Instacart

Analyzing Baskets. Predicting Behavior.

Fast Facts for Busy Reviewers

Goal:
Understand when, what, and how customers shop to improve loyalty, recommendations, and marketing strategies.

Tools Used:
Python (Pandas, Seaborn, Matplotlib) • Jupyter Notebook • Excel

Key Wins:

🕙 Peak shopping hours = 10 AM to 4 PM
🔁 5% of users account for over 50% of all orders
👨‍👩‍👧 Families make up the largest shopping group

Strategic Takeaways:

Launch promos before the weekend to catch peak traffic
Create loyalty bundles for frequent shoppers
Push premium items to high-spend, low-frequency users

Explore the Code Behind the Insights

GitHub

The Challenge

Instacart wanted to better understand its customer base, what people are buying, when they’re shopping, and how the platform could improve recommendations and build loyalty.

So, I dug into real-world order data to explore shopping patterns, segment different types of customers, and uncover what really drives repeat purchases.

Problem

Help Instacart optimize product suggestions and promotions by analyzing user behavior and shopping frequency.

Tools Used

To clean, process, and analyze the data, the following tools were used:

Python (Pandas, Matplotlib, Seaborn)
Jupyter Notebook
Excel

Key Insights

Weekend orders spike especially on Saturdays

A small group of loyal customers (top 5%) drives most of the repeat orders
High spenders don't necessarily order often

Business Suggestions

Create rewards programs for loyal shoppers

Send weekend promos based on peak activity
Offer high-spending users incentives to increase frequency

Data Cleaning Highlights

Before diving into analysis, I made sure the dataset was in good shape. Here’s what I cleaned up:

Fixed column names and standardized formatting
Removed duplicates and checked for missing values
Reclassified variables like spending flags and loyalty segments
Created new flags to help group customers by behavior

🥕 I used .isnull() to check for missing data across orders and product tables, making sure nothing slipped through before moving forward with analysis.

🥕 This before/after table shows how I handled common issues like inconsistent column names, null values, and duplicate rows.

🥕 Visual Tip

I included simple visuals like this to highlight transformations in a clear way no need to guess what changed behind the scenes.

Snapshot of Key Cleaning Improvements

Analytical Build

This analysis went way beyond filtering rows or cleaning columns. I built a streamlined dataset from scratch by merging raw Instacart files and creating new features to actually understand how people shop. Every step I took helped bring more clarity into what customers were buying, when they bought it, and what patterns repeated across time.

🥕 Clustering by Order Type

Not every customer shops the same way. Some stock up weekly like clockwork. Others show up once and vanish. I wanted to dig into those patterns so I grouped users based on how often they reordered, when they shopped, and how many orders they placed overall.

By clustering order behavior, I could better understand who’s already likely to come back and who might need a little nudge.

🥕 Engineered Features

After building a clean, streamlined dataset, I started engineering new features to actually reflect how people shop. I grouped prices into spending tiers, mapped peak shopping hours, and labeled the busiest days of the week. These added layers turned generic rows into real behavior patterns showing not just what people bought, but when, how often, and how much it mattered.

Created Variables:

price_label – grouped product prices into Low, Mid, and High tiers
busiest_day – labeled each day of the week by how many orders it held
order_hour_of_day – helped pinpoint when orders peaked during the day
busiest_period_of_day – clustered order hours into Morning, Afternoon, and Evening

Code Pulled From:
4.7 Deriving New Variables.ipynb

🥕 Order Rhythm Patterns

To understand when customers actually shop, I broke order times into hours, days, and busiest periods. Weekends were big, but it turns out weekday afternoons also saw repeat activity. Orders peaked midday, but there was a clear drop-off after 6 p.m. These timing patterns help identify the most responsive windows for campaigns or feature launches.

Created Timing Insights:

order_hour_of_day – shows when users placed orders throughout the day
busiest_day – ranks days of the week by order volume
busiest_period_of_day – classifies hours into Morning, Afternoon, and Evening blocks

Code Pulled From:
4.7 Deriving New Variables.ipynb
4.9 Data Visualization with Python.ipynb

These engineered features made it possible to move past surface-level metrics and dig into the real story behind shopping behavior. Without them, the insights wouldn’t have had the same clarity or impact and definitely wouldn’t have translated into anything actionable.

Want to see the full build in action?
Jupyter Notebooks on GitHub

Key Visuals & What They Show

🥕 Orders by Day of the Week

Most orders were placed on weekends, especially Sundays (~85,000 orders) and Mondays (~78,000 orders). This tells us promotions and ads should go live by Friday night to catch peak weekend traffic.

🥕 Customer Loyalty:

After filtering out inactive users (those with fewer than 5 orders), I found that 62.5% of returning customers placed more than 10 orders. That’s over 1.4 million regular shoppers, making this group ideal for investing in when it comes to loyalty programs and personalized offers.

🥕 Regional Shopping Patterns:

Shoppers in the Midwest ($12.72) and South ($12.25) regions spend the most on average, while "Other" regions and the West lag behind with lower spending (around $11.30). These insights can help Instacart prioritize premium promotions in high-spending areas and offer discount-based incentives where needed.

🥕 Family-Oriented Shoppers :

Couples with 1–2 dependents made up the largest household group at 1.3 million users, while both Singles (650K) and Families with 3+ dependents (657K) also represented sizable segments. Household size clearly shapes product needs. Larger families may be looking for bulk or family-friendly deals, while smaller households might prioritize convenience and flexibility.

Main Insights (Recap)

🥕 Big Spenders = Frequent Shoppers

Around 30% of high-spending users placed fewer than 5 orders, highlighting a major opportunity for Instacart to re-engage these customers through incentives or premium loyalty perks. These shoppers spend more per order but shop less frequently and that’s a gap worth closing.

🥕 Weekends = Peak Order Days

Sunday and Monday had the highest order volumes, with over 84,000 and 78,000 orders respectively. In contrast, Friday and Saturday saw fewer than 60,000 orders. This confirms that the weekend rush starts early, and promotions should go live by Friday night to capture peak traffic.

🥕 Repeat Shoppers = Loyalty Potential

40.9% of orders came from repeat users, and among returning customers, 62.5% placed more than 10 orders. That’s a loyal audience worth prioritizing in retention strategies like rewards programs and personalized deals.

🥕 Staples = Stickiness

Produce (Dept 4) led with 7.5 million products ordered, followed by Dairy & Eggs (Dept 16) at 4 million making them the two most-purchased departments. These fresh essentials play a key role in driving return visits and reinforcing loyalty. Their consistent demand makes them ideal targets for retention promos and personalized product recommendations.

🥕 Regional Spending Varies

Customers in the Midwest ($12.72) and South ($12.25) spend the most per order, indicating strong potential for premium product promotions in these regions. In contrast, the West ($11.32) and Other regions ($11.38) show lower average spending, suggesting they might respond better to discount-based strategies. These insights support a localized marketing approach, tailored to regional spending behaviors.

From Data to Decisions

This analysis moved beyond baskets and orders to understand how people actually shop over time. By connecting purchasing behavior with frequency, timing, and category patterns, the data surfaced clear opportunities to improve retention, engagement, and promotion strategy.

Business Decisions Informed by the Data

04. Time promotions around natural shopping peaks
Order volume consistently surged over weekends. Launching promotions by Friday evening increases visibility when customers are already primed to shop.

❯❯❯❯ Meet customers when intent is highest, not randomly.

02. Treat frequent shoppers as the retention core
A large segment of customers placed 20 or more orders. These shoppers drive repeat revenue and long-term value, making loyalty programs and personalized offers far more effective than broad discounts.

❯❯❯❯ Retention grows fastest when loyalty is rewarded intentionally.

01. Anchor promotions around staple categories
Departments like Produce and Dairy & Eggs appeared in most baskets. Featuring these staples in promotions increases relevance and keeps offers aligned with everyday needs.

❯❯❯❯ Promotions work best when they support routine behavior.

03. Re-engage high-spend, low-frequency customers
A meaningful portion of high spenders ordered fewer than five times. These customers already see value, but lack habit. Targeted perks, bundles, or subscription-style incentives can reduce friction and bring them back.

❯❯❯❯ Revenue opportunity often sits in behavior gaps, not new users.

What This Changes

Instead of treating all shoppers the same, this reframes growth as a behavioral design problem. Timing, frequency, and habit matter more than blanket discounts.

Better decisions come from understanding how people shop, not just what they buy.