How OptiMix Works

OptiMix uses Automatic Differentiation Variational Inference (ADVI) — a modern approximate inference algorithm — to estimate marketing channel effects without requiring you to write statistical code.

Unlike Markov Chain Monte Carlo (MCMC) methods, which can take hours or days to converge, ADVI runs in minutes on a standard laptop. The result is a full posterior distribution over every channel contribution — credible intervals, not just point estimates.

The core model is a hierarchical Bayesian regression estimating how each channel contributes to your target KPI (revenue, leads, or conversions), accounting for:

• -Baseline (Organic) — sales driven by factors outside paid media
• -Saturation — diminishing returns at higher spend levels
• -Adstock — carry-over effects that decay over time
• -Trend & Seasonality — macroeconomic patterns in your data
• -Control Variables — price, promotions, competitor activity

Every coefficient has a full posterior distribution. Every assumption is inspectable. The model reports its own uncertainty — so you know when to trust a recommendation and when you need more data.

OptiMix never requires tracking pixels, customer-level data, or raw transactions. The model runs on aggregated weekly spend and revenue data — no PII, no cookies, no third-party sharing.

This makes OptiMix compliant with GDPR and CCPA out of the box.

OptiMix requires a minimum of 26 weeks of weekly spend and revenue data across two or more marketing channels — roughly six months of history. Most SMBs with any paid digital presence already meet this.

26+ weeks → tight credible intervals. 52+ weeks → seasonal patterns distinguishable from channel effects. 2+ years → trend vs. media impact separated cleanly.