The Interview Question

"You notice that revenue dropped 15% week-over-week. Walk me through how you would investigate and diagnose the root cause."

This is a classic structured problem-solving question that tests your analytical thinking, business acumen, and ability to work under pressure.

Why Companies Ask This

Interviewers are evaluating:

1. Structured Thinking — Can you decompose a complex problem systematically?
2. Business Knowledge — Do you understand revenue drivers?
3. Hypothesis-Driven Approach — Can you form and test hypotheses efficiently?
4. Data Literacy — Do you know which data to look at and when?
5. Communication Under Pressure — Can you explain your thinking clearly?

The Revenue Investigation Framework

Phase 1: Clarify & Scope (2 minutes)

Before diving into data, clarify:

• What metric dropped? (Gross revenue? Net revenue? By product line?)
• Time period? (When did it start? Is it a single event or trend?)
• Segmentation? (Is it across all users or specific segments?)
• External factors? (Holidays? Competitor actions? Economic events?)

Phase 2: Decompose & Segmentation (5 minutes)

Break the revenue metric into its components:

Revenue = Users × Conversion Rate × Average Order Value × Frequency

Or for subscription businesses:

Revenue = (Existing Subscribers × Retention Rate + New Subscribers) × ARPU

Phase 3: Hypothesize & Prioritize (3 minutes)

Generate hypotheses and prioritize by likelihood and impact:

High likelihood, high impact — Test first High likelihood, low impact — Quick check Low likelihood, high impact — Don't ignore Low likelihood, low impact — Skip for now

Phase 4: Investigate & Validate (10 minutes)

Systematically test each hypothesis with data.

Phase 5: Synthesize & Recommend (5 minutes)

Pull findings together and provide actionable recommendations.

Example: Investigating a 15% Revenue Drop at Uber

Phase 1: Clarify

"Before diving in, let me clarify: Is this 15% drop in gross bookings, net revenue after driver payouts, or profit? And is it a sudden drop this week, or has it been declining over several weeks?"

Interviewer: "It's gross bookings, dropped suddenly this week compared to last week."

Phase 2: Decompose

Gross Bookings = Trips × Average Trip Value

Where:
Trips = Active Riders × Trips per Rider
Average Trip Value = Distance × Price per Mile + Surcharges

Phase 3: Hypotheses

hypotheses = [
    {
        'hypothesis': 'Fewer active riders (acquisition or retention issue)',
        'likelihood': 'medium',
        'impact': 'high',
        'data_needed': 'DAU, new user signups, churn rate',
    },
    {
        'hypothesis': 'Existing riders taking fewer trips',
        'likelihood': 'medium',
        'impact': 'high',
        'data_needed': 'trips per active rider, frequency distribution',
    },
    {
        'hypothesis': 'Average trip value decreased (shorter trips, lower prices)',
        'likelihood': 'high',
        'impact': 'medium',
        'data_needed': 'trip distance, pricing, surge multipliers',
    },
    {
        'hypothesis': 'External event (weather, competitor, holiday)',
        'likelihood': 'medium',
        'impact': 'medium',
        'data_needed': 'calendar events, weather data, competitor news',
    },
    {
        'hypothesis': 'Product bug or pricing error',
        'likelihood': 'low',
        'impact': 'high',
        'data_needed': 'error logs, pricing system status',
    },
    {
        'hypothesis': 'Seasonal pattern (this is normal)',
        'likelihood': 'medium',
        'impact': 'n/a',
        'data_needed': 'year-over-year comparison, seasonal indices',
    },
]

Phase 4: Investigation

import pandas as pd
import numpy as np

def investigate_revenue_drop(bookings_df, users_df, trips_df):
    """
    Systematic investigation of revenue drop.
    """
    results = {}
    
    # 1. Top-level decomposition
    this_week = bookings_df[bookings_df['week'] == 'current']
    last_week = bookings_df[bookings_df['week'] == 'previous']
    
    results['revenue_change'] = (
        (this_week['gross_bookings'].sum() - last_week['gross_bookings'].sum()) /
        last_week['gross_bookings'].sum()
    )
    
    # 2. Driver analysis: Users vs. Frequency vs. Value
    metrics = {
        'active_users': this_week['active_users'].sum() / last_week['active_users'].sum(),
        'trips_per_user': (
            this_week['trips'].sum() / this_week['active_users'].sum()
        ) / (
            last_week['trips'].sum() / last_week['active_users'].sum()
        ),
        'avg_trip_value': (
            this_week['gross_bookings'].sum() / this_week['trips'].sum()
        ) / (
            last_week['gross_bookings'].sum() / last_week['trips'].sum()
        ),
    }
    
    results['driver_decomposition'] = metrics
    
    # 3. Geographic segmentation
    geo_impact = (
        this_week.groupby('city')['gross_bookings'].sum() /
        last_week.groupby('city')['gross_bookings'].sum() - 1
    ).sort_values()
    
    results['geo_impact'] = geo_impact.head(5)  # Worst performing cities
    
    # 4. User segment analysis
    segment_impact = (
        this_week.groupby('user_segment')['gross_bookings'].sum() /
        last_week.groupby('user_segment')['gross_bookings'].sum() - 1
    ).sort_values()
    
    results['segment_impact'] = segment_impact
    
    # 5. Time-of-day analysis
    hourly_change = (
        this_week.groupby('hour')['trips'].sum() /
        last_week.groupby('hour')['trips'].sum() - 1
    )
    
    results['hourly_pattern'] = hourly_change
    
    # 6. Cohort analysis (are new users behaving differently?)
    new_user_bookings = (
        this_week[this_week['is_new_user']]['gross_bookings'].sum() /
        last_week[last_week['is_new_user']]['gross_bookings'].sum() - 1
    )
    
    existing_user_bookings = (
        this_week[~this_week['is_new_user']]['gross_bookings'].sum() /
        last_week[~last_week['is_new_user']]['gross_bookings'].sum() - 1
    )
    
    results['new_vs_existing'] = {
        'new_user_change': new_user_bookings,
        'existing_user_change': existing_user_bookings,
    }
    
    return results

# Example output:
# revenue_change: -0.15
# driver_decomposition: {
#     'active_users': 0.95,      # -5% (not the main driver)
#     'trips_per_user': 0.92,    # -8% (significant)
#     'avg_trip_value': 1.00     # 0% (not a factor)
# }
# 
# → Primary driver: Existing users taking fewer trips
# → Next question: Why? Weather? Competitor? Product change?

Amazon-Specific Investigation Approach

The "Working Backwards" Method

Amazon prefers starting from the customer and working backwards:

1. Customer Impact — Which customers are affected?
2. Root Cause — What changed in their experience?
3. System Impact — What technical or process change caused it?
4. Business Impact — What's the financial impact?
5. Corrective Action — How do we fix it and prevent recurrence?

Amazon Revenue Streams

For Amazon, revenue drops could be in:

• E-commerce sales — Product purchases
• AWS — Cloud services
• Advertising — Ad revenue
• Subscriptions — Prime memberships
• Third-party seller services — Marketplace fees

Each has different drivers and investigation paths.

Uber-Specific Investigation Approach

The "Marketplace Health" Framework

Uber's revenue depends on marketplace liquidity:

Revenue = Requests × Acceptance Rate × Completion Rate × Average Fare

Investigate each component:

• Requests — Are fewer people opening the app?
• Acceptance Rate — Are drivers declining more rides?
• Completion Rate — Are more rides being cancelled?
• Average Fare — Are trips shorter or is surge pricing lower?

Common Mistakes to Avoid

1. Jumping to conclusions without data — Form hypotheses, then test
2. Not decomposing the metric — Revenue is an aggregate; break it down
3. Ignoring base rates — Is 15% actually unusual for this time period?
4. Forgetting external factors — Check calendar, weather, competitors
5. Not considering data quality — Is the metric accurate?
6. Over-complicating — Sometimes it's simple (holiday, bug, data pipeline issue)

How to Structure Your Answer

Minute 1-2: Clarify the problem Minute 3-5: Decompose the metric Minute 6-8: Generate and prioritize hypotheses Minute 9-14: Investigate each hypothesis with data Minute 15-17: Synthesize findings Minute 18-20: Recommend actions and next steps

Quiz: Test Your Understanding

Related Topics

• Metrics Design
• Funnel Analysis
• Business Impact Quantification
• Data Storytelling
• Statistical Thinking