Uber & Bolt Driver Analytics
Gabriel Souza
Uber & Bolt Driver Analytics
A data analytics portfolio project built on real ride-hailing data collected manually over 6 months (March–August 2026) as a working TVDE driver in Lisbon, Portugal.
Project Overview
This project analyses 2,541 rides across two platforms (Uber and Bolt) to answer practical questions about driver earnings, ride efficiency, and operational strategy. Unlike most portfolio projects that use generic datasets, every data point here was recorded from real working shifts.
The analysis is structured in four layers:
Excel — manual data collection and storage
Power BI dashboard — interactive KPIs and visual exploration
Python notebooks — statistical analysis and hypothesis testing
SQL (SQLite) — business queries replicating and extending the Python analysis
Data Collection
How the data was collected
All data was collected manually by the driver after each working shift, directly from the Uber and Bolt driver apps. Each ride was recorded individually into an Excel spreadsheet.
The data collection process for each ride:
Open the driver app (Uber or Bolt) after completing a ride
Navigate to ride history and open the individual ride details
Record the following fields manually into the Excel spreadsheet:
Platform, Category, Date, Start Time, End Time
Duration (minutes), Origin postal code, Destination postal code
Distance (km), Client fare (€), Driver earnings (€)
Repeat for every ride at the end of each shift
This manual process ensures full control over the data and allows for immediate quality checking — any inconsistency (e.g. duration not matching start/end times) was identified and corrected at the point of entry.
Data source examples
The screenshots below show real examples of the raw data as it appears in the Uber and Bolt driver apps before being recorded into the spreadsheet. All personal passenger information is hidden by the platforms for privacy reasons.
Uber ride examples
Bolt ride examples
Screenshots from the Uber and Bolt driver apps are used for educational and portfolio purposes only, to illustrate the data collection methodology. All passenger data is automatically hidden by the platforms.
Why this data is reliable
No automated scraping — every row was entered manually and verified at the point of collection
Exact values — earnings and distances are taken directly from the official platform receipts, not estimated
Consistent format — the same fields were recorded for every ride across the entire collection period
Platform receipts as source of truth — the driver earnings shown in the app represent the final confirmed amount after any platform adjustments
Collection period: March 4 – August 2026 Total rides analysed: 2,541 Working days: 152 Platforms: Uber and Bolt
This project was developed between March and August 2026. The analysis covers the full March–August 2026 period to capture both spring and summer mobility patterns in Lisbon.
Data Schema
Columns used in analysis
Column Description Type Ride_ID Unique ride identifier String Platform Uber or Bolt String Category Service tier (UberX, Economy, Bolt, etc.) String Date Ride date Date Start_Time Ride start time Time End_Time Ride end time Time Duration_Min Ride duration in minutes Float Origin_PostCode Postal code of ride origin String Dest_PostCode Postal code of ride destination String Distance_Km Ride distance in kilometres Float Client_Fare_EUR Total amount paid by passenger Float Driver_Earnings_EUR Net amount received by driver Float
Columns excluded from analysis
Column Reason for exclusion Platform_Fee_EUR Focus is on driver perspective, not platform economics Fee_Pct Same reason as above Portagem Toll reimbursement — passes through, does not affect net earnings Dinamica Bolt-only surge data — not available on Uber, excluded for consistency Tip Inconsistent and cannot be relied upon as regular income
Operational Parameters
Acceptance criteria
Rides are accepted under the following conditions:
Minimum 0.50 €/km
Minimum 12 €/hour Example: A €6 ride must be 12 km long and last a maximum of 30 minutes.
Working schedule
Start: 12:30–13:00
End: 21:30–22:00
Rest days: variable (no fixed day off)
having a few days with runs outside of that schedule for personal reasons
Earnings targets
Daily: 100€ gross
Weekly: 500€ net (after fuel and fleet costs)
Minimum wage reference
The reference minimum wage used is 920€ gross (2026). The gross value was chosen over net because tax deductions vary by personal situation (marital status, dependants).
Cost Structure and Net Profit Calculation
Fleet fee
The vehicle is operated under a fleet agreement. The fleet charges a fixed fee of 30€ per week, regardless of the number of days worked or rides completed. This is a common arrangement in the Portuguese TVDE market — some fleets charge a fixed weekly amount, others charge a percentage of earnings.
Fuel cost — how it was calculated
Step 1 — Measuring real fuel consumption:
At the start of the project, a full-tank-to-full-tank fuel consumption test was conducted over one working week:
Measurement Value Odometer at start 206,448 km Odometer at end 207,780 km Distance driven 1,332 km Fuel used 68.75 litres Consumption result 5.16 L/100km
Step 2 — Fuel price (period average):
The vehicle is fuelled exclusively at PRIO stations. Rather than using a single week's price (which fluctuates), the fuel cost calculation uses the average diesel price across the full collection period (March–August 2026):
Average diesel price at PRIO: 1.892 €/litre
Step 3 — Cost per km:
Fuel cost per km = (5.16 / 100) x 1.892 = 0.0976 €/km
Step 4 — Proportional allocation:
The vehicle is also used for personal trips outside working hours. To avoid overestimating operational costs, fuel cost is applied only to the km recorded in the dataset (ride km), not to total km driven:
Fuel cost = Total ride km x 0.0976 €/km
This ensures the net profit calculation reflects only costs directly attributable to TVDE activity.
Net profit cascade
For every aggregation in this project, net profit follows this calculation:
Where Weeks_Worked is calculated as:
This formula is identical in Python, SQL and Power BI — verified by cross-validation.
Business Questions Answered
Driver economics
How much does a TVDE driver actually earn per hour (net)?
How many days does it take to reach the minimum wage (920€ gross)?
What is the real cost of fuel per km driven?
What is the net profit after all operational costs?
Operational efficiency
Does working more hours per day increase €/hour? (Q1)
Do longer rides pay more per km? (Q2)
Is there a statistically significant difference between Uber and Bolt? (Q3)
Does more rides per day lead to higher €/hour? (Q4)
Temporal patterns
Is there a consistent weekly pattern? (Q5)
Which hour of the day generates the most profit? (Q6)
Is there a trend over time in €/hour earnings? (Q7)
Do weekends behave differently from weekdays? (Q8)
Ride profile
What is the real distribution of distances and fares? (Q9)
Is there an optimal ride distance? (Q10)
What is the profile of the most frequent ride? (Q11)
Acceptance criteria analysis
What % of rides meet the acceptance criteria? (Q12)
At what time do more rides meet the acceptance criteria? (Q13)
How close are results to the daily and weekly targets? (Q14)
Is the working schedule (12:30–22:00) well dimensioned? (Q15)
Does more km per day lead to higher net profit? (Q16)
Geographic mobility
Which postal code areas generate the most rides?
What are the most frequent origin to destination flows?
Which zones generate the highest earnings per ride?
How do geographic hotspots change by time period and day type?
Methodology Notes
Why median over mean?
All ride distributions are right-skewed — a small number of long, high-value rides pull the mean upward. The median is used as the primary central tendency measure throughout this analysis.
Statistical tests used
Pearson correlation + linear regression — for continuous variable relationships (hours vs €/hour, distance vs €/km)
Mann-Whitney U test — for platform and day-type comparisons, chosen over t-test because ride earnings do not follow a normal distribution
Outlier treatment
Outliers were identified using the IQR method (1.5x fence). All flagged outliers were inspected individually and retained — they represent genuine long-distance rides (max 54 km), not data entry errors.
Weeks worked calculation
Fleet cost is calculated using the difference between the first and last recorded date, divided by 7 and rounded. This matches the Power BI DAX DATEDIFF calculation exactly, ensuring consistency across all tools.
Cross-validation
Net profit after all costs was validated across all three tools:
Tool Net Profit Power BI 12,869.07 € Python 12,869.07 € SQL 12,869.07 €
All three return identical results, confirming consistency across the full analysis pipeline.
Key Findings
(Full March–August 2026 dataset — 2,541 rides across 152 working days)
Net profit per hour: 18.93 €
Days to reach minimum wage: 11 days (920€ gross)
84.3% of rides meet both acceptance criteria — strategy is working
Friday is the best day (116€ avg daily earnings, 27.2 €/h)
16:00–18:00 is the golden window — 91–94% acceptance rate
Bolt outperforms Uber in acceptance rate (88.2% vs 80.6%) — statistically significant (p = 0.0019)
More rides per day = higher €/hour (r = 0.422, p < 0.001)
Short rides (0–3 km) are most efficient (2.22 €/km, 42.9 €/h) but represent only 8.5% of rides
The 5–12 km bracket is the core market (53.6% of all rides)
Postal codes 1300 and 1500 are the top origin and destination zones, accounting for ~21% of all ride starts
Power BI Dashboard Preview
Static previews of the 4-page interactive dashboard, generated from the full March–August 2026 dataset.
Note: These are static screenshots. To interact with the dashboard (filter by platform, drill into categories, explore the map), openpowerbi/uber-bolt-driver-analytics.pbixin Power BI Desktop.
Overview — KPIs, earnings and minimum wage tracker
Driver Performance — daily earnings trend over time
Platforms — Uber vs Bolt volume and category breakdown
Map — geographic distribution of ride origins across Lisbon
Project Structure
How to Run
Run notebooks in order:
notebooks/01_data_cleaning_and_analysis.ipynb
notebooks/02_sql_analysis.ipynb
notebooks/03_mobility_analysis.ipynb
Tools Used
Tool Purpose Excel Manual data collection and storage Python + pandas Data cleaning and statistical analysis matplotlib + seaborn Data visualisation scipy Statistical testing SQLite (via Python) SQL query analysis folium Geographic visualisation library Power BI Interactive dashboard Claude (Anthropic) AI-assisted development — code review, debugging and documentation GitHub Version control and portfolio hosting
Author
Gabriel Souza — TVDE driver and data analyst Lisbon, Portugal · 2026 LinkedIn: https://www.linkedin.com/in/gabriel-souza-5bb6123a8/ GitHub: https://github.com/gabriel-souza-data
Like this project
Posted Sep 8, 2026
Analyzed 2,541 real Uber & Bolt rides using Python, SQL, and Power BI, cross-validating net profit to the cent across all three tools.
Likes
0
Views
0
Timeline
Mar 4, 2026 - Aug 31, 2026
