Causal action models

Know which action
changes the outcome.

Randomize price, product, and message on a simulation of your market. Validate weekly against real behavior.

Test a decisionInspect the evidence
0.79–0.88Mintel validationSimilarity to Mintel’s 2,000-person conjointWeeklyHuman checkValidated against real behaviorGradedReport confidenceConfidence and limits with every report

Causal action brief · illustrative

Which price × message increases choice?

Illustrative Price by Message experiment conditions
PriceValueProofSpeed
$29$2M$8M$14M
$49$11M$56M$24M
$69$5M$31M$18M

Used inside category leaders.

Independent check0.73Mean rank agreement43 design-filtered human studies across 9 domains. Failures included.Read the paper

// What we do

Start with the decision.
The method follows.

What should we charge? Can we take an increase?Discrete choice with priceWillingness-to-pay, demand curve, revenue-optimal price
Which features and claims actually earn their keep?MaxDiff / best–worstRanked priorities, and where they differ by segment
What should the product or pack be?Build-your-own + adaptive choiceThe winning configuration, and what it costs to serve
Why should a retailer give us the shelf?Category shelf simulationSource of volume: category growth vs. brand switching
How many will they buy, and how often?Volumetric choiceUnit forecasts, basket and repeat effects
Which message or concept wins, and why?Concept test + complete rankingRanked concepts with the reason-why behind each
Who do we go after first?Preference-based segmentationSegment size, value, and the lever that moves each
What if a competitor cuts price next quarter?Market simulator on the fitted modelShare and margin under any scenario you type in

Anyone can ask. Every completed decision makes the next report better. No consultants. No inbox full of static reports.

A self-improving
business researcher.

  1. Randomize

    The action.

  2. Estimate

    Effect and interval.

  3. Validate

    Real behavior.

  4. Update

    The causal map.

Recursive self-improvement.

Each answer is checked against real behavior, then updates the shared causal model for every team.

Randomize an action, estimate its effect, validate it against real behavior, and update the causal map before the next decision.

What decision are you
about to make?

// Evidence record

Proof compounds.

Four public checkpoints. Each changes what the system learns next.

  1. 2023

    Program

    Selected by UChicago.

    Selected for UChicago’s Transform Cohort 2, run by the Polsky Center and Data Science Institute.

    UChicago · Transform Cohort 2

  2. 2024

    Validation

    The validation record began.

    43 design-filtered human studies across nine domains. Failures included; working paper, not peer reviewed.

    Read the working paper

  3. 2025

    Market shift

    Simulation became continuous research.

    a16z calls simulated societies a continuous, dynamic advantage over one-time research.

    a16z · AI market research

  4. 2026

    Recursive self-improvement

    World simulation becomes the research interface.

    Autonomous AI research agents run experiments, learn from outcomes, update the shared causal model, and use that evidence in the next decision. Gartner named Subconscious one of four leaders in causal modeling.

    Gartner · G00838481