Recommendation: Valvoline should launch a free, ad-free maintenance-tracking app (reminders, a maintenance log, recall alerts) that brings drivers back for service, with booking as the in-app transaction.
A group consulting project (AN 306) analyzing 988,310 Google Play apps to answer four business questions for Valvoline. Built with Python, pandas, scikit-learn and Plotly.
Valvoline runs about 2,400 service centers and completes 30M+ services a year. Its Instant Oil Change app is well rated (4.4★), but it only books visits: nothing brings drivers back to the app between oil changes. The project asked whether a maintenance-tracking app could close that gap.
| Business question | Answer | Key evidence |
|---|---|---|
| 1. Is there room for a better car app? | Yes. Drivers rate this category poorly. | Auto & Vehicles ranks 44th of 48 categories by average rating (3.89★ vs 4.10★ store-wide). |
| 2. Do recurring-use apps earn deeper engagement than one-time apps? (primary) | Yes. | Maintenance trackers and reminder apps earn 2.0× the median ratings per install of one-time apps such as store locators (0.025 vs 0.012), and rate 0.45★ higher (4.22★ vs 3.77★). |
| 3. What predicts installs, and do permissions matter? | Engagement does; permissions barely matter. | In a Random Forest model of Auto & Vehicles installs, Rating Count and Rating carry 88% of the weight; all 12 permission features together carry about 4%. |
| 4. Which revenue model should the app use? | Free and ad-free, built around booking. | 97.7% of 4.0★+ car apps are free to download, and the most common model (58.2%) is free with no ads or in-app purchases. |
Feedback on the presentation suggested letting the app add each predicted service date to the driver’s calendar, with a link to book. The data points to an open gap: none of the 35 recurring maintenance apps in the analysis request Calendar access, and only 177 of 7,403 Auto & Vehicles apps (2.4%) do. The Calendar permission also carries almost no weight in the install model (about 0.001), so asking for it when a driver adds a reminder shouldn’t cost installs. The data shows the gap, not the effect, so the next step is a pilot that measures how many calendar reminders turn into bookings.
| My Firestone | Valvoline Instant Oil Change | Car Care Companion (target) | |
|---|---|---|---|
| App type | Recurring | One-time | Recurring |
| Star rating | 3.50★ | 4.40★ | 4.5★+ |
| Ratings per install | 0.0197 | 0.0245 | 0.041+ (top quarter of recurring apps) |
Firestone shows a service brand can run a recurring maintenance app, but drivers rate it near the bottom of the recurring group. Valvoline already has the goodwill (4.40★) with a one-time app. Car Care Companion puts that goodwill into the recurring format, where Firestone is weak.
Google-Playstore.csv (2.3M apps) ─┐
├─ 01_data_cleaning_and_join.ipynb ─→ df_clean.parquet ─→ 02_analysis.ipynb ─→ charts
googleplay-app-permission.json ─┘ (988K apps × 44 columns)
| Notebook | What it does |
|---|---|
01_data_cleaning_and_join.ipynb |
Streams the large permissions file and joins it to the app metadata, removes apps with fewer than 10 ratings, and engineers size, monetization and permission features. |
02_analysis.ipynb |
Trains the install models, compares recurring and one-time apps, places Valvoline and Firestone against both groups, and builds the charts. |
Both notebooks run in Google Colab and read their data from Google Drive. The raw data files aren’t included because of their size. Run notebook 01 once to create df_clean.parquet, then run notebook 02.