Data Processing

Learn how we transform raw data into insights for our customers and partners.

CPM Prediction and Estimating Spend

Introduction

We estimate specific cost per thousand impressions (CPM) rates for each creative via our proprietary models.

By incorporating direct buy and programmatic data we tune our CPM estimates for size, position, time, type, and marketplace dynamics. Our models use machine-learning to assign a CPM to each impression of each creative.

Desktop and Mobile (Display and Video) CPM

We use the following inputs to predict the CPM of each creative:

  • Location on the page (above vs. below the fold)
  • Creative size (e.g. 300x250)
  • Creative type (display, video, etc.)
  • Site
  • Direct vs. Indirect (Programmatic)
    • Direct: This model is trained using actual rate cards from agencies
    • Programmatic: This model is trained using actual programmatic clearing prices from buys made by Pathmatics on tracked sites

Social and OTT CPM

Each quarter we project Facebook, Instagram, Twitter, and OTT spend, impressions and CPM based on historical quarterly earnings reports and other industry sources. We use a number of the most reliable industry sources who estimate CPMs on each of the social channels we track and OTT to determine our initial CPM estimates.

Each quarter, we review the global year over year change in price per ad reported in Facebook’s earnings report along with details in Twitter’s earnings report to confirm our CPMs follow a similar pattern.

We also update history for the previous quarter to account for differences between our projected figures and what is announced in each social platform's quarterly earnings.

Estimating Spend

Each impression served to our crawlers and panelists is assigned a CPM, which when combined with impressions results in our spend estimates.

Was this article helpful?

Have more questions? Submit a request