Most models assume one stable relationship between spend and revenue (one ROAS curve, one elasticity forever).
Reality disagrees:
Using a single fixed relationship leads to strategic lag.
"Are we operating under the same rules as before?"
"Has channel responsiveness structurally changed?"
"Did saturation fundamentally alter returns?"
"Are past learnings still valid?"
"Should we re-learn or keep optimizing?"
Markov Switching Regression assumes there are multiple hidden regimes, and the system switches between them probabilistically.
Instead of forcing one truth, it allows different slopes, sensitivities, and behaviors across phases.
Think of it as: “Different playbooks for different chapters.”
The Problem
Scaling rules stop working suddenly.
What It Reveals
Distinct regimes before and after saturation.
Decision Enabled
Stop scaling tactics that belong to the past.
The Problem
Performance breaks without explanation.
What It Reveals
Structural change in channel response.
Decision Enabled
Relearn instead of over-optimizing.
The Problem
Early growth playbooks fail later.
What It Reveals
Transition from expansion to efficiency regimes.
Decision Enabled
Switch strategies on time.
The Problem
Same campaigns stop working over time.
What It Reveals
Regime shifts in response curves.
Decision Enabled
Refresh strategy, not just creatives.
The Problem
Forecasts rely on outdated relationships.
What It Reveals
Which regime assumptions are active.
Decision Enabled
Avoid extrapolating the wrong past.
SpendSignal uses this model as a relationship integrity checker.
Specifically:
This prevents the silent killer: Optimizing perfectly for a world that no longer exists.
Instead of “Channel elasticity = 0.8”, you see:
The decision insight: "Stop scaling. The rules changed."
No. HMMs detect states; this changes the regression itself.
Yes. Transitions can occur both ways.
Only if abused. SpendSignal constrains regime count tightly.