Prediction problems don’t exist in isolation.
Forecasting next month’s revenue? You might need predicted demand first. Predicting customer lifetime value? That could depend on predicted churn probability. Estimating production costs? You probably need predicted material prices.
In the real world, models feed into models. But most ML platforms treat each project as an island.
The chaining problem
Consider a manufacturing company forecasting monthly production costs. The cost depends on:
- Material prices (which fluctuate)
- Energy costs (seasonal patterns)
- Predicted demand (which drives volume)
You could build one massive model with all these factors. But that’s often impractical:
- Different data frequencies (daily prices, monthly demand)
- Different data sources (supplier APIs, internal forecasts)
- Different update cycles (demand forecasts run weekly)
- Different teams owning different predictions
What you actually want is to use your demand forecast as an input to your cost forecast. Chain the models.
Linked projects in Kanva
Kanva now supports cross-project predictions. You can link projects together and use one model’s predictions as features in another.
Here’s how it works:
1. Build your upstream model first
Create and train your demand forecasting project as usual. Once it’s working, it becomes available as a data source for other projects.
2. Link it as a driver
In your cost forecasting project, add the demand project as an “external driver.” Kanva treats its predictions like any other feature—available for training and inference.
3. Automatic resampling
Different frequencies? Kanva handles it. If your demand forecast is weekly but your cost model runs monthly, the predictions get resampled automatically.
4. Recursive prediction
When you run predictions, Kanva chains the calls. Need next month’s cost forecast? It first generates next month’s demand prediction, then uses that as input.
Why this matters
Cross-project predictions unlock workflows that were previously awkward:
Hierarchical forecasting. Predict at the product level, aggregate to category, then to total. Each level can be its own project with its own features.
Sequential dependencies. Model A predicts customer acquisition. Model B predicts their first-month behavior. Model C predicts lifetime value. Chain them together.
Team collaboration. The demand team owns their forecast. The finance team uses it in their cost model. Each team controls their piece, but the pieces connect.
Interpretability compounds. Each model in the chain has its own explanations. You can trace a final prediction back through the entire sequence of reasoning.
The practical details
Some specifics on how linking works:
- Linked projects appear in the “External Drivers” section of feature engineering
- You choose which output columns to include
- Resampling uses sensible defaults (you can override)
- The dependency graph is visible in project settings
- Circular dependencies are prevented
This feature works with all project types—forecasting, classification, and regression.
Getting started
If you have projects that logically depend on each other, try linking them:
- Open your downstream project (the one that should use another project’s predictions)
- Go to Feature Engineering > External Drivers
- Select the upstream project to link
- Choose which predictions to include
- Re-train with the new feature
The models now work together. And when you make predictions, the chain runs automatically.
Cross-project predictions are available now. Questions? Reach out at hello@human-driven.ai