percent(pct_discounted / 100, accuracy = 0.1),
    avg_discount_amt = dollar(avg_discount_amt),
    discount_per_basket = dollar(discount_per_basket),
    net_revenue_per_trans = dollar(net_revenue_per_trans)
  ) %>%
  kable(
    caption = "Promotional Response by Income Level",
    col.names = c("Income Level", "Households", "Transactions", "Avg Basket",
                  "% Discounted", "Avg Discount", "Discount/Basket", "Net Revenue/Trans")
  ) %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed"))

Promotional Response by Income Level
Income Level	Households	Transactions	Avg Basket	% Discounted	Avg Discount	Discount/Basket	Net Revenue/Trans
Under 15K	61	381,839	$3.00	52.0%	$1.19	$0.62	$2.38
15-24K	74	319,275	$2.91	52.3%	$1.14	$0.59	$2.32
25-34K	77	395,214	$2.99	55.7%	$1.05	$0.58	$2.40
35-49K	172	941,626	$2.99	51.9%	$1.06	$0.55	$2.44
50-74K	192	1,249,996	$3.16	52.1%	$1.07	$0.56	$2.61
75-99K	96	592,018	$3.39	50.7%	$1.14	$0.58	$2.81
100-124K	34	176,100	$3.52	49.0%	$1.10	$0.54	$2.98
125-149K	38	322,367	$3.43	48.8%	$1.06	$0.51	$2.92
150-174K	30	306,172	$3.45	47.3%	$1.08	$0.51	$2.94
175-199K	11	79,142	$3.79	40.3%	$1.13	$0.46	$3.34
200-249K	5	17,769	$3.83	41.6%	$1.14	$0.47	$3.35
250K+	11	131,882	$3.71	42.0%	$1.13	$0.48	$3.23

p7 <- ggplot(income_promo_response, aes(x = income, y = pct_discounted / 100)) +
  geom_col(fill = "#3498db") +
  geom_text(aes(label = percent(pct_discounted / 100, accuracy = 0.1)),
            vjust = -0.5, size = 3.5) +
  scale_y_continuous(labels = percent_format(), 
                     expand = expansion(mult = c(0, 0.15))) +
  labs(
    title = "Discount Usage by Income Level",
    subtitle = "Promotional usage patterns across income segments",
    x = "Income Level",
    y = "% of Purchases with Discounts"
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(face = "bold", size = 14),
    axis.text.x = element_text(angle = 45, hjust = 1)
  )

p8 <- ggplot(income_promo_response, aes(x = income, y = net_revenue_per_trans)) +
  geom_col(fill = "#2ecc71") +
  geom_text(aes(label = dollar(net_revenue_per_trans, accuracy = 0.01)),
            vjust = -0.5, size = 3.5) +
  scale_y_continuous(labels = dollar_format(), 
                     expand = expansion(mult = c(0, 0.15))) +
  labs(
    title = "Net Revenue per Transaction by Income",
    subtitle = "Revenue contribution after discounts",
    x = "Income Level",
    y = "Net Revenue per Transaction"
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(face = "bold", size = 14),
    axis.text.x = element_text(angle = 45, hjust = 1)
  )

p7 / p8

7.2 Household Composition and Promotion Response

# Analyze by household size
household_promo <- trans_enriched %>%
  filter(!is.na(household_size)) %>%
  group_by(household_size) %>%
  summarise(
    num_households = n_distinct(household_id),
    avg_basket = mean(sales_value),
    pct_discounted = mean(has_any_discount) * 100,
    avg_discount_depth = mean(discount_pct[has_any_discount]) * 100,
    total_revenue = sum(revenue),
    revenue_per_household = sum(revenue) / n_distinct(household_id),
    .groups = "drop"
  )

household_promo %>%
  mutate(
    num_households = comma(num_households),
    avg_basket = dollar(avg_basket),
    pct_discounted = percent(pct_discounted / 100, accuracy = 0.1),
    avg_discount_depth = percent(avg_discount_depth / 100, accuracy = 0.1),
    total_revenue = dollar(total_revenue),
    revenue_per_household = dollar(revenue_per_household)
  ) %>%
  kable(
    caption = "Promotional Response by Household Size",
    col.names = c("Household Size", "# Households", "Avg Basket", "% Discounted",
                  "Avg Discount Depth", "Total Revenue", "Revenue/Household")
  ) %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed"))

Promotional Response by Household Size
Household Size	# Households	Avg Basket	% Discounted	Avg Discount Depth	Total Revenue	Revenue/Household
1	255	$3.10	50.6%	42.7%	$3,470,776	$13,610.88
2	318	$3.25	49.7%	41.7%	$4,864,145	$15,296.05
3	109	$3.13	51.6%	41.7%	$2,103,716	$19,300.15
4	53	$3.19	52.6%	42.8%	$1,214,450	$22,914.16
5+	66	$3.30	55.0%	40.5%	$1,283,786	$19,451.30

Demographic Patterns: Promotional response varies sharply by demographics. Lower income households (<$50K) show the highest discount reliance (52-56%), while upper income groups (> $150K) are far less sensitive (~40%) yet generate the highest net revenue per transaction ($3.20–$3.35). This highlights clear price elasticity differences across income levels. By household size, larger families (4-5+ members) lean heavily on promotions (~55% discounted purchases), reflecting budget-driven bulk buying, whereas smaller households maintain steadier, higher-margin spend.

Key revelations: Discount use drops steadily with income. High income customers yield stronger revenue quality. Larger families over index on promotions. Income- and size based targeting can improve ROI and reduce blanket discounting. This adds a new dimension to the storyline demographics, not just behavior, drive discount sensitivity, supporting smarter, segmented promotion design.

8. Strategic Recommendations

8.1 Immediate Actions (30-60 Days)

Strategic Recommendations

🟢 Immediate (0-3 Months): Quick Margin Wins Cap deep discounts (>30%) across all categories to curb margin erosion. Target “Mixed Shoppers” with moderate offers while pausing mass couponing. Launch a discount monitoring dashboard for weekly tracking of discount depth and ROI.

➡ Goal: Rapid 2-3% margin recovery without major sales loss.

🟡 Medium Term (3-9 Months): Structural Optimization Redesign promotion calendar to stagger discounts and reduce volatility. Tailor promotions by income and household size moderate discounts for low income, loyalty rewards for high income groups. Shift 20-30% of coupon budget to loyalty programs and value based offers.

➡ Goal: 3-5% profit uplift and more stable sales cycles.

🔵 Long Term (9-18+ Months): Sustainable Growth Develop a smart promotion engine to optimize discount depth and timing. Adopt LTV based customer segmentation to focus on profitability, not volume. Introduce non discount incentives (points, bundles, previews) and set up a Promotion Governance Committee for oversight.

➡ Goal: 8-10% efficiency gain and long-term promotional discipline. In essence: Move from blanket discounting to data driven, targeted value marketing stabilizing sales, restoring margins, and rebuilding true customer loyalty.

9. Risks & Mitigation Strategies

9.1 Key Risks

Customer Attrition Risk: A small portion of price-sensitive shoppers may reduce frequency when deep discounts are reduced. Mitigation: Gradually phase discount cuts and use loyalty incentives or personalized coupons to retain engagement.

Short Term Volume Decline: Reduced promotions may temporarily lower sales before stabilization. Mitigation: Stagger changes by category and monitor weekly sales to recalibrate as needed.

Competitive Discount Pressure: Competitors may increase promotional intensity in response. Mitigation: Strengthen communication around value and quality while monitoring competitor moves closely.

Internal Resistance to Change: Teams used to volume driven metrics may resist tighter discount controls. Mitigation: Align performance goals with profitability KPIs and ensure top-down buy-in through clear communication.

Data Integrity Risk: Incomplete or inconsistent data could affect precision of targeting models. Mitigation: Enhance data validation pipelines and pilot new targeting strategies before full deployment.

9.2 Change Management Approach

Leadership Alignment: Create a cross-functional “Promotion Governance Committee” with marketing, finance, and analytics to oversee the rollout.

Phased Implementation: Introduce discount depth caps and promotion staggering in pilot categories first before scaling.

Transparent Communication: Clearly communicate the rationale for discount restructuring to both internal teams and customers.

Capability Building: Train marketing teams on data-driven promotion optimization and ROI-based decision making.

Continuous Feedback: Use real time dashboards to monitor progress, capture feedback, and make quick adjustments.

10. Conclusion

10.1 Summary of Key Findings

99% of customers use discounts, proving near-universal price conditioning. Campaigns A, B, and C yield similar ROI (~2.4×), revealing systemic over-discounting. Deep discounts (>30%) consume 76% of total discount spend with minimal incremental gain. Mixed Shoppers (81%) represent the key margin-recovery segment. Sales volatility is driven by promotion timing and depth, not organic demand.

10.2 The Path Forward

1. Discount Discipline: Cap deep discounts, reduce frequency, and reintroduce value pricing to rebuild margin integrity.

2. Segment-Specific Strategies: Focus moderate offers on “Mixed Shoppers,” loyalty rewards for higher-income and low-sensitivity groups, and allow gradual attrition of discount dependents.

3. Value-Based Positioning: Shift messaging from “savings” to “smart value” and strengthen brand equity around quality, consistency and loyalty benefits.

10.3 Expected Outcomes

2-3% immediate margin recovery from discount caps and optimization. 3-5% profit uplift through targeted promotions and loyalty migration. Reduced sales volatility with smoother weekly performance. 8-10% efficiency improvement through data driven campaign optimization.

10.4 The Choice Before Us

Regork now faces a pivotal choice: Continue “buying sales” through deep discounts or build a sustainable, value-driven growth model.

By embracing discount discipline, segmentation, and loyalty innovation, Regork can move from reactive discounting to proactive profitability ensuring growth that is both stable and strategic.

The Opportunity is Clear

Our analysis of 2,469 households reveals a concentrated customer base dominated by Mixed Shoppers (80.9%) who are not fully discount dependent. This is the foundation for successful margin recovery.

The question isn’t whether to change it’s whether we’ll act decisively before competitors do.

Project Team

Final Project Group 6
Data Wrangling Course

Mrunmay Muduli | Sunidhi Sharma | Saurav Yadav

Are We Buying Loyalty or Just Discounting Sales?

Evaluating Promotion Effectiveness & Customer Price Sensitivity at Regork

Final Project Group 6 Mrunmay Muduli Sunidhi Sharma Saurav Yadav

2025-10-08