1. Knowledge Base
  2. Platform Tutorials

Why INCRMNTAL Requires Spend to Measure Contribution - and How Promotions Are Different

Understanding why spend variation is essential for causal measurement - and how INCRMNTAL distinguishes between paid media and promotional triggers.

Why INCRMNTAL Requires Spend Per Channel

INCRMNTAL’s model is designed to measure the causal contribution of each marketing channel by analyzing how changes in spend affect outcomes. For the model to evaluate whether a channel is driving incremental value, there must be actual spend activity associated with that channel.

When a channel receives no spend (e.g., it’s provided as a 100% discount or barter), the model lacks the variation it needs to detect causality. With no financial input or changes to observe, the model cannot distinguish whether any resulting conversions were driven by that channel or would have occurred regardless. In other words, zero spend means zero signal, making it impossible for INCRMNTAL to generate a valid prediction for that channel’s impact.

To ensure channels are measurable, they must have some level of paid activity, even if small or intermittent, so the model can learn from fluctuations and generate reliable insights.

How INCRMNTAL Measures Promotions Without Spend

Even though promotions (like discounts, sales events, or in-app offers) don’t carry media spend, INCRMNTAL can still measure their impact due to the different way they affect user behavior.

Promotions typically cause sharp, time-bound, and significant changes in performance. These shifts are often detectable as causal events because they act like external interventions or 'shocks to the system'. INCRMNTAL treats these events as non-media factors and can incorporate them as contextual inputs in the model.

 

Key Differences Between Promotions and Zero-Spend Media

Feature

Promotions

Media Channels (with 0 spend)

Has direct media spend

❌ No

❌ No

Causes observable system shock

✅ Yes (behavioral trigger)

❌ No variation = no observable effect

Treated as model input

✅ Yes (as a contextual factor)

❌ No - not a measurable variable

Can be measured by INCRMNTAL

✅ Yes

❌ No