AI Audience Analytics for Publishers

AI-powered audience analytics system for publishers

Publishers accumulate rich data on reader behavior, but only use the top-level metrics (views, bounces). Deep audience analytics with ML transforms this data into actionable insights for editorial and sales teams.

Audience segmentation

RFM + Behavioral Clustering:

Behavioral segmentation of readers goes beyond demographics:

import pandas as pd
import numpy as np
from sklearn.preprocessing import StandardScaler
from sklearn.cluster import KMeans

def segment_readers(reader_events_df, n_segments=6):
    """
    Сегментация читателей по поведенческим признакам.
    reader_events_df: события (статьи, время, scroll_depth, shares)
    """
    # Агрегация на уровне читателя
    reader_features = reader_events_df.groupby('reader_id').agg({
        'article_id': 'count',                    # Frequency
        'session_duration': 'mean',               # вовлечённость
        'scroll_depth_pct': 'mean',               # глубина чтения
        'days_active': 'nunique',                 # активные дни
        'category': lambda x: x.mode()[0],       # любимая категория
        'shares': 'sum',                          # виральность
        'direct_visit': 'mean',                  # лояльность (не трафик из поиска)
        'last_visit': lambda x: (pd.Timestamp.now() - pd.to_datetime(x).max()).days
    }).reset_index()
    reader_features.columns = ['reader_id', 'articles_read', 'avg_session',
                               'avg_scroll', 'active_days', 'top_category',
                               'shares', 'direct_ratio', 'recency_days']

    # Нормализация
    numeric_cols = ['articles_read', 'avg_session', 'avg_scroll',
                    'active_days', 'shares', 'direct_ratio', 'recency_days']
    scaler = StandardScaler()
    X = scaler.fit_transform(reader_features[numeric_cols].fillna(0))

    # K-Means кластеризация
    kmeans = KMeans(n_clusters=n_segments, random_state=42, n_init=10)
    reader_features['segment'] = kmeans.fit_predict(X)

    return reader_features

Typical segments: - Loyalists: come directly, read every day, deep scroll - Casual browsers: sometimes come from social networks, read only the headline - Topic specialists: read a lot in one category (audience for niche newsletters) - Social sharers: read little themselves, but actively share - Churning users: were active, stopped

Content analytics

Content Performance Scoring:

Article evaluation is based not only on views, but also on quality metrics:

Topic Resonance Analysis:

Which topics resonate most with different audience segments: - NLP content clustering (BERTopic) → topic clusters - Matrix "Topic × audience segment" → editorial insights - The editors see: "Loyalist readers want more analytics, Casual readers want more listicles"

Subscriber Churn Forecast

Subscriber Health Score:

Dynamic scoring of each subscriber: - Decrease in reading frequency → yellow flag - Unsubscribe from email newsletter → red flag - Pattern: did not open emails for 3 weeks + did not visit the site → P(churn_30d) = 0.7

Win-back campaigns:

If P(churn) > 0.6: - Personalized selection of the best articles for the period of absence - Special offer upon subscription expiration (if LTV > cost of offer) - Re-engagement email series with increasing incentives

Attribution and monetization insights

Content-Subscriber Attribution:

Which articles actually lead to subscriptions? - Multi-touch attribution: the reader read 8 articles before subscribing - Markov Chain attribution: each article gets its fair share of credit for conversion - Editorial: invest in creating content type X because it converts

Development time: 2–4 months for an audience analytics platform with segmentation, churn prediction, and content scoring.