Stories from the Data — Vidio Plays EDA

Dataset: 1% sample of play events (~107K rows, Feb 1–16 2020)
Goal: Understand user viewing behavior, content patterns, and determine which platform earns the keenest viewers.

0. Setup & Data Loading

import pandas as pd
import matplotlib.pyplot as plt
import matplotlib.colors as mcolors
import seaborn as sns
import warnings

warnings.filterwarnings('ignore')

# Slide theme colors
VIDIO_PINK = '#E6175C'
VIDIO_LIGHT = '#FF3D7F'
VIDIO_DARK = '#B30E46'
VIDIO_BLUSH = '#FFB8CE'
VIDIO_BG = '#FDF0F0'

# Build a custom palette from the slide gradient
vidio_palette = [VIDIO_PINK, VIDIO_LIGHT, VIDIO_DARK, VIDIO_BLUSH, '#D4145A', '#FF6699', '#8C0B3A']
vidio_cmap = mcolors.LinearSegmentedColormap.from_list('vidio', [VIDIO_BLUSH, VIDIO_PINK, VIDIO_DARK])

sns.set_theme(style='whitegrid')
sns.set_palette(vidio_palette)
plt.rcParams['figure.figsize'] = (12, 5)
plt.rcParams['figure.dpi'] = 100
plt.rcParams['figure.facecolor'] = VIDIO_BG
plt.rcParams['axes.facecolor'] = '#FFFFFF'
plt.rcParams['axes.edgecolor'] = '#DDDDDD'
plt.rcParams['text.color'] = '#2A1215'
plt.rcParams['axes.labelcolor'] = '#2A1215'
plt.rcParams['xtick.color'] = '#555555'
plt.rcParams['ytick.color'] = '#555555'

pd.set_option('display.max_columns', None)
pd.set_option('display.width', 140)

df = pd.read_csv('data/plays_1.csv.gz', on_bad_lines='skip', low_memory=False)

# Parse datetime columns
df['play_time'] = pd.to_datetime(df['play_time'], errors='coerce')
df['end_time'] = pd.to_datetime(df['end_time'], errors='coerce')

print(f"Shape: {df.shape[0]:,} rows x {df.shape[1]} columns")
print(f"Date range: {df['play_time'].min()} → {df['play_time'].max()}")
df.head(3)

Shape: 106,811 rows x 41 columns
Date range: 2020-02-01 09:26:40+00:00 → 2020-02-16 16:59:36+00:00

	hash_content_id	hash_play_id	hash_visit_id	hash_watcher_id	hash_film_id	hash_event_id	is_login	playback_location	platform	play_time	end_time	referrer	average_bitrate	bitrate_range	total_bytes	buffer_duration	referrer_group	completed	utm_source	utm_medium	utm_campaign	player_name	has_ad	flash_version	os_name	os_version	browser_name	browser_version	app_name	autoplay	is_premium	app_version	city	play_duration	content_type	stream_type	title	category_name	film_title	season_name	genre_name
0	0096dafb1049ee942e8e7cbad5abf4a46dc92e3995caac...	0162f124f5bd61592d9bca6aaa3b1b6097a00b62ef33c9...	27dfffe2bb74f8767caeb64ac4c92e9eaf4b11a28ef7e1...	e7c2138fd9d4047356066a01de46fc22ee9955a6fd6679...	NaN	c81bd64a212d3b5cb227499a6b4a3dc607748ae5133e48...	False	embed	web-mobile	2020-02-02 10:21:20+00:00	2020-02-02 10:21:20+00:00	https://m.liputan6.com/bola/read/4169085/marc-...	300000.0	200-500	0	0.0	liputan6	False	NaN	NaN	NaN	videojs	True	0,0,0	Android	9	Chrome Mobile WebView	62.0.3202	vidio	False	False	NaN	NaN	0	vod	NaN	Kiprah Eks Pemain Juventus yang Dikaitkan deng...	Sports	NaN	NaN	NaN
1	01ddda2f8667719625bd0afa58537fd95a166c93233b8b...	5f6c722782c919a9cb6254e94086462a6c34794dc741b6...	90b2b454e70a2c13cf9e8ca987d7c4f914c7d819532c5d...	d265ea2b8df584c51de3b2c6aebb6a0415384a18d1a148...	NaN	90a514e451e400c4c32ac42815920a66c500c1c2ea347d...	False	embed	web-mobile	2020-02-02 12:17:33+00:00	2020-02-02 12:17:53+00:00	https://m.bola.net/italia/lautaro-martinez-abs...	600.0	0-200	750	0.0	bolanet	False	NaN	NaN	NaN	videojs	True	0,0,0	Android	9	Chrome Mobile	79.0.3945	vidio	False	False	NaN	NaN	19	vod	NaN	Pindah ke Inter Milan, Christian Eriksen Jadi ...	Sports	NaN	NaN	NaN
2	01ddda2f8667719625bd0afa58537fd95a166c93233b8b...	66c32aa6aa6a63f7d85d1bd1b8b6b295477c88794888d8...	f79cab25855841a83e6aa01eccbf3d167491a868977b77...	d2585a4b468b7cdbebc053ffc5b49cebd58d6e738bccdd...	NaN	c278001db02102ece22d428b9235332856cba4754b63a0...	False	embed	web-mobile	2020-02-02 01:26:42+00:00	2020-02-02 01:27:07+00:00	https://m.bola.net/italia/lautaro-martinez-abs...	300000.0	200-500	900000	0.0	bolanet	False	NaN	NaN	NaN	videojs	True	0,0,0	Android	9	Samsung Internet	10.2	vidio	False	False	NaN	NaN	24	vod	NaN	Pindah ke Inter Milan, Christian Eriksen Jadi ...	Sports	NaN	NaN	NaN

# Missing values overview
missing = df.isnull().sum()
missing = missing[missing > 0].sort_values(ascending=False)
missing_pct = (missing / len(df) * 100).round(1)

pd.DataFrame({'missing_count': missing, 'missing_pct': missing_pct})

	missing_count	missing_pct
city	106811	100.0
utm_campaign	106773	100.0
utm_source	106724	99.9
utm_medium	106724	99.9
genre_name	101586	95.1
film_title	100995	94.6
season_name	100993	94.6
hash_film_id	100993	94.6
app_version	72129	67.5
stream_type	67874	63.5
autoplay	38960	36.5
category_name	38937	36.5
completed	38937	36.5
browser_version	36038	33.7
flash_version	34850	32.6
os_name	33725	31.6
browser_name	33666	31.5
referrer	5761	5.4
average_bitrate	4946	4.6
player_name	3267	3.1
os_version	405	0.4

1. Data Cleaning & Feature Engineering

# Drop columns that are ~100% null
drop_cols = ['city', 'utm_source', 'utm_medium', 'utm_campaign']
df = df.drop(columns=drop_cols)

# Fix boolean columns that were read as object due to bad lines
for col in ['completed', 'is_login', 'has_ad', 'autoplay', 'is_premium']:
    if df[col].dtype == 'object':
        df[col] = df[col].map({'True': True, 'False': False, True: True, False: False})

# Feature engineering
df['hour'] = df['play_time'].dt.hour
df['day_of_week'] = df['play_time'].dt.day_name()
df['date'] = df['play_time'].dt.date
df['is_engaged'] = df['play_duration'] >= 60  # watched at least 1 minute

print(f"Shape after cleaning: {df.shape}")
print(f"\nNew columns: hour, day_of_week, date, is_engaged")
print(f"Engaged plays (>=60s): {df['is_engaged'].sum():,} ({df['is_engaged'].mean()*100:.1f}%)")

Shape after cleaning: (106811, 41)

New columns: hour, day_of_week, date, is_engaged
Engaged plays (>=60s): 44,933 (42.1%)

2. Overall Traffic Patterns

fig, axes = plt.subplots(1, 3, figsize=(18, 5))

# Daily play volume
daily = df.groupby('date').size()
axes[0].plot(daily.index, daily.values, marker='o', linewidth=2, color=VIDIO_PINK, markerfacecolor=VIDIO_DARK)
axes[0].set_title('Daily Play Volume')
axes[0].set_xlabel('Date')
axes[0].set_ylabel('Number of Plays')
axes[0].tick_params(axis='x', rotation=45)

# Hourly distribution
hourly = df.groupby('hour').size()
axes[1].bar(hourly.index, hourly.values, color=VIDIO_LIGHT, edgecolor=VIDIO_PINK, linewidth=0.5)
axes[1].set_title('Plays by Hour of Day (UTC)')
axes[1].set_xlabel('Hour')
axes[1].set_ylabel('Number of Plays')

# Day of week
dow_order = ['Monday', 'Tuesday', 'Wednesday', 'Thursday', 'Friday', 'Saturday', 'Sunday']
dow = df['day_of_week'].value_counts().reindex(dow_order)
axes[2].bar(range(7), dow.values, color=VIDIO_PINK, edgecolor=VIDIO_DARK, linewidth=0.5)
axes[2].set_xticks(range(7))
axes[2].set_xticklabels(['Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat', 'Sun'])
axes[2].set_title('Plays by Day of Week')
axes[2].set_ylabel('Number of Plays')

plt.tight_layout()
plt.show()

3. Content Analysis

fig, axes = plt.subplots(1, 3, figsize=(18, 5))

# Content type distribution
ct = df['content_type'].value_counts()
axes[0].bar(ct.index, ct.values, color=[VIDIO_PINK, VIDIO_LIGHT, VIDIO_BLUSH])
for i, (val, cnt) in enumerate(ct.items()):
    axes[0].text(i, cnt + 500, f'{cnt/len(df)*100:.1f}%', ha='center', fontweight='bold', color=VIDIO_DARK)
axes[0].set_title('Content Type Distribution')
axes[0].set_ylabel('Number of Plays')

# Top 10 titles
top_titles = df['title'].value_counts().head(10)
axes[1].barh(top_titles.index[::-1], top_titles.values[::-1], color=VIDIO_PINK)
axes[1].set_title('Top 10 Titles by Play Count')
axes[1].set_xlabel('Number of Plays')

# Top categories
cat = df['category_name'].value_counts().head(10)
axes[2].barh(cat.index[::-1], cat.values[::-1], color=VIDIO_LIGHT)
axes[2].set_title('Top 10 Categories')
axes[2].set_xlabel('Number of Plays')

plt.tight_layout()
plt.show()

# Premium vs Free
premium = df['is_premium'].value_counts()
print("Premium vs Free content:")
print(f"  Free:    {premium[False]:>7,} plays ({premium[False]/len(df)*100:.1f}%)")
print(f"  Premium: {premium[True]:>7,} plays ({premium[True]/len(df)*100:.1f}%)")

Premium vs Free content:
  Free:    105,886 plays (99.1%)
  Premium:     925 plays (0.9%)

4. User & Session Analysis

fig, axes = plt.subplots(1, 3, figsize=(18, 5))

# Login vs Anonymous
login = df['is_login'].value_counts()
axes[0].bar(['Anonymous', 'Logged In'], [login[False], login[True]], 
            color=[VIDIO_BLUSH, VIDIO_PINK])
for i, cnt in enumerate([login[False], login[True]]):
    axes[0].text(i, cnt + 500, f'{cnt/len(df)*100:.1f}%', ha='center', fontweight='bold', color=VIDIO_DARK)
axes[0].set_title('Login vs Anonymous Plays')
axes[0].set_ylabel('Number of Plays')

# Playback location
loc = df['playback_location'].value_counts()
axes[1].bar(loc.index, loc.values, color=[VIDIO_PINK, VIDIO_LIGHT])
for i, (val, cnt) in enumerate(loc.items()):
    axes[1].text(i, cnt + 500, f'{cnt/len(df)*100:.1f}%', ha='center', fontweight='bold', color=VIDIO_DARK)
axes[1].set_title('Playback Location')
axes[1].set_ylabel('Number of Plays')

# Top referrer groups
ref = df['referrer_group'].value_counts().head(8)
axes[2].barh(ref.index[::-1], ref.values[::-1], color=VIDIO_PINK)
axes[2].set_title('Top Referrer Groups')
axes[2].set_xlabel('Number of Plays')

plt.tight_layout()
plt.show()

5. Platform Deep-Dive

Central question: Which platform do users keenly use to watch Vidio?

We define "keen" viewership not just by volume, but by engagement quality — how long users watch, whether they complete content, and how committed they are (login, repeat usage).

# Platform overview — play volume
platform_counts = df['platform'].value_counts()

fig, ax = plt.subplots(figsize=(10, 5))
bars = ax.bar(platform_counts.index, platform_counts.values, color=vidio_palette[:len(platform_counts)])
for bar, cnt in zip(bars, platform_counts.values):
    ax.text(bar.get_x() + bar.get_width()/2, cnt + 500, 
            f'{cnt:,}\n({cnt/len(df)*100:.1f}%)', ha='center', fontweight='bold', color=VIDIO_DARK)
ax.set_title('Play Volume by Platform')
ax.set_ylabel('Number of Plays')
plt.tight_layout()
plt.show()

# Build comprehensive platform metrics table
platform_metrics = df.groupby('platform').agg(
    total_plays=('hash_play_id', 'count'),
    unique_watchers=('hash_watcher_id', 'nunique'),
    median_duration=('play_duration', 'median'),
    mean_duration=('play_duration', 'mean'),
    engagement_rate=('is_engaged', 'mean'),
    login_rate=('is_login', 'mean'),
    avg_bitrate=('average_bitrate', 'mean'),
    has_ad_rate=('has_ad', 'mean'),
)

# Ensure numeric types and round
platform_metrics = platform_metrics.apply(pd.to_numeric, errors='coerce').round(3)

# Plays per unique watcher (repeat usage)
platform_metrics['plays_per_watcher'] = (
    platform_metrics['total_plays'] / platform_metrics['unique_watchers']
).round(2)

# Completion rate (VOD only — livestreaming has no 'completed')
vod = df[df['content_type'] == 'vod']
completion = vod.groupby('platform')['completed'].apply(lambda x: x.astype(float).mean()).round(3)
platform_metrics['completion_rate_vod'] = completion

platform_metrics

	total_plays	unique_watchers	median_duration	mean_duration	engagement_rate	login_rate	avg_bitrate	has_ad_rate	plays_per_watcher	completion_rate_vod
platform
app-android	25955	24083	120.0	884.198	0.744	0.763	300000.000	0.650	1.08	0.323
app-ios	1781	1673	60.0	566.040	0.526	0.755	122684.309	0.536	1.06	0.271
tv-android	5930	4989	45.0	1129.359	0.471	0.253	300000.000	0.000	1.19	0.211
tv-tizen	1031	922	60.0	903.986	0.506	0.638	300000.000	0.000	1.12	0.220
tv-webos	153	153	0.0	0.163	0.000	0.007	300000.000	0.000	1.00	0.000
web-desktop	10233	9939	32.0	594.904	0.415	0.055	48141.791	0.697	1.03	0.256
web-mobile	61728	61248	26.0	125.819	0.277	0.011	174329.385	0.926	1.01	0.237

# Visualize key engagement metrics side by side
fig, axes = plt.subplots(2, 3, figsize=(18, 10))

platforms = platform_metrics.index.tolist()
colors = vidio_palette[:len(platforms)]

# Median play duration
axes[0, 0].bar(platforms, platform_metrics['median_duration'], color=colors)
axes[0, 0].set_title('Median Play Duration (seconds)')
axes[0, 0].tick_params(axis='x', rotation=20)
for i, v in enumerate(platform_metrics['median_duration']):
    axes[0, 0].text(i, v + 1, f'{v:.0f}s', ha='center', fontweight='bold', color=VIDIO_DARK)

# Engagement rate
axes[0, 1].bar(platforms, platform_metrics['engagement_rate'] * 100, color=colors)
axes[0, 1].set_title('Engagement Rate (% plays >= 60s)')
axes[0, 1].tick_params(axis='x', rotation=20)
for i, v in enumerate(platform_metrics['engagement_rate'] * 100):
    axes[0, 1].text(i, v + 0.5, f'{v:.1f}%', ha='center', fontweight='bold', color=VIDIO_DARK)

# Login rate
axes[0, 2].bar(platforms, platform_metrics['login_rate'] * 100, color=colors)
axes[0, 2].set_title('Login Rate (%)')
axes[0, 2].tick_params(axis='x', rotation=20)
for i, v in enumerate(platform_metrics['login_rate'] * 100):
    axes[0, 2].text(i, v + 0.5, f'{v:.1f}%', ha='center', fontweight='bold', color=VIDIO_DARK)

# Plays per watcher
axes[1, 0].bar(platforms, platform_metrics['plays_per_watcher'], color=colors)
axes[1, 0].set_title('Plays per Unique Watcher')
axes[1, 0].tick_params(axis='x', rotation=20)
for i, v in enumerate(platform_metrics['plays_per_watcher']):
    axes[1, 0].text(i, v + 0.02, f'{v:.2f}', ha='center', fontweight='bold', color=VIDIO_DARK)

# VOD Completion rate
axes[1, 1].bar(platforms, platform_metrics['completion_rate_vod'].fillna(0) * 100, color=colors)
axes[1, 1].set_title('VOD Completion Rate (%)')
axes[1, 1].tick_params(axis='x', rotation=20)
for i, v in enumerate(platform_metrics['completion_rate_vod'].fillna(0) * 100):
    axes[1, 1].text(i, v + 0.5, f'{v:.1f}%', ha='center', fontweight='bold', color=VIDIO_DARK)

# Average bitrate
axes[1, 2].bar(platforms, platform_metrics['avg_bitrate'] / 1000, color=colors)
axes[1, 2].set_title('Average Bitrate (kbps)')
axes[1, 2].tick_params(axis='x', rotation=20)
for i, v in enumerate(platform_metrics['avg_bitrate'] / 1000):
    axes[1, 2].text(i, v + 1, f'{v:.0f}k', ha='center', fontweight='bold', color=VIDIO_DARK)

plt.suptitle('Platform Engagement Metrics Comparison', fontsize=14, fontweight='bold', y=1.02, color=VIDIO_DARK)
plt.tight_layout()
plt.show()

# Play duration distribution by platform (box plot, capped at 600s for readability)
fig, ax = plt.subplots(figsize=(12, 5))
plot_df = df[df['play_duration'] <= 600]

sns.boxplot(data=plot_df, x='platform', y='play_duration', palette=vidio_palette, ax=ax)
ax.set_title('Play Duration Distribution by Platform (capped at 600s)')
ax.set_ylabel('Play Duration (seconds)')
ax.set_xlabel('Platform')
plt.tight_layout()
plt.show()

# Content type preference by platform
ct_platform = pd.crosstab(df['platform'], df['content_type'], normalize='index') * 100

ct_platform.plot(kind='bar', stacked=True, figsize=(10, 5), 
                 color=[VIDIO_BLUSH, VIDIO_PINK, VIDIO_DARK])
plt.title('Content Type Preference by Platform')
plt.ylabel('Percentage (%)')
plt.xlabel('Platform')
plt.legend(title='Content Type', bbox_to_anchor=(1.05, 1))
plt.xticks(rotation=20)
plt.tight_layout()
plt.show()

ct_platform.round(1)

content_type	catchup	livestreaming	vod
platform
app-android	0.7	69.8	29.5
app-ios	1.2	76.0	22.8
tv-android	0.0	95.3	4.7
tv-tizen	0.0	91.2	8.8
tv-webos	0.0	99.3	0.7
web-desktop	0.7	45.7	53.6
web-mobile	0.2	13.0	86.8

# Normalized ranking across engagement dimensions
rank_cols = ['median_duration', 'engagement_rate', 'login_rate', 
             'plays_per_watcher', 'completion_rate_vod', 'avg_bitrate']

rank_df = platform_metrics[rank_cols].rank(ascending=True)
rank_df['composite_score'] = rank_df.mean(axis=1).round(2)
rank_df = rank_df.sort_values('composite_score', ascending=False)

print("Platform Engagement Ranking (higher = better)")
print("="*60)
rank_df

Platform Engagement Ranking (higher = better)
============================================================

	median_duration	engagement_rate	login_rate	plays_per_watcher	completion_rate_vod	avg_bitrate	composite_score
platform
app-android	7.0	7.0	7.0	5.0	7.0	5.5	6.42
tv-tizen	5.5	5.0	5.0	6.0	3.0	5.5	5.00
app-ios	5.5	6.0	6.0	4.0	6.0	2.0	4.92
tv-android	4.0	4.0	4.0	7.0	2.0	5.5	4.42
web-desktop	3.0	3.0	3.0	3.0	5.0	1.0	3.00
web-mobile	2.0	2.0	2.0	2.0	4.0	3.0	2.50
tv-webos	1.0	1.0	1.0	1.0	1.0	5.5	1.75

# Heatmap of normalized metrics
fig, ax = plt.subplots(figsize=(10, 4))

norm_df = platform_metrics[rank_cols].apply(lambda x: (x - x.min()) / (x.max() - x.min()))
norm_df.columns = ['Median Duration', 'Engagement Rate', 'Login Rate', 
                   'Plays/Watcher', 'VOD Completion', 'Avg Bitrate']

sns.heatmap(norm_df, annot=True, cmap=vidio_cmap, fmt='.2f', 
            linewidths=1, ax=ax, vmin=0, vmax=1)
ax.set_title('Platform Engagement Heatmap (0 = lowest, 1 = highest)', fontsize=13)
plt.tight_layout()
plt.show()

6. Quality of Experience

fig, axes = plt.subplots(1, 3, figsize=(18, 5))

# Bitrate distribution by platform
sns.boxplot(data=df, x='platform', y='average_bitrate', palette=vidio_palette, ax=axes[0])
axes[0].set_title('Bitrate Distribution by Platform')
axes[0].set_ylabel('Average Bitrate')
axes[0].tick_params(axis='x', rotation=20)

# Buffer duration vs play duration (sampled for performance)
sample = df[df['buffer_duration'] > 0].sample(min(5000, len(df)), random_state=42)
axes[1].scatter(sample['buffer_duration'].clip(upper=100), 
                sample['play_duration'].clip(upper=600),
                alpha=0.3, s=10, color=VIDIO_PINK)
axes[1].set_title('Buffer Duration vs Play Duration')
axes[1].set_xlabel('Buffer Duration (capped at 100s)')
axes[1].set_ylabel('Play Duration (capped at 600s)')

# Ad impact on play duration
ad_dur = df.groupby('has_ad')['play_duration'].median()
axes[2].bar(['No Ad', 'Has Ad'], [ad_dur[False], ad_dur[True]], 
            color=[VIDIO_BLUSH, VIDIO_PINK])
for i, v in enumerate([ad_dur[False], ad_dur[True]]):
    axes[2].text(i, v + 1, f'{v:.0f}s', ha='center', fontweight='bold', color=VIDIO_DARK)
axes[2].set_title('Median Play Duration: Ad vs No Ad')
axes[2].set_ylabel('Median Play Duration (s)')

plt.tight_layout()
plt.show()

7. Key Insights & Conclusions

Insight 1: Web-mobile dominates in volume but not engagement

Web-mobile accounts for ~58% of all plays, making it the largest traffic source. However, high volume does not necessarily indicate keen viewership — many of these plays originate from embedded players on partner sites (liputan6, kapanlagi, merdeka) and may represent casual, short-duration viewing.

Insight 2: Livestreaming drives a significant portion of traffic

~37% of all plays are livestreaming content, primarily TV channels (SCTV, Indosiar, RCTI, TRANS TV). These are the most-played titles by far, indicating strong demand for live TV streaming.

Insight 3: Most users watch without logging in

77% of plays come from anonymous (non-logged-in) users. Premium content is only 0.9% of plays. This suggests a predominantly ad-supported, casual viewing audience — with an opportunity to convert engaged viewers into registered/premium users.

Insight 4: Referrer traffic is key

Embedded playback (50.8%) slightly edges out direct plays. Top referrer groups — kapanlagi, merdeka, liputan6 — are major Indonesian media portals. These partnerships drive substantial traffic to Vidio's content.

Insight 5: The keen viewers live on app-android

"Keen" viewership should be measured by engagement quality, not just volume. A platform where users watch longer, complete more content, log in, and return for more plays indicates truly keen viewers.

We evaluated all 7 platforms across 6 engagement dimensions:

Metric	What it measures	Why it matters
Median Duration	How long users actually watch	Longer = more invested
Engagement Rate	% of plays >= 60 seconds	Filters out bounces
Login Rate	% of logged-in users	Shows platform commitment
Plays per Watcher	Repeat usage intensity	Keen users come back
VOD Completion Rate	% content watched to end	Shows content value
Average Bitrate	Streaming quality	Higher = better experience

Result: app-android is the platform users most keenly use to watch Vidio.

Evidence from the composite engagement ranking:

Rank	Platform	Composite Score	Key Stats
1	app-android	6.42	Median duration 120s, 74.4% engagement, 76.3% login rate, 32.3% VOD completion
2	tv-tizen	5.00	60s median, 50.6% engagement, 63.8% login
3	app-ios	4.92	60s median, 52.6% engagement, 75.5% login
4	tv-android	4.42	45s median, highest plays/watcher (1.19)
5	web-desktop	3.00	32s median, low login (5.5%)
6	web-mobile	2.50	26s median, only 27.7% engagement, 1.1% login
7	tv-webos	1.75	Too few plays (153) for meaningful comparison

Why app-android users are the keenest viewers:

1. Longest watch sessions — 120s median duration, nearly 5x longer than web-mobile (26s)
2. Highest engagement rate — 74.4% of plays last more than 60 seconds (vs 27.7% on web-mobile)
3. Highest login rate — 76.3% of app-android users are logged in, showing account commitment
4. Best VOD completion — 32.3% of VOD content is watched to completion
5. Intentional usage — installing an app requires deliberate effort, unlike clicking an embedded web player
6. Max bitrate streaming — 300kbps average, indicating users choosing higher quality

In contrast, web-mobile has the highest volume (58% of all plays) but the lowest engagement — most of this traffic comes from casual embedded plays on partner news sites (kapanlagi, merdeka, liputan6), where users may not even intentionally seek out Vidio content.