Mapping universal causation required architecture no one had built before. So we built it.
The world’s first Large Causal Architecture.
The breakthrough technology that makes enterprise consciousness possible.
The intelligence
Behind the intelligence.
FOUNT – Large Causal Architecture
Most forecasting tools tell you what happened. FOUNT tells you why – and what comes next.
Built on a groundbreaking transformer-based architecture, FOUNT goes beyond spotting patterns in data. It understands cause and effect, uncovering which factors drive change in others, not just which ones tend to move together.
Neural Data Integration & Causal Processing
FOUNT brings together data from across your business — sales, weather, customer behaviour and more — and makes sense of it all in one place. Its transformer-based neural network embeds causal relationships directly into the data from the start, learning shared patterns across different data types while keeping track of what makes each one unique. The result is a unified understanding of your data that never loses sight of how different variables actually influence each other.
Finding Cause-and-Effect Relationships
FOUNT doesn’t just spot correlations – it finds the real drivers. Following Judea Pearl’s causal inference framework, it identifies which factors cause changes in others, models what happens when you deliberately shift one variable, and answers “what if” questions about scenarios that haven’t happened yet. Causes before effects. Always.
Large-Scale Pattern Recognition
Trained on millions of multivariate data points across diverse domains, FOUNT breaks down data silos and learns from the combined, interconnected impact of every influencing factor. At this scale, hidden synergies and emergent effects become visible — the kind that smaller models simply can't see. The more data it sees, the sharper and more reliable the signal.
- Uncovers hidden variable interactions invisible to smaller-scale approaches
- Separates genuine causal effects from random noise
- Learns universal patterns and domain-specific nuances simultaneously
Cross-Domain Learning & Adaptive Fine-Tuning
FOUNT trains across multiple domains at once, identifying common patterns and transferring knowledge between them. It understands the universal principles that cut across industries, while staying sensitive to what makes each domain unique. And when you need precision, its hierarchical fine-tuning adapts the model to your specific context — your data, your business, your goals — without losing any of the foundational intelligence built through shared learning
Large-Scale Causal Foundation Models:
Benchmarks broken.
The M5 competition: the world’s most rigorous forecasting test.
The winners improved benchmarks by 20-22%.
We beat the winners.
Then we beat everyone else.
All major time series competitions.
All major competitors — from siloed analytics to traditional AI.
One single model outperforming hundreds.
| Model | WMRSSE |
| Poem365 | 0.515 |
| IN_STU | 0.5204 |
| Matthias | 0.5281 |
| TS Mixer (Google) | 0.568 |
| TFT | 0.579 |
| DeepAR (Amazon) | 0.611 |
How Fount thinks.
Variables don’t exist in isolation. Everything influences everything else.
Fount captures these interconnected relationships — the ripple effects, the feedback loops, the evolving patterns that traditional models can’t see.
It adapts as systems change, learning new causal patterns in real-time.
- Cross-Domain Causation Modeling
Cross-industry variable influence. - Interconnected Factor Analysis
Ripple effects others miss. - Dynamic Relationship Learning
Adapts as systems evolve. - Explainatory Intelligence
Reveals why, not just what.
Under the hood.
Four breakthrough technologies unified into one architecture. Together, they amplify each other.
The result: Intelligence that doesn't just process your data, it understands your business.
CAUSAL AI
Actual causation
Ascends Pearl’s causal ladder by enabling interventional ‘what-if’ reasoning through counterfactual analysis, moving beyond traditional pattern recognition to understand true cause-and-effect relationships.
DEEP LEARNING
Hidden interconnected patterns
Captures complex multivariate non-linear relationships and interconnected impacts across large datasets, effectively handling multiple KPIs and intricate variable dependencies.
MULTIVARIATE OPTIMIZATION
The optimal state
Simultaneously optimizes multiple interconnected marketing variables to maximize ROI across the entire mix, considering complex factor interactions for holistic business outcomes.
AGENT SWARM ARCHITECTURE
Unified intelligence
Deploys specialized AI agents working collaboratively on different analytical aspects, leveraging collective intelligence that exceeds individual AI capabilities for complex problem-solving.
Trained on
250
billion
transactions
$5 Trillion
in spend data
Real-time
daily data collection
Over a
trillion
data points
Continuous learning
from every new data signal
See what Data Poem can do for you.
Let’s talk about how we can help you grow your business.