Content Demand Forecasting
“How much will this new show be watched?”
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
By submitting, you accept the Terms and Privacy Policy.
Loved by data scientists, ML engineers & CXOs at
A real-world example
How much will this new show be watched?
Studios invest $200M+ per year in original content with limited ability to predict which titles will perform. Marketing spend is allocated uniformly rather than proportional to predicted demand. Content that underperforms wastes production budget; content that overperforms gets under-marketed. For a studio producing 50 originals per year, accurate demand forecasting saves $30-50M in misallocated production and marketing spend.
Quick answer
Graph neural networks forecast viewership demand for new content before launch by connecting creator track records, genre trends, similar title performance, and trailer engagement signals in a content graph. Studios producing 50 originals per year save $30-50M in misallocated production and marketing spend by predicting which titles will over- or under-perform.
Approaches compared
4 ways to solve this problem
1. Comparable title analysis (manual benchmarking)
Pick 3-5 similar past titles and average their performance as a demand estimate. The standard approach in content planning meetings.
Best for
Quick directional estimates when you need a number for budget discussions. Simple and intuitive.
Watch out for
Highly subjective -- different analysts pick different comparables and get different estimates. Cannot account for how genre trends have shifted since the comparable was released, or how this creator's trajectory has changed.
2. Regression on content attributes
Train a model on content features (genre, budget, cast star-power, release window) to predict viewership. Linear or tree-based models.
Best for
Systematic baseline that removes human bias from comparable selection. Scales across a large content portfolio.
Watch out for
Treats each title independently. Cannot model the interaction between a creator's track record and the current genre trend, or how trailer engagement converts differently by subscriber segment.
3. Social listening / trailer signal models
Track social media mentions, trailer views, and search volume as leading indicators of demand. Build regression models on these pre-launch signals.
Best for
Captures real-time demand signals in the weeks before launch. Good for adjusting marketing spend close to release.
Watch out for
Only available in the final weeks before launch -- too late for production or early marketing decisions. Social buzz doesn't always convert to viewership (trailers can go viral for the wrong reasons).
4. KumoRFM (relational graph ML)
Connect content metadata, genres, creators, similar titles, and trailer engagement into a graph. The GNN learns how creator track records interact with genre trends and how trailer signals convert to viewership by segment.
Best for
Pre-launch predictions with enough lead time to optimize marketing budget and scheduling. Captures multi-hop signals that linear models miss: 'this creator's rising trajectory + trending genre + strong trailer completion rate among the target demographic.'
Watch out for
Requires a content library with historical performance data across creators, genres, and marketing metrics. More valuable for studios with 20+ titles in their back catalog.
Key metric: Studios save $30-50M annually by using pre-launch demand forecasts to optimize marketing allocation and content scheduling across 50+ original titles.
Why relational data changes the answer
Content demand is not determined by any single attribute. A thriller from a hot director with a strong trailer sounds like a hit, but demand depends on how the thriller genre is trending right now, whether the director's last project was in a similar genre, how the trailer's completion rate compares to similar trailers, and whether the release window competes with other high-profile launches. These signals live in different tables (content, creators, genres, trailers, schedules), and the interactions between them matter more than any individual signal.
Relational models connect the full content graph. They learn that Director Nakamura's track record of 8.2M average viewers is more predictive for sci-fi thrillers than for rom-coms, that a 72% trailer completion rate converts to higher viewership when the genre trend is 'rising' vs 'stable,' and that similar titles from 2 years ago over-performed because the genre was peaking. On the RelBench benchmark, relational models score 76.71 vs 62.44 for single-table baselines. For content forecasting, that accuracy gap means the difference between greenlighting a $60M production with a realistic demand estimate and flying blind.
Forecasting content demand from comparable titles is like a talent scout evaluating a baseball prospect by finding 3 players with similar stats. The scout misses that the prospect is left-handed (rare advantage), plays in a different league, and is trending upward. Graph-based forecasting reads the full scouting report: the player's trajectory, the league context, the team fit, and the competitive landscape, producing a projection grounded in connected signal rather than surface-level comparisons.
How KumoRFM solves this
Graph-powered intelligence for media platforms
Kumo connects content metadata, genres, creators, similar titles, and trailer engagement into a graph. The model learns demand signals from the content graph: how a creator's track record interacts with genre trends, how trailer engagement converts to viewership by subscriber segment, and how similar titles' performance trajectories predict new content demand. Predictions are available before launch, enabling optimized marketing spend and scheduling.
From data to predictions
See the full pipeline in action
Connect your tables, write a PQL query, and get predictions with built-in explainability — all in minutes, not months.
Your data
The relational tables Kumo learns from
CONTENT
| content_id | title | genre | budget | creator_id |
|---|---|---|---|---|
| NEW01 | Dark Protocol | Thriller/Sci-Fi | $45M | DIR001 |
| NEW02 | Summer House | Rom-Com | $18M | DIR002 |
| NEW03 | Iron Circuit | Action | $60M | DIR003 |
GENRES
| genre | trend_score | avg_completion | subscriber_penetration |
|---|---|---|---|
| Thriller/Sci-Fi | Rising | 68% | 32% |
| Rom-Com | Stable | 75% | 28% |
| Action | Declining | 62% | 35% |
CREATORS
| creator_id | name | avg_viewership_m | hit_rate |
|---|---|---|---|
| DIR001 | J. Nakamura | 8.2M | 60% |
| DIR002 | S. Okafor | 4.5M | 40% |
| DIR003 | R. Zhang | 12.1M | 45% |
SIMILAR_TITLES
| content_id | similar_to | similarity_score | viewership_m |
|---|---|---|---|
| NEW01 | Title-X | 0.89 | 9.4M |
| NEW02 | Title-Y | 0.82 | 5.1M |
| NEW03 | Title-Z | 0.91 | 14.2M |
TRAILER_VIEWS
| content_id | views_7d | completion_rate | social_shares |
|---|---|---|---|
| NEW01 | 4.2M | 72% | 180K |
| NEW02 | 1.8M | 65% | 45K |
| NEW03 | 6.1M | 58% | 320K |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT SUM(WATCH_HISTORY.minutes_watched, 0, 30, days) FOR EACH CONTENT.content_id WHERE CONTENT.release_date > '2025-03-01'
Prediction output
Every entity gets a score, updated continuously
| CONTENT_ID | TITLE | PREDICTED_VIEWERS_30D | CONFIDENCE |
|---|---|---|---|
| NEW01 | Dark Protocol | 11.2M | High |
| NEW02 | Summer House | 4.8M | Medium |
| NEW03 | Iron Circuit | 9.6M | Medium |
Understand why
Every prediction includes feature attributions — no black boxes
Content NEW01 -- Dark Protocol (Thriller/Sci-Fi)
Predicted: 11.2M viewers in first 30 days (High confidence)
Top contributing features
Trailer engagement conversion rate
72% completion
28% attribution
Creator J. Nakamura track record
8.2M avg
24% attribution
Genre trend score
Rising
20% attribution
Similar title performance
9.4M viewers
17% attribution
Social share velocity (first 7 days)
180K shares
11% attribution
Feature attributions are computed automatically for every prediction. No separate tooling required. Learn more about Kumo explainability
PQL Documentation
Learn the Predictive Query Language — SQL-like syntax for defining any prediction task in 2–3 lines.
Python SDK
Integrate Kumo predictions into your pipelines. Train, evaluate, and deploy models programmatically.
Explainability Docs
Understand feature attributions, model evaluation metrics, and how to build trust with stakeholders.
Frequently asked questions
Common questions about content demand forecasting
How do you forecast viewership for new content before launch?
Connect creator track records, genre trend data, similar title performance, and trailer engagement metrics in a graph model. The model learns how these signals interact -- a rising genre plus a proven creator plus strong trailer engagement compounds into higher predicted viewership than any linear combination suggests. Predictions are available weeks before launch, enabling optimized marketing and scheduling.
What data do you need for content demand forecasting?
Content metadata (genre, budget, cast, director), historical performance of similar titles, creator track records, genre trend indicators, and trailer/marketing engagement metrics (views, completion rate, social shares). More connected data means better predictions. The model improves significantly when you add subscriber segment-level preferences.
How accurate is content demand forecasting with AI?
Graph-based models predict 30-day viewership within 20-30% MAE for most titles, with higher accuracy for titles in established genres with proven creators. The business value comes from correctly rank-ordering your portfolio: knowing which titles will over-perform (increase marketing) and which will under-perform (reduce spend) saves $30-50M annually for a studio producing 50 originals.
How do you decide marketing spend for new content?
Use pre-launch demand forecasts to allocate marketing proportional to predicted performance. Titles with high predicted demand get heavy marketing pushes; titles with moderate predictions get standard campaigns. This replaces the common practice of uniform marketing allocation, which over-invests in under-performers and under-invests in potential hits.
What is the ROI of content demand forecasting?
Studios producing 50 originals per year save $30-50M by right-sizing production budgets and marketing spend based on predicted demand. The model pays for itself if it correctly identifies even 2-3 under-performers early enough to reduce marketing spend, or 2-3 over-performers early enough to increase promotional investment.
Bottom line: A studio producing 50 originals per year saves $30-50M by accurately forecasting demand before launch. Kumo's content graph connects creator track records, genre trends, and trailer signals to predict viewership with enough lead time to optimize marketing and scheduling.
Related use cases
Explore more media & entertainment use cases
Topics covered
One Platform. One Model. Infinite Predictions.
KumoRFM
Relational Foundation Model
Turn structured relational data into predictions in seconds. KumoRFM delivers zero-shot predictions that rival months of traditional data science. No training, feature engineering, or infrastructure required. Just connect your data and start predicting.
For critical use cases, fine-tune KumoRFM on your data using the Kumo platform and Research Agent for 30%+ higher accuracy than traditional models.
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
