Documentation
Documentation
Start typing to search…

Causal Attribution

Note: To use Causal Attribution, you must be subscribed to both MTA and MMM solutions. If you're not subscribed, please reach out to [email protected] for assistance.

Introduction

In the complex landscape of modern marketing, understanding the true impact of various marketing efforts is crucial for optimizing spend and maximizing return on investment (ROI). Causal attribution has emerged as a powerful tool in this pursuit, offering marketers a cross media and granular view of their marketing effectiveness.

Understanding Causal Attribution

At its core, causal attribution combines two powerful methodologies:

  1. Incrementality from Marketing Mix Modeling (MMM) and/or Experiments
  2. Granular, real-time numbers generated by Self-reported/ Tracked / Platform Attribution

By integrating these approaches, causal attribution offers a calibrated assessment of marketing performance by quantifying the true incremental impact of each ad channel (and associated tactics, campaigns, ad sets and ads) within the marketing mix. It does this by aligning platform-reported metrics with incrementality factors derived from a Marketing Mix Model or controlled experiments.

This method effectively addresses attribution bias, a common issue where marketing platforms often overestimate the impact of their own contributions.

It also addresses the problems of having to choose between different sets of rule-based attribution models, which all suffer from the problem of incomplete data (signal loss due to privacy restrictions, for example).

Causal attribution delivers a more balanced, objective view of performance across all channels, enabling a clearer understanding of what truly drives conversions.

The Importance of Causal Attribution

Causal attribution addresses several key challenges in marketing measurement:

  1. Generating right hypotheses based on fluctuations or anomalies in attribution metrics for Experiments

  2. Monitor incremental performance at granular ad set and ad levels near-real-time, allowing for quick tactical changes within a channel (compared to more strategic with the results of a marketing mix model or experiment)

Methodology of Causal Attribution

The core formula for causal attribution is:

Incremental Conversions = Incrementality Factor (from MMM or Experiments) * Platform-Reported numbers (pConversions)

Types of Conversions supported on Lifesight:

  1. Revenue
  2. Non-Revenue
    1. Orders
    2. Qualified Leads
    3. Signups/ Registrations

Let's break down the components:

  1. Incremental ROAS (iROAS):
    This is derived from Marketing Mix Modeling. It represents the additional return generated for each dollar spent on a particular marketing activity, above what would have occurred without that activity.

  2. Platform Reported ROAS (pROAS):
    This is the return on ad spend as reported by the advertising platform itself. It often includes all conversions that occurred after exposure to an ad, regardless of whether the ad truly caused the conversion.

  3. Incrementality Factor:
    By dividing iROAS and pROAS, we get a measure of the true incremental revenue generated by the marketing activity. This incrementality factor is helpful because it reveals the true additional value generated by a marketing channel beyond platform-reported figures, guiding more effective budget allocation and strategic decisions for genuine growth.

    1. An incrementality factor of 1 means the platform-reported ROAS is fully incremental – every dollar of return reported was genuinely caused by the ad.
    2. A factor less than 1 (e.g., 0.6) indicates that only a portion (e.g., 60%) of the pROAS is truly incremental. The rest would have likely occurred organically.
    3. A factor greater than 1 is rare but could suggest synergistic effects or under-reporting by the platform.

  4. Platform Conversions:
    The conversion numbers reported by the ad platform; this includes metrics like Revenue, Orders, Leads, Signups, Registrations, Add to Carts, App Installs and Custom Conversions.

To effectively visualize data on Lifesight, a benchmark must be defined. Lifesight utilizes two distinct benchmark categories:

📘

When setting a new benchmark for Causal Attribution, ensure you meet the following conditions:

  1. Integrate the ad channels you use
  2. At least 1 scenario adopted ( If you want to use a scenario as benchmark, else ignore)
  1. Scenario-Based: those derived from a scenario created using MMM Model
  2. Non-Scenario Based: Those created independent of scenarios. These consists of:
    1. mCPA (Registration) - Use this if you your KPI is Registrations
    2. mCPA (Leads) - Use this if you your KPI is Leads
    3. mROAS - Use this if you your KPI is Revenue
    4. iROAS
    5. mCPA (Orders) - Use this if you your KPI is Orders

The recommendations for adjusting spend at the ad channel level (campaign, ad set, or ad) are determined by the chosen benchmark category.

1. Scenario-Based Benchmarks

When using scenario-based benchmarks, recommendations are based on the percentage difference between the current spend and the optimized (or recommended) spend.

  • If the percentage change > +5%: The recommendation is to Scale Up Spend.
  • If the percentage change < -5%: The recommendation is to Reduce Spend.
  • Otherwise (between -5% and +5%, inclusive): The recommendation is to Maintain Spend.

Example (Scenario-Based):

Consider Tactic T1:

  • Current Spend = $10,000
  • Recommended Spend (Optimized Spend) = $15,000
  • Percentage Change = (($15,000 - $10,000) / $10,000) * 100% = +50%
  • Since +50% is greater than +5%, the recommendation is to Scale Up Spend by $5,000 (which is the difference, or 50% of the current spend). The same is shown for all the various campaigns, adsets and ads part of the tactic.

2. Non-Scenario Based Benchmarks

When using non-scenario based benchmarks, recommendations are determined by comparing the performance metric of a channel or tactic (such as Marginal Return on Ad Spend - mROAS, or Incremental Return on Ad Spend - iROAS) against a predefined benchmark value set in the benchmarks modal.

  • If the performance metric (e.g., mROAS or iROAS) for the channel or tactic is below the benchmark value set in the modal: The recommendation is to Reduce Spend.
  • If the performance metric (e.g., mROAS or iROAS) for the channel or tactic is above the benchmark value set in the modal: The recommendation is to Scale Up Spend.

Example (Non-Scenario Based):

Assume the benchmark MROAS set in the benchmarks modal is 3.5.

  • Scenario A: Tactic T2

    • Actual MROAS for Tactic T2 = 2.8
    • Since 2.8 (Actual mROAS) is less than 3.5 (Benchmark MROAS), the recommendation for Tactic T2 is to Reduce Spend.

  • Scenario B: Tactic T3

    • Actual iROAS for Tactic T3 = 4.2 (assuming iROAS is the chosen metric for this tactic)
    • Benchmark iROAS set in modal = 3.5 (assuming the same benchmark value applies or is set for iROAS)
    • Since 4.2 (Actual iROAS) is greater than 3.5 (Benchmark iROAS), the recommendation for Tactic T3 is to Scale Up Spend.

Implementation Process

Implementing causal attribution involves several key steps:

  1. Data Collection:
    Gather comprehensive data from all marketing channels, including spend, impressions, clicks, and conversions. This also includes broader business data like sales, pricing, and external factors (e.g., seasonality, competitor activities).

  2. Marketing Mix Modeling:
    Develop and run an MMM model to determine the incremental impact of marketing activities at a high level. This generates the iROAS values.

  3. Multi-Touch Attribution:
    Implement an MTA solution to capture granular, user-level interactions across marketing touchpoints.

  4. Data Integration:
    Combine the MMM and MTA data, ensuring alignment in metrics and time periods.

  5. Causal Model Development:
    Create a model that applies the causal attribution formula, integrating the iROAS from MMM with the detailed attribution data from MTA.

  6. Validation and Testing:
    Validate the model results against historical data and through controlled experiments where possible.

  7. Ongoing Calibration:
    Regularly update the model with new data and refine it based on changing market conditions and business objectives.


In-Depth Examples

Let's explore two detailed examples to illustrate the power of causal attribution:

Example 1: E-commerce Holiday Campaign

An online retailer runs a multi-channel holiday campaign including social media ads, search engine marketing, email marketing, and display advertising.

Traditional attribution might show:

ChannelConversionsRevenue
Social Media1000$100,000
Search800$80,000
Email500$50,000
Display200$20,000

However, causal attribution reveals:

Snapchat:
  iROAS: 1.2, pROAS: 2.0
  The Incrementality Factor for Snapchat is: 1.2/2.0 = 0.6
  Incremental Revenue = 0.6 * $100,000 = $60,000

Facebook:
  iROAS: 1.5, pROAS: 1.6
  The Incrementality Factor for Facebook is: 1.5/1.6 = 0.9375
  Incremental Revenue = 0.9375 * $80,000 = $75,000

Google Top of Funnel (BOF):
  iROAS: 3.0, pROAS: 1.0
  The Incrementality Factor for Google BOF is: 3.0/1.0 = 3.0
  Incremental Revenue = 3.0 * $50,000 = $150,000

Google Bottom of Funnel (TOF):
  iROAS: 0.5, pROAS: 2.0
  The Incrementality Factor for Google BOF is: 0.5/2.0 = 0.25
  Incremental Revenue = 0.25 * $20,000 = $5,000

This analysis shows that while Snapchat and Google BOF advertising appeared to perform well in platform reporting, it actually had a negative ROI when considering incrementality. Conversely, Google TOF, which seemed to perform moderately in direct attribution, actually drove the highest incremental revenue.

Example 2: Mobile App User Acquisition

A mobile gaming company is running user acquisition campaigns across various channels. They're particularly interested in understanding which channels drive not just installs, but long-term user value.

Traditional attribution might show:

ChannelInstalls30-day revenue
Network A10,000$50,000
Network B8,000$40,000
Network C5,000$25,000

Causal attribution, incorporating long-term user value data, reveals:

Network A:
  iROAS: 0.8, pROAS: 2.5
  The Incrementality Factor for Google BOF is: 0.8/2.5 = 0.24
  Incremental Revenue = 0.24 * $50,000 = $12,000

Network B:
  iROAS: 1.2, pROAS: 2.0
  The Incrementality Factor for Google BOF is: 1.2/2.0 = 0.6
  Incremental Revenue = 0.6 * $40,000 = $24,000

Network C:
  iROAS: 1.5, pROAS: 1.7
  The Incrementality Factor for Google BOF is: 1.5/1.7 = 0.882
  Incremental Revenue = 0.882 * $25,000 = $22,058

This analysis shows that while Network A appeared to drive the most installs and early revenue, it actually had the lowest incremental impact. Network B, which seemed to perform moderately in direct attribution, actually drove the highest incremental revenue when considering long-term user value.

Challenges and Limitations

While powerful, causal attribution is not without its challenges:

  1. Data Requirements: It requires extensive data from multiple sources, which can be challenging to collect and integrate.

  2. Model Complexity: Developing and maintaining accurate MMM (Marketing Mix Modeling) and MTA (Multi-Touch Attribution) models requires significant expertise and ongoing effort.

  3. External Factors: Accounting for all external factors that might influence outcomes (e.g., competitor actions, economic conditions) is difficult and can impact model accuracy.

  4. Model Drift: Over time, models may become less effective due to changes in spend, market conditions, or data quality, requiring updates and recalibration.

Use causal attribution when:

  • You need to understand the true incremental impact of your marketing efforts
  • You're dealing with complex, multi-channel marketing campaigns
  • You want to optimize budget allocation across channels
  • You need to account for both short-term and long-term marketing effects
  • You're looking to understand the interaction effects between different marketing channels

Conclusion

Causal attribution represents a significant advancement in marketing measurement, offering a more granular and actionable understanding of marketing performance. By combining the strategic insights of MMM with granular attribution data from MTA, marketers can make more informed decisions, optimize their marketing mix, and drive better business outcomes.

While implementing causal attribution requires significant effort and expertise, the insights it provides can be transformative for marketing organizations, enabling truly data-driven decision making and maximizing the impact of marketing investments.

Updated about 2 months ago

What’s Next