Ironhack Data Analytics | May 2026
Spotify Hit Formula
What does a Top 50 song sound like — and how can we invest smarter?
Claire & Diana | Data Analytics Project for Warner Music GroupBy
The Business case
The Noise
Record labels receive thousands of demos annually. What separates a chart-topping hit from the rest?
The Risk
Investment decisions are largely based on "gut feel." Missed opportunities equal lost revenue.
There is a need for an investment filter to identify songs
with high commercial potential before committing A&R resources.
Quantitative Solution
Audio Feature Analysis: Deep dive into Spotify Top 50 metrics to decode hit DNA.
Modern Data Stack: Built a SQL + Python pipeline to spot promising audio profiles.
Investment Filters: Develop a data-driven filtering mechanism for A&R teams.
Genre-Specific Criteria: Testing if one "Universal Formula" beats specialized genre rules.
Dual Hypotheses
H1: Universal Formula
Top 50 songs share a signature audio profile that WMG can use to identify commercially promising tracks at scale
H2: Genre-Specific Formula
The audio profile differs by genre, requiring genre-specific filter for investment criteria
Project pipeline
STEP 1
Setup & Data Loading
Imports, raw data preview
STEP 2
Data Quality & Cleaning
Column renaming, type casting, null/duplicate.
Create SQL database
STEP 3
Formula Analysis
Universal hit formula baseline (H1)
STEP 4
Genre Analysis
Genre audio profiles differentiation (H2)
STEP 5
Conclusions & Obstacles
WMG view on filtering mechanism for A&R teams.
Data Acquisition & Cleaning
Kaggle Top 50 Spotify 2019
Primary Attributes: Danceability, Energy, Valence, BPM, Loudness, Acousticness, Speechiness, Liveness.
Note: Valence= Emotional positivity (low=sad, high=happy)
Data Hygiene: No missing values detected. Standardized track names and handled duplicate titles via Python cleaning.
Database schema (ERD)
Miro
DrawDB
H1: Defining the Universal Formula
looked at which metrics had the lowest variability across the entire dataset
Looked at the coefficient of variation:
σ (standard deviation)
CV (%) = ─────────── x 100
μ (mean)
H1 partially confirmed: Danceability, length and energy have the least variability, suggesting artists have the least freedom in these metrics.
Liveness, acousticness and speechiness have the highest variability, meaning they may be less important factors.
H1: What separates the top?
Which metrics differentiate the top performing songs?
Ranked & binned songs in groups of 5
Normalized values due to differences in dynamic range.
value- μ (mean)
z-score = ───────────
σ (standard deviation)
H1 partially confirmed: High variability across rankings. BPM, speechiness and valence differentiate top rank from other ranks.
H1: What separates the top?
Looked at the relationships between metrics across the binned dataset.
Evaluated the correlation between each metric and popularity using Pearson’s correlation coefficient.
H1 partially confirmed: BPM (0.54), Speechiness (0.48), and Valence (−0.54) exhibited the strongest correlations with popularity. Though the relationships were moderate in strength.
H2: is there a differentiation per genre?
The starting point -> metrics differentiating top ranked songs from others:
H2 Confirmed: Audio profiles are genre-specific.
The metrics (normalized) DO variate among all genres.
H2: Genre clusters vs. Top songs metrics
Comparing genre clusters differences against top 5 songs based on the 3 metrics (Beats Per Minute (BPM), valence and speechiness).
H2 Confirmed: Top-tier success cannot be assessed by an universal filter as its dependent on genre-specific profiles.
High Beats & Vocals
(BPM+++,
speechiness ++)
Mellow & Low-Tempo
(speechiness-, BPM+)
High "Feel-Good" Vibes
(valence+++)
Top performing songs
Obstacles
Small sample� (50 songs) — exploratory only; limited predictive power
No control group� without non-charting songs, we can't isolate what makes a hit
Single year (2019)� streaming trends shift; these findings may not generalize
Conclusions
Consistent Core
Features� Danceability, length,
and energy.
Variation in Expressive Features� BPM, speechiness
and valence.
Genre-specific Sound Profiles� "One-size-fits-all" filter ineffective. Sounds profiles are
genre dependent.
The Takeaway: WMG should apply genre-specific filters rather than a single universal rule to increase investment accuracy.
Questions?
Thank you for your attention.
Claire & Diana | Ironhack Data Analytics
Dataset: kaggle.com/datasets/leonardopena/top50spotify2019
H1: Do the top songs adhere closest to the formula?
We identified danceability, length and energy as the most consistent metrics
The top songs largely fall within the optimal range for these qualities, only one outlier was seen
H1 partially confirmed: Energy, length, and danceability remained relatively consistent across songs, while top-performing tracks were differentiated by higher BPM, greater speechiness, and lower valence
Sources