Enterprise causal analysis. Beyond marketing.
The operational dynamics your stack cannot see.
We run causal inference on the dataset of your choice and surface the dynamics your existing tools cannot reach. Marketing, supply chain, churn, manufacturing yield, fraud, retention. One engine, your data, any operational question.
Three platforms, three answers, one budget to defend. Your GA4 export already holds the answer, and the first finding is free and never leaves your browser. Get your first finding free
Your dashboards show what happened. We show what caused it.
The questions your existing tools cannot answer.
Every operational dashboard you own is correlational by design. Causal inference answers the questions that come after the dashboard, the ones that actually move budgets and roadmaps.
Marketing
Which channel actually drove the revenue, and which one was just claiming credit?
Supply chain
Which upstream signal causes the late-delivery cascade three weeks downstream?
Churn
Which product behavior precedes churn 60 days out, and is the predictor a cause or a co-symptom?
Manufacturing
Which line variable is causally responsible for the yield drop, vs the variables that just correlate with it?
Retention
Which onboarding step causally lifts week-12 retention, vs steps users complete because they were going to stay anyway?
Fraud
Which behavioral pattern actually causes a fraud loss, vs the patterns that show up alongside it because of selection bias?
Pricing
If we raise this SKU 8%, how much volume do we actually lose, controlling for season, competitor, and substitution?
Workforce
Which management practice causes attrition to drop, vs the practices that look correlated because high-performers self-select?
Whatever the variable, the math is the same. Tell us your operational question; we will tell you if it is causally answerable from your data.
One engine. Any dataset. Any operational question.
The math does not care about your vertical. The same causal framework, applied domain by domain:
Drug Research
Does this compound cause the therapeutic effect, or is something else going on?
- Data in:
- Patient cohorts, dosage, biomarkers, adverse events
- Insight out:
- Causal effect estimation, confounder identification
Faster Phase II signal detection. Fewer expensive dead ends.
Sports Performance
Which training interventions actually cause performance gains?
- Data in:
- Training load, biometrics, nutrition, match data
- Insight out:
- Causal training effects, injury risk isolation
Know the difference between overtraining and peaking. Before game day.
Supply Chain
What actually causes stockouts? Not what correlates. What causes them.
- Data in:
- Lead times, inventory, demand signals, disruption events
- Insight out:
- Root cause of disruptions, causal demand forecasting
Stop firefighting symptoms. Fix the cause.
Manufacturing
Which process parameters cause defects? Which ones just look guilty?
- Data in:
- Production sensors, raw materials, environment, defect logs
- Insight out:
- Causal root cause analysis, process optimization
Fewer defects, because you fixed the right variable.
Finance & Risk
What causes drawdowns? What causes defaults? VaR will not tell you.
- Data in:
- Market data, positions, macro indicators, credit data
- Insight out:
- Causal risk factors, stress testing, intervention analysis
Risk models that explain why, not just how much.
Energy & Emissions
Which interventions actually cause emissions reductions? Which ones just look good in the report?
- Data in:
- Energy consumption, output, weather, equipment, monitoring
- Insight out:
- Verified causal impact, CSRD-ready reporting
Prove your sustainability claims. With math, not marketing.
Waste Processing
Which variables cause sorting yield to change? And by how much?
- Data in:
- Sensor data, throughput logs, material composition
- Insight out:
- Root cause of contamination, throughput bottlenecks
Illustration: a 1% purity gain across 100K tons a year. Do the math on that.
Marketing Attribution
Which channel actually causes incremental sales? Not which one was last-clicked.
- Data in:
- Ad spend, GA4, Shopify conversions
- Insight out:
- True channel contribution, budget reallocation
Stop crediting channels that just happened to be nearby when someone converted.
Four steps. Your data stays where it is.
Every engagement follows the same causal framework, regardless of the dataset. The domain changes. The rigor and the security posture do not.
01
Your dataset
Marketing logs, supply chain telemetry, customer events, manufacturing sensor feeds, HR systems, financial transactions. If it has variables and outcomes, we can work with it. Data residency on your terms: VPC, on-prem read replica, or anonymized export. No SDK, no 90-day onboarding.
02
Causal model
We build a directed acyclic graph that maps the actual cause-and-effect structure of your operation. Not a regression. Not a correlation matrix. An auditable map of what causes what across the variables you care about.
03
Causal graph
X causes Y. Z is a confounder. W is a mediator. Every relationship quantified. Every claim scored for confidence and identifiability. The kind of answer that survives a board challenge.
04
Intervention
Change X by this amount, expect Y to change by this amount. Prescriptive. Actionable. With sensitivity bounds. The answer your COO or Chief Data Officer needs to commit to the budget cycle.
The engine is the constant. Your domain is the variable.
The pilot
The Causality Audit. 90 days. One operational question.
Your dataset of choice. Our engine. You pick the question and the success criteria. We agree on them in writing before any data moves. Miss the criteria and the engagement closes there. No annual conversion, no extension. The math is the math.
Week 1-2
Ingest
Your data, your terms: VPC, read-only access, anonymized export, whatever your security review accepts. We map the candidate causal graph and pin down what your current analytics think is happening. This is the baseline. We either beat it by week 12 or the engagement closes there.
Week 3-6
Discover
The engine separates causal signal from correlated noise. We surface relationships your dashboards do not show because dashboards are not designed to: confounders, mediators, the variables that look causal until you control for the right thing. Every claim quantified, every claim auditable.
Week 7-10
Validate
We do not just model, we test. A/B experiments, holdout groups, natural experiments inside your historical data. Model predictions vs. observed reality. Every causal claim gets a confidence score and a sensitivity range. No hand-waving.
Week 11-12
Prescribe
Change X by this amount, expect Y to change by this amount, with bounds. Scenario modeling. Executive summary your COO can sign. Regulator-ready when applicable. The kind of deliverable that ends the meeting because the answer is obvious.
- Duration
- 90 days
- Risk
- No annual lock-in
- Conversion
- 100% credited to annual
Your data, your terms
We adapt to your security review, not the other way around.
Three integration patterns, picked in the kickoff. Whichever your team is comfortable with is the one we use.
VPC / Private cloud
We deploy the engine inside your AWS, GCP, or Azure VPC. Your data never leaves your perimeter. Audit logs in your SIEM.
Read-only replica
You provision a read replica of the relevant tables. We connect through your existing VPN or IP-allowlisted endpoint. Zero write access ever.
Anonymized export
You ship a hashed-key, aggregated CSV or Parquet. The model runs causally on the structure without ever seeing identifiable rows.
- Data residency
- EU primary, regional on request
- Encryption
- TLS 1.3 in transit, AES-256 at rest
- Audit
- SOC 2-aligned, GDPR-compliant
- Egress
- None unless explicitly granted
Frequently asked questions
What is a causal inference platform?
A causal inference platform uses statistical and machine learning methods to determine cause-and-effect relationships from data, rather than just identifying correlations. The Causality Engine constructs directed acyclic graphs (DAGs) to map which variables cause changes in outcomes, identifies confounders and mediators, and validates causal claims through experimentation. This goes beyond traditional analytics which can only show that two variables move together, not that one causes the other.What is the difference between correlation and causation in data analytics?
Correlation means two variables move together. Causation means one variable directly causes a change in another. Most analytics tools, dashboards, and attribution platforms report correlations and present them as if they were causal. The Causality Engine uses causal discovery algorithms, directed acyclic graphs, and incrementality testing to prove actual cause-and-effect relationships. This distinction matters because acting on correlations can lead to wasted resources, while acting on causation leads to predictable outcomes.What does an enterprise engagement with Causality Engine actually deliver?
An enterprise engagement runs causal inference on the corporate dataset of the customer's choice and surfaces the operational dynamics their existing analytics stack cannot see. Beyond marketing attribution, this covers supply chain root cause, churn causation vs co-symptom, manufacturing yield drivers, fraud pattern causality, pricing elasticity controlling for confounders, employee retention causation, and any other multivariate operational question. The deliverable is a quantified causal graph with confidence and sensitivity bounds, plus prescriptive interventions the COO or Chief Data Officer can act on in the next budget cycle.What industries can use causal inference?
Causal inference is domain-agnostic. Any operation with multivariate data and outcomes can benefit. The method applies to marketing attribution, supply chain operations, manufacturing yield, customer churn and retention, fraud detection, pricing and demand elasticity, workforce attrition, drug research, sports performance, energy and emissions, and waste processing. The math does not care about the vertical; what matters is whether the operational question is causally identifiable from the available data.What is the Causality Audit?
The Causality Audit is a 90-day enterprise pilot engagement where Causality Engine runs causal inference on the customer's chosen dataset to answer one high-value operational question. It follows four phases: ingest the data on the customer's security terms (VPC, read-only replica, or anonymized export) and map the candidate causal graph (Week 1-2), discover causal relationships separating signal from correlation noise (Week 3-6), validate every claim with A/B experiments and holdout groups (Week 7-10), and deliver prescriptive recommendations with confidence and sensitivity bounds (Week 11-12). Success criteria are defined upfront in writing. On success, 100% of the pilot fee credits toward an annual contract. If criteria are missed, the engagement closes there with no annual conversion and no extension; the buyer is not locked into anything beyond the pilot fee.Where does Causality Engine run on enterprise data?
Three integration patterns, picked in the kickoff. (1) VPC / private cloud: the engine deploys inside the customer's AWS, GCP, or Azure VPC; data never leaves their perimeter; audit logs flow into their SIEM. (2) Read-only replica: the customer provisions a read replica of the relevant tables; we connect through their existing VPN or IP-allowlisted endpoint; zero write access. (3) Anonymized export: the customer ships a hashed-key, aggregated CSV or Parquet; the model runs causally on the structure without ever seeing identifiable rows. Data residency defaults to EU primary, with regional placement on request. TLS 1.3 in transit, AES-256 at rest, SOC 2-aligned controls, GDPR compliant.How does the Causality Engine work?
The Causality Engine follows four steps: (1) Data Ingestion - any multivariate dataset from sensor feeds, transaction logs, clinical data, or ad spend; (2) Causal Model - the engine constructs a directed acyclic graph (DAG) of cause and effect, not a regression or correlation matrix; (3) Causal Graph - every relationship is quantified with X causes Y, Z is a confounder, W is a mediator, all scored for confidence; (4) Intervention - prescriptive recommendations stating if you change X by this amount, expect Y to change by this amount. The domain changes but the mathematical rigor does not.What is a directed acyclic graph (DAG) in causal inference?
A directed acyclic graph (DAG) is a mathematical structure used in causal inference to represent cause-and-effect relationships between variables. Each node represents a variable, and each directed edge represents a causal relationship. 'Acyclic' means there are no feedback loops. The Causality Engine automatically constructs DAGs from your data using causal discovery algorithms, identifying which variables cause changes in outcomes, which are confounders (common causes), and which are mediators (intermediate steps in a causal chain).How is causal inference different from machine learning?
Traditional machine learning excels at prediction (what will happen) but cannot answer causal questions (what causes what, and what happens if we intervene). A machine learning model can predict that customers who see a TikTok ad are more likely to buy, but it cannot tell you whether the ad caused the purchase or whether those customers would have bought anyway. Causal inference, as implemented by the Causality Engine, answers the intervention question: if we change X, what happens to Y? This is the difference between forecasting and understanding.Can the Causality Engine be used for drug research and clinical trials?
Yes. The Causality Engine can ingest patient cohort data, dosage information, biomarkers, and adverse event records to determine whether a compound causes a therapeutic effect or whether the observed effect is confounded by other variables. This enables faster Phase II signal detection and fewer false positives by separating true causal drug effects from spurious correlations in clinical data.How does causal inference apply to supply chain optimization?
In supply chain management, the Causality Engine ingests lead times, inventory levels, demand signals, and disruption events to identify the root causes of stockouts and cost overruns. Rather than reacting to symptoms, the platform identifies what actually causes disruptions, enabling preventive action. This includes causal demand forecasting that accounts for true drivers rather than correlated signals.What data does the Causality Engine need?
The Causality Engine works with any multivariate dataset that has variables and outcomes. This includes sensor feeds, transaction logs, clinical trial data, advertising spend data, production line sensors, energy consumption records, financial market data, and more. Data can be ingested via APIs, CSV uploads, or direct database connections. There is no requirement for a specific SDK or pixel installation, and no 90-day onboarding period.
Who runs the pilot
Joris van Huët
Founder & CEO · Utrecht, Netherlands
Founder and CEO of Causality Engine. Builds causal-inference attribution software for Ecommerce brands on GA4. Runs every demo himself.
Speaks about: Causal Attribution · ecommerce · Digital Marketing
Tell us the question.
We will tell you if it is causally answerable.
No sales deck, no demo theater. One conversation about your operational problem, your dataset, and the success criteria we will agree to in writing before any data moves.