What-If Scenarios

Coming Soon

Test counterfactual interventions against live simulations. Ask Shadow "what would happen if…" and receive projected outcomes with full confidence bounds — before making any changes to production.

Example Scenarios

Based on the current Yelp demo simulation, here are scenarios you could test:

Reduce iOS Verification Gating

What happens to first-review rates if we remove the 2 extra verification steps for iOS reviewers on new NYC listings?

Early Review Drop-off
L1 DeterministicL2 Statistical
+4.2pp first-review rate · iOS NYC
Coming Soon

This scenario will be available in the next release

Review-Request Nudge

Simulate emailing consumers who viewed but didn't review a newly listed restaurant within 48 hours.

Early Review Drop-offWeb Activation Stall
L2 StatisticalL3 AI Simulation
+2,800 incremental reviews / 30d
Coming Soon

This scenario will be available in the next release

Restore Android Notifications

Model the recovery trajectory if Android notification delivery returns to Q3 levels via an improved permission flow.

Android Re-engagement Decay
L2 Statistical
+1.8pp review rate · Android LA
Coming Soon

This scenario will be available in the next release

Mobile-Web Re-engagement

What if we add a lightweight mobile-web re-engagement path for consumers who bounce before leaving a first review?

Web Activation Stall
L1 DeterministicL2 Statistical
+3.1pp conversion · web
Coming Soon

This scenario will be available in the next release

How What-If Works

Counterfactual Branching

Fork any active simulation at a specific point in time and apply a hypothetical change. Shadow computes the divergent trajectory using the same statistical models.

Side-by-Side Comparison

Compare the baseline projection against one or more what-if scenarios. See divergence points, confidence intervals, and net impact in a unified view.

Traceable To A Layer

L1 bounds what the system makes possible, L2 computes the divergent trajectory. L3 explains the result in words when you ask — it never decides the outcome. Every scenario shows which layer produced what.

Explicit Uncertainty

Every scenario includes confidence bounds. If a what-if introduces too much uncertainty, Shadow flags it as SPECULATIVE rather than presenting false precision.

Scenario Simulation Flow
1Define InterventionWhat change to test
2Select SegmentsWhich cohorts to simulate
3Run SimulationL1 → L2 · L3 on request
4Compare OutcomesBaseline vs scenario
5Export DecisionShare with stakeholders
Shadow — know what happens next, before it happens