Granularity | Definition
Granularity in marketing mix modeling (MMM) refers to the level of detail at which data is aggregated across both temporal dimensions (daily, weekly, monthly) and structural dimensions (channel, campaign, creative, geography), with these choices fundamentally affecting model sensitivity, accuracy, and practical utility for decision-making. The granularity decisions made during model design create tradeoffs between statistical power, operational complexity, and strategic insight depth that profoundly influence what questions the model can reliably answer.
Temporal granularity determines how quickly models detect changes and respond to market dynamics. For example, daily granularity enables detection of weekend vs. weekday patterns and rapid response to campaign launches, but requires 365 data points per year compared to 52 for weekly data—potentially creating noise and instability if marketing activities don’t vary sufficiently day-to-day. A retail brand running always-on digital campaigns might find that daily data captures meaningful variation, while a B2B software company with quarterly campaign flights would see mostly noise in daily data and benefit from weekly or monthly aggregation.
The choice must balance the need for timely insights against statistical requirements for stable coefficient estimation, with consideration for how quickly marketing effects manifest through lag effects and adstock decay.
Structural granularity determines which strategic questions models can address. Channel-level granularity (all digital, all television) enables broad portfolio allocation but cannot distinguish display from video within digital. Campaign-level granularity reveals which specific initiatives drive results but requires substantially more data and risks multicollinearity when campaigns run simultaneously. Geographic granularity enables regional budget optimization but demands sufficient sales volume in each region to detect marketing effects above baseline noise.
Take the case of a national retailer with 200 stores: They might model at DMA-level granularity for the top 20 markets generating 70% of revenue, while aggregating remaining markets into regions—balancing actionable insight for major markets against statistical stability for smaller ones.
The strategic challenge lies in matching granularity to available data and business decisions. Finer granularity always sounds appealing but often degrades model quality when data becomes too sparse to reliably estimate effects. Successful MMM implementations often use weekly temporal granularity and channel-level structural granularity as a baseline, adding finer detail only where business decisions genuinely require it and data volume supports robust estimation.
Kochava MMM’s hierarchical modeling framework enables flexible granularity that adapts to data availability, borrowing statistical strength across related units through partial pooling. This enables detailed insights where data supports them while maintaining stability through Bayesian modeling approaches that prevent overfitting in data-sparse contexts.
