Multi-touch Attribution

Author: Robin Jia，Tommy Lin

Last Update：2024/06/14

 

1. Model Introduction

The Multi-touch Attribution Model Analysis focuses on exploring the differences in contribution of different ad touchpoints in the user conversion path. By adopting different attribution logic (such as first touch, last touch, linear weighting, etc.) to allocate conversion contributions, this model can help comprehensively evaluate the role of various ad touchpoints in promoting conversions, find the key touchpoint combinations that have the most critical impact on conversions, and further optimize touchpoint strategies and budget allocation to achieve overall marketing efficiency improvement. This model has built-in multiple mainstream attribution models and supports flexible adjustment of various key parameters, which can meet the analysis needs of different attribution analysis scenarios.

Through this model, the following business questions can be analyzed:

• In addition to Amazon's basic attribution model, what other attribution models can help understand the impact of ads on conversions?
• How to optimize your ad budget through multi-touch attribution (MTA) models?
• What are the expected results of optimizing the ad budget using other models?

 

2. Model Interpretation

This model mainly consists of three parts. The first part is the model selection menu, where up to 7 different attribution models can be selected. Based on the selection result of the model, the second part shows the attribution results of each ad event according to the selected attribution model and compares them with Amazon's default attribution. The third part goes further and provides budget optimization suggestions and result predictions based on the attribution results from the previous step.

In the model's top menu, the predefined model is for fixed Amazon users, all Amazon ad types, and all ASINs under the store. You can select a time range that includes data for each month. In addition, it allows switching the conversion goal attributed by the model, purchase or new customer purchase.

In the predefined model, the default attribution for each ad type varies based on the characteristics of the ad type:

• Amazon DSP attributes based on impressions in the past 14 days.

• Sponsored Products, Sponsored Brands, and Sponsored Display attribute based on clicks in the past 7 days.

Here you can customize according to your needs in the custom model. See customization for details.

In the attribution model selection module, users can select an attribution model from the preset mainstream attribution model library. Each model represents a specific attribution logic. Different attribution logics assign different weights to each touchpoint in the conversion path, representing different value judgment perspectives. The coexistence of multiple models helps to evaluate the attribution problem more objectively and comprehensively.

MTA provides 7 attribution models to choose from, as follows, among which the two model-based models are more recommended.

Attribution Model	Type	Description
Markov Chain	Model-Based	The Markov Chain model, through multiple advertising touchpoint paths, calculates the probability of transitioning from one touchpoint to another, and determines the contribution of a touchpoint to conversion by assessing the impact on conversion rate when that touchpoint is removed.
Shapley Value		Multi-touch attribution using the Shapley Value is a cooperative game theory approach that fairly allocates credit to each marketing touchpoint in a customer's journey based on its marginal contribution to the final conversion.
Last Touch	Rule-Based	Last Touch attribution assigns all conversion credit to the last ad the user interacted with.
First Touch		First Touch attribution assigns all conversion credit to the first ad the user interacted with.
Linear		The Linear attribution model equally distributes credit to each ad in the conversion path.
Time Decay		The Time Decay model assigns more credit to ad interactions that are closer in time to the conversion, reflecting the timeliness of the interactions.
Position-Based		The Position-Based attribution model places special emphasis on the first and last ad interactions, while also considering those in between.

Based on the selected attribution model, the attribution result overview module will intuitively present the attribution contribution analysis results of various ad touchpoints. Through the bar chart, users can see at a glance the contribution proportion of each type of touchpoint under the current attribution logic and its difference from other touchpoints. In the bar chart, the bar on the left represents Amazon's official attribution result, and the bar on the right represents the attribution result of the selected attribution model. By comparing the results, the potential value of ad events that perform ordinarily in Amazon's ad reports can be discovered. Clicking the download button can download the chart data spreadsheet.

For the Markov Chain attribution model, an additional interactive chart is provided. The chart represents all the conversion paths a user may experience before purchasing or not purchasing. Each circle represents an ad event touchpoint, and users passing through this touchpoint represent those who have been reached by this ad event.

By clicking on different ad touchpoints, the conversion paths related to this ad event will be removed. Through the window on the right, you can see the change in overall conversion rate after removing this ad event, thereby calculating the removal effect and attributing ad events.

After gaining insights into the value contribution of various types of touchpoints, users can further utilize the budget optimization module driven by attribution to view the optimized budget allocation results. Here, the historical spending proportion of the current Amazon ad account on each ad type (within the selected time range) will be compared, and the recommended budget proportion will be given.

Clicking on the second tab is the result prediction for the recommended budget proportion. Based on the current optimized budget allocation plan, the system will quantify and simulate the expected return improvement space after optimization. It includes the prediction of attribution results for each ad type and the prediction of total attribution results. If the bar is above the coordinate axis, it represents a predicted increase in attribution results; conversely, it represents a predicted decrease in attribution results. Since the final attribution and performance of ads are affected by various factors, the prediction results here are for reference only. Clicking the download button can download the chart data spreadsheet.

The table at the bottom summarizes all the data in the charts above and allows you to directly view the specific data.

 

3. Model Customization

The custom model conditions of this model provide the function of filtering by ad account and support Amazon Purchase Data Insights. The ad section below can customize ad groupings. By default, it counts separately by ad type (DSP, SP, SB, SD). Custom attribution events can also be selected, with a maximum of 10 different groupings defined.

Click the edit icon to select the ad campaigns included in the ad grouping. Note that a single ad grouping can only include ad campaigns under the same ad type.

After defining the ad groupings, you can set the attribution events in the attribution module below, including the event type (impression/click) and lookback period

• The attribution event determines the criteria for the model to judge whether an ad grouping has an impact on conversion. When impression is selected, as long as the ad in the grouping has an impression before the conversion occurs, it is considered to have an impact on the conversion and the attribution is calculated. When click is selected, the ad in the grouping needs to have a click before the conversion occurs to be considered to have an impact on the conversion. Impressions without clicks will not be calculated for attribution.
• The lookback period determines the longest effective impact time of the selected event type before the conversion occurs. For example, if impression is selected as the attribution event and 14 days is selected as the lookback period, then ad grouping impression events within 14 days before a user makes a purchase will be calculated for attribution, while ad events beyond 14 days before the purchase are considered to have no impact on the purchase and will not be calculated for attribution.
• When selecting consistent, all ad events will adopt the same attribution method.
• When selecting custom, you can set different attribution methods for each ad event

Users can choose the interaction type and lookback period based on the nature of the ad event. Ad events that are more biased towards upper-funnel traffic are more suitable for impression as the interaction type and a longer lookback period, while ad events that are more biased towards lower-funnel conversion are more suitable for click as the interaction type and a shorter lookback period.

In addition, this model also supports customizing the monitored ASINs. After setting, only the conversion attribution results of the set ASINs will be counted.

 

4. Model Data

The Multi-touch Attribution Analysis model comprehensively utilizes multi-source data such as ad reach and user conversion. By linking and analyzing the time sequence characteristics of users' ad exposure and conversion behaviors, it quantifies the contribution of different ad exposures to the final conversion. To ensure the reliability of attribution analysis results, the following points need to be focused on when interpreting attribution data:

• Setting of the attribution time window. The setting of the attribution time window will significantly affect the attribution analysis results. If the window is set too short, some key touchpoints in the earlier period may be missed; if the window is set too long, too much noise data may be mixed in. Therefore, it is necessary to carefully evaluate the reasonable range of the attribution time window based on the product's purchase decision cycle, historical data patterns, etc.
• Handling of impressions/clicks. When performing attribution analysis, it is necessary to consider how to calculate the way users are reached by a certain type of ad. Generally, you can choose to maintain a consistent impression or click reach method, or you can choose to set different reach interaction methods for different ads according to the characteristics of different ad types.
• Omission of offline touchpoints. The current attribution analysis methods mainly capture online ad touchpoints that can be directly observed, while brand exposures in offline scenarios (such as a user seeing an offline ad of a brand and then returning online to search and purchase) are often difficult to incorporate into the attribution system. Therefore, when interpreting attribution results, it is necessary to recognize their limitations and carefully evaluate the potential impact of offline factors.
• Impact of external factor fluctuations. There are often many uncontrollable factors in the marketing environment, such as competitor activities, seasonal holidays, etc., which may disturb the attribution analysis results. Therefore, when comparing and interpreting attribution results from different periods, it is necessary to examine whether these external factors have changed significantly and appropriately control the marketing environment.
• Evaluation and optimization of attribution model application. The purpose of attribution analysis is to optimize marketing decisions and improve ROAS. The budget optimization data is provided based on the attribution logic and model set by the user. Therefore, after attribution analysis, it is also necessary to test the actual effect improvement brought by real ad evaluation attribution budget optimization, and continuously adjust the attribution logic and analysis parameters based on this to ensure that attribution analysis continues to generate value.

Glossary

Type	Term	Description
Dimension	Event	Impression or click that attributon model
	Ads Account	The account used for managing and executing advertising campaigns.
	Model	The algorithm or method used to attribute conversions or other desired outcomes to specific events or touchpoints.
	Goal	The primary objective or result that an advertising campaign aims to achieve.
	Attribution	The process of assigning credit to different marketing touchpoints for their role in achieving the campaign goals.
Metrics	Total Purchase	The total number of purchases made as a result of the advertising campaign.
	Total NTB Purchase	The total number of purchases made by new-to-brand (NTB) customers, indicating first-time buyers influenced by the campaign.
	Actual Cost	The total expenditure on the advertising campaign.
	Suggest Budget	The recommended budget allocation for future advertising efforts, based on past performance and future goals.
	Forecast Result Change	The predicted change in campaign performance metrics as a result of proposed adjustments or ongoing trends.