The Science

How we actually decode algorithms.

This isn't marketing buzzword stuff. Here's a look under the hood at the real data science, machine learning, and statistical modeling that powers everything we do at Solarium.

The Problem

Every platform runs on an algorithm. Most marketers ignore them.

TikTok's ForYou page, YouTube's recommendation engine, Google's search rankings, Instagram's Explore tab — they all use machine learning models to decide what content gets shown to whom.

Most agencies treat these like black boxes. They follow "best practices" and hope for the best. We take a different approach: we build our own models to predict what the algorithms will do.

Our Approach

Reverse-engineering ranking systems with real data science.

We collect thousands of data points from content performance across each platform — engagement rates, watch time, save ratios, share depth, posting time, format, hashtags, and dozens more.

Then we apply regression analysis, random forests, gradient boosting, and neural networks to identify which variables have the highest predictive power for content distribution. The result is a custom model that tells us — with measurable accuracy — what will perform before you hit publish.

# Simplified example of our modeling pipeline
from solarium import AlgoModel

# Train on 10,000+ content data points
model = AlgoModel(platform="tiktok")
model.fit(features, engagement_scores)

# Score new content before publishing
score = model.predict(new_content)
# → viral_probability: 0.847
# → optimal_length: 34s
# → best_posting_window: 6-8pm EST
Key Algorithm Signals

What the algorithms are actually measuring.

These are the variables our models have identified as having the highest predictive power across platforms. Each one is weighted differently depending on the platform.

Engagement Velocity

How quickly likes, comments, and shares accumulate in the first minutes after posting.

r = 0.89
👀

Watch-Through Rate

Percentage of viewers who watch your content to completion. The single strongest signal on video platforms.

r = 0.94
🔖

Save / Bookmark Ratio

Saves indicate deep value — algorithms weight these 3-5x more than likes.

r = 0.82
📤

Share Depth

Not just whether people share, but how far the share chain extends. Indicates virality potential.

r = 0.78
💬

Comment Depth

Reply chains matter more than comment count. Algorithms reward genuine conversation.

r = 0.71
🔄

Replay Signals

Content that gets re-watched signals high value. Algorithms boost it to new audiences.

r = 0.86

r = correlation coefficient with content distribution in our models

Platform Intelligence

Every algorithm works differently.

We've built separate models for each major platform because they each weight signals differently. Here's a high-level look at what matters where.

T

TikTok ForYou Algorithm

TikTok's recommendation engine prioritizes watch-through rate above all else, followed by engagement velocity in the first 30 minutes and share depth. Sound choice and trend alignment act as secondary boosters.

watch_through_rate early_engagement share_depth replay_rate sound_trend_score
Y

YouTube Recommendation Engine

YouTube optimizes for session time — how long a viewer stays on the platform after watching your video. CTR on impressions and satisfaction signals (likes vs. "not interested") are heavily weighted.

session_time impression_ctr satisfaction_score audience_retention subscribe_after_watch
I

Instagram Explore & Reels

Instagram's algorithm weights relationship signals (DMs, profile visits) alongside engagement stacking — the combination of saves + shares + comments matters more than any single metric.

save_ratio share_to_dm engagement_stack relationship_score content_freshness
G

Google Search & SGE

Google's core ranking combines E-E-A-T trust signals, topical authority, passage-level relevance, and now AI Overview triggers. Entity relationships and structured data are becoming increasingly critical.

eeat_score topical_authority passage_relevance entity_mapping ai_overview_trigger
Staying Ahead

Algorithms change. We detect it in hours, not months.

Platform algorithms are updated constantly — sometimes multiple times per week. Most marketers don't realize their strategy stopped working until they see a month of declining metrics.

We run continuous drift detection on our models. When the relationship between signals and distribution changes, we know immediately. Your strategy adapts before the impact hits.

Think of it like a weather forecast for algorithms. We can't control the weather, but we can make sure you're never caught in the rain.

Curious what the algorithms think of your brand?

The free audit will show you exactly how your content scores across our models. No cost, no commitment — just data.

Get Your Free Audit → View Pricing