Published in


Using Census Data For Advertising Campaign Funnel Analysis

In a privacy centric advertising world, the insights are pivoting from individuals to audiences. In some ways, not all, that macro approach is better, especially when it comes to optimizing the conversion funnel.

In a typical funnel, an impression is served, it either engages the user or not, and then in a perfect world the landing page seals the deal and results in a conversion. Google has invested tremendously in AI to optimize the impression, but the creative and landing page remain an art as much as a science.

What if we add a bit more science to the art of creatives and landing pages?

Thats where Segmentology comes in. It turns conversion funnel performance data from Google Ads, DV360, or CM360 into a performance map of hundreds of consumer characteristics. Analysts can quickly identify which user attributes correlate with success or failure at every stage of the funnel. The creative team can directly use these attributes to build creatives and landing pages that will resonate with the users they are trying to convert.

Currently Segmentology leverages US Census data to create maps of user attributes. Each stage of the funnel has its own attribute map, one for clicks which relate to creatives, and the other conversions which relate to landing pages. Filters are provided in the dashboard to enable comparing maps of various campaigns.

The map has 4 quadrants with different meanings, in this case evaluating clicks. The same meanings apply to the conversion map where the sublect is the landing page attributes.

  1. Upper Left low impressions / high clicks
    User attributes that relate to the creative but are not being served.
  2. Upper Right high impressions / high clicks
    User attributes that should be more present in every creative.
  3. Lower Right high impressions / low clicks
    User attributes that do not relate to the creative but are being served.
  4. Lower Left low impressions / low clicks
    User attributes that are probably outside the brand.

Additionally Segmentology provides a breakdown of the conversion funnel so the Creative Team can analyze exactly which user characteristics are causing a fall off in performance.

The perfect conversion funnel. These creatives and landing pages are working like magic. However, does the brand match the user attributes? Should the brand or the campaign change?

The conversions and by extension landing pages are failing to close the deal. How can the landing pages be tuned to appeal to these demographics more?

The landing page is great, but the clicks and by extension creatives are the weak link in the middle for these user attributes. How can the creatives be tuned to appeal to these demographics more?

Whats the data science?

The core of Segmentology is a statistically significant correlation between key performance indicators from a campaign and the US Census.

  1. A KPI report is downloaded from Google Ads, CM360, or DV360.
  2. The report impressions, clicks, and conversions are normalized as percent.
  3. The Census data is normalized as percent and categories are added.
  4. A correlation over Postal Code is done between the report and Census.
  5. The correlation is tested for significance using Pearsons Test.
  6. All correlations are reduced to:
    1 = positive correlation
    0 = no correlation
    -1 = negative correlation
  7. The dashboard counts all correlations in each category to map attributes.

The KPIs are impressions, clicks, and conversions, but can be extended to include store visits, or online orders if that data is available. There are over 22K postal codes in the USA and over 220 dimensions in the US census, resulting in a correlation across 5M data points within Segmentology.

In some cases the correlation does not produce any statistically significant results. Try increasing the number of days for the KPI reports or the scope of the campings included. No correlation is a valid answer, it signals that the brand is not exhibiting any affinity towards specific user attributes.

Whats the infrastructure?

Technologies like BigQuery and DataStudio are now allowing the processing display of massive amounts of information in a single chart. Combined with Python to help move the data between APIs, Segmentology leverages both to display hundreds of dimensions combined with interactive dropdown filters that recalculate the data on demand.

How is it deployed?

Segmentology is a standard StarThinker recipe with 5 deployment options:

  • AppEngine — Full browser based UI with multiple user login.
  • Cloud Function — Scheduled deployment to Google Cloud Functions.
  • AirFlow — Run recipes using cloud composer / airflow framework.
  • Colab — Quick browser deployment using Google Colaboratory.
  • Developer — Full test suite in a local development machine.
  • Command Line — Quick easy to install utility for running recipes.
  • PIP Package — Use StarThinker as a Python module.

For a quick start, leverage the Colab Deployment by starting on these pages for Google Ads, CM360, and DV360.

How can it be customized?

The solution is built as an example of a pattern that can easily be extended to datasets outside the US Census, including:

In addition the solution itself can be modified at several layers:

  1. Dashboard — completely editable, remove or add pages, filters, or charts. When adding a KPI dimension, copy and paste existing charts, then replace the dimension to preserve formatting and logic. Be aware of filters applied to each chart.
  2. BigQuery DataSet — StarThinker recipes do not overwrite existing views allowing them to be customized with new SQL logic or JOINS even when running in a scheduled deployment. Segmentology can be completely customized within BigQuery dataset via views.
  3. Recipe — Each StarThinker recipe, in any deployment, can be extended with additional tasks from the Solution Gallery or Script Library. The Segmentology recipe includes many configuration settings for statistical significance thresholds, fields to correlate, and values to use as filters.
  4. Python — The Census normalize and correlate tasks are defined in a open source python package, which can be modified.
  5. Tests — A test recipe can be quickly deployed via StarThinker framework.

How will it impact the business?

Crafting compelling creatives and engaging landing pages are arguably two of the most critical steps of the conversion funnel. They are also the two steps in the user journey most under the control of the advertising agency. Segmentology’s extensible offering across Google Ads, CM360, and DV360 enables a great conversion funnel offering to any advertising partner.



At gTech, we believe every ad operations team should be faster, nimbler, and able to use all their data sources to drive client impact. To that end, we’ve created StarThinker, a simple and intuitive web UI that allows users to create, edit, run, and schedule data pipelines consis

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store