Causal inference platform for the physical world

Understand Normal. Explain Deviations.
Improve the Outcome.

Your enterprise has far more examples of what works than what fails: its richest, least-used operational intelligence. Parabole learns what drives normal performance, then uses causality to explain deviations and identify what can change the outcome.

CRUDE FEED CRUDE FEED FIRED HEATER FIRED HEATER ATMOSPHERIC COLUMN ATMOSPHERIC COLUMN OVERHEAD CONDENSER OVERHEAD CONDENSER PRODUCTS PRODUCTS DOWNSTREAM · REFINERY DOWNSTREAM · REFINERY SYS OK SYS OK DATA LINK DATA LINK RECORDING RECORDING PANEL OPERATOR PANEL OPERATOR NORMAL-BEHAVIOR MONITOR NORMAL-BEHAVIOR MONITOR LIVE LIVE CAUSAL RELATIONSHIP MONITOR CAUSAL RELATIONSHIP MONITOR RE-ESTIMATED CONTINUOUSLY RE-ESTIMATED CONTINUOUSLY FIRED-HEATER OUTLET TEMP FIRED-HEATER OUTLET TEMP green = learned normal green = learned normal ATMOS. COLUMN ΔP ATMOS. COLUMN ΔP OVERHEAD REFLUX RATIO OVERHEAD REFLUX RATIO ripples building →flaggedripples building →flaggedripples building →flaggedripples building →flaggedripples building →flaggedripples building →flaggedripples building →flaggedripples building →flaggedripples building →flaggedripples building →flaggedripples building →flaggedripples building →flagged NOW NOW ← earlier ← earlier DEVIATION DETECTED flow reverses flow reverses link dropped link dropped new coupling new coupling link dropped link dropped new path new path FIRED HEATER FIRED HEATER OVERHEAD COND. OVERHEAD COND. PUMP-AROUND PUMP-AROUND FEED PUMP FEED PUMP ATMOS. COLUMN ATMOS. COLUMN SIDE STRIPPER SIDE STRIPPER FEED PREHEAT FEED PREHEAT REFLUX DRUM REFLUX DRUM BOTTOMS PUMP BOTTOMS PUMP STRUCTURE MATCHES THE LEARNED MODEL STRUCTURE MATCHES THE LEARNED MODEL COUPLING SURGE · heater outlet intensifying into column LOOP REVERSAL · reflux return direction flipped LINK DROPPED · condenser-to-drum coupling lost NEW COUPLING · heater bypassing column to condenser COUPLING WEAKENED · pump-around draw fading TOPOLOGY SHIFT · stripper path lost, new bottoms route learned link learned link changed / new changed / new
Control-room view: instability ripples through the signals as the causal structure shifts (links strengthen, reverse, or drop); that shift is the early indicator, ahead of the deviation flag

Illustrative view using simulated data. Refresh frequency is configurable.

Scroll down to explore why Parabole and causal inference
Why Parabole

Built for the decisions that have no precedent.

Many AI systems learn primarily from historical patterns. TRAIN also models the relationships that define normal operation, reducing dependence on examples of every failure mode.

01 / Cause, not catalog

Reads deviations without precedent

Parabole models the relationships that govern the process, then reads what breaks them. A deviation with no history in this plant or the fleet is still a signal, with a stated cause.

02 / One causal model

Procurement to logistics, held together

Recipe, first-pass yield, machine reliability, and material flow evaluated in a single causal model instead of systems that never talk to each other.

03 / In your environment

Deployed in your tenant, auditable by design

Customer data is not pooled across customers under the approved deployment and data-use terms. Existing control systems keep setpoint authority. Each supported platform-generated decision records its model version, inputs, causal path, and binding constraints.

Read why Parabole, in full →

Customers include global energy majors and sector leaders in industrials and manufacturing across Europe and the US, applying the causal-inference platform to complex production and operations problems in upstream, midstream and downstream.

30–40%
Lower subsurface prediction error against a kriging baseline.
25%
Faster pipeline leak detection through causal consistency monitoring.
2–5%
Fired heater energy-efficiency gain, with roughly 10% lower emissions.
30%
Decrease in unplanned shutdowns through alarm rationalization.
The platform

Learn normal. Everything else is a signal.

TRAIN holds what the data says the process does and what your engineers say it should do as one reference, then reads live conditions against it: modeling business as usual, detecting the residual, and explaining cause and consequence. See how the platform works →

Industries

One decision, owned by one team, to start.

Each engagement begins with a single decision that is hard to get right. None requires a catalog of past incidents.

Bring us the decision and we will walk through the causal model behind it.

Insights

Latest Insights.

Explore All Insights →

Get started

Bring us one decision that is hard to get right.

Bring the engineer who knows the process. We will walk through the causal model behind it.

Request a demo