Attribute credit to multiple acquisition touch points

Attributing the success of marketing activities is difficult when you can't pinpoint which activities led users to a desired outcome. For example, a user might visit your website after seeing a Google ad, interacting with a Facebook post, and watching a TikTok video. You can attribute credit to one or more of these activities in several ways. Attributing success to property values, known as multi-touch attribution, provides context that informs future marketing plans.

Restrictions

• Users on the Starter and Plus plans can create a single channel view.

Pre-built attribution models

Amplitude includes common, pre-built attribution models that you can configure on your metric.

When measuring unique users, none of these models generate attribution group totals that sum to the count in the Overall row. Each unique user can appear in multiple attribution groups because attribution applies to events, not unique users. A unique user can perform events attributed to both channel X and channel Y. In this case, the user appears in rows for both X and Y, which can double-count the user when summing the rows.

• First Touch: Gives all credit for the selected metric to the first property value within the selected lookback window relative to the date the metric occurred. With the event totals attribute, this sums to 100%.
• Last Touch: Gives all credit for the selected metric to the last property value within the selected lookback window relative to the date the metric occurred. With the event totals attribute, this sums to 100%.
• Linear: Distributes credit for the selected metric equally across all property values within the selected lookback window relative to the date the metric occurred. For example, with two properties each receives 50% credit, and with three properties each receives 33.3%.
• Participation: Allocates credit for the selected metric fully to all property values within the selected lookback window relative to the date the metric occurred. For example, with two properties each receives 100% credit, and with three properties each receives 100%. The total rows can exceed the overall total because multiple properties from the same event can receive credit.
• U-Shaped: Biases credit to the first and last values for the selected property. With two touch points, the middle 20% is added equally to the first and middle touch points (50%, 50%). With four touch points, the middle two touch points share the 20% (40%, 10%, 10%, 40%).
• J-Shaped: Distributes credit for the selected metrics in a way that biases credit to the more recent values for the selected property. With two touch points, the first 20% is added equally to the last and middle touch points (30%, 70%). With four touch points, the final two touch points share the 20% (10%, 10%, 20%, 60%).
• Inverse J-Shaped: Distributes credit for the selected metrics in a way that biases credit to the first values for the selected property. With two touch points, the last 20% is added equally to the first and middle touch points (70%, 30%). With four touch points, the last two touch points share the 20% (60%, 20%, 10%, 10%).
• Data Driven: With this model, Amplitude Analytics relies on a probabilistic algorithm based on first-order Markov chains. Every customer journey—a sequence of channels or touch points—is represented as a chain in a directed Markov graph. Each node is a possible state (a channel or a touch point), and the edges represent the probability of transition between states. Amplitude Analytics removes the nodes one by one and estimates the impact of removing each node on the conversion rate. Each channel receives credit in proportion to its removal effect. Use this model with properties that don't have a large number of unique values (50 or fewer work best). Learn more about the algorithm at analyzecore.com.

Configure an attribution model

In a data table, configure an attribution model on each metric column:

1. On the column, click …, then click Attribution.
2. Select an attribution model and configure a lookback window. Optionally, apply the attribution model to all columns in the table.
3. Click Apply to confirm the change and review the table results with the attribution model applied.

Create a custom attribution model

If the pre-built attribution models don't meet your needs, create a custom model. You must be an Admin or Manager to create a custom attribution model. Follow these steps:

1. On the column, click …, then options, then click Attribution.
2. Select Custom from the model dropdown to display options for configuring your custom model.
3. Set a name and description for the model so others know how to interpret it.
4. Choose a custom weighting for your model.
   • The first weight applies to the first touch.
   • The last weight applies to the last touch.
   • The middle weight distributes evenly across all touches between. If there are no touches between, the first and last touch each receive half of the middle weight.

5. Set the default lookback window for the model. Optionally, lock the window to ensure that others using this model can only use that lookback window.
6. Decide whether to share the custom model with others in your organization.
7. Optionally, exclude property values from attribution. This is useful when you don't want to assign credit to a particular value, such as direct website visits or email.
8. Click Save to confirm the change, save the model for yourself or others to use later, and review the table results with the attribution model applied.

Use cases

• Acquisition channel credit: When you analyze the effectiveness of organic and paid investments, use acquisition channels with a multi-touch attribution model to understand how each channel contributes to driving KPI outcomes. Depending on your business model and user behavior, analyze which attribution model fits best, and make investment decisions based on each channel's contribution to your target metric.
• Comparing attribution models: In longer conversion cycles with multi-session user flows, compare the same metric with different attribution models applied. This data helps you discover which attribution model reflects the most efficient marketing investment, and which stage of the customer buying cycle a campaign affects. For example, when attributing to advertising campaigns, find which campaigns tend to be the first interaction (awareness) that a customer has, the last (high intent), or somewhere between (research).
• Content: Use attribution to find how often users viewed content and how that content participated in driving a business KPI outcome. Knowing that content has a low bounce or exit rate or longer time on page is helpful, but you can clarify the business impact by generating a conversion rate based on different attribution models.
• Internal campaigns: Like paid off-platform advertising investments, marketing teams invest time and creative talent to generate offers and brand-building content that drives KPI outcomes. Attribution on the impact of those marketing efforts informs your content marketing teams which types of offers and creatives drive the most short- and long-term business value.
• Paid channels with LTV: Combine your attribution model with your behavior-based LTV calculations to see a fuller perspective of how much value a paid channel or campaign drives. This data unlocks potential for greater investments in channels that drive the most long-term business value.

Supported attribution types by metric

Each metric type supports a specific set of attribution types:

• Uniques

  • first touch
  • last touch
  • participation
  • markov

• Conversion

  • first touch
  • last touch
  • participation

• Event totals

  • first touch
  • last touch
  • participation
  • linear
  • j-shaped
  • inverse j-shaped
  • u-shaped
  • custom
  • markov

• property sum, revenue total, and formula (clauses: uniques, total, propsum)

  • first touch
  • last touch
  • participation
  • linear
  • j-shaped
  • inverse j-shaped
  • u-shaped
  • custom

  Attribution options differ between uniques and event total attribution types because a unique user is a less clear unit to split across multiple channels or campaigns. Only attributions that clearly assign a whole user to a single channel or to many channels are used, such as First, Last, or Participation.

Attribution with multiple properties

The attribution model applies only to the property in the outermost group-by level. Properties in inner group-by levels don't undergo attribution modeling. Instead, they inherit their values from the attributed event.

In this example, the attribution model applies only to Channel, the outermost group-by property.

utm_source, as an inner group-by property, derives its value from the attributed events rather than applying attribution separately.

For more information about how Amplitude organizes results, or why a sum may not match the overall value, go to Group-bys: How Amplitude prunes and orders chart results.

Attribution example calculation

This example highlights the differences between attribution models and lookback windows.

A user has three touch points before the Sign Up event, each with a different UTM source:

The following sections show example combinations of attribution model and lookback window, and the resulting credit attributed to each UTM source.

• First Touch
  • Lookback Window: 30 Days
  • Credit: Google: 100%
  • Explanation: All credit goes to the first touch within the last 30 days, which is Google on 2022-05-01.
• First Touch
  • Lookback Window: 7 Days
  • Credit: Facebook: 100%
  • Explanation: All credit goes to the first touch within the last 7 days, which is Facebook on 2022-05-07.
• Last Touch
  • Lookback Window: 7 Days
  • Credit: TikTok: 100%
  • Explanation: All credit goes to the last touch within the last 7 days, which is TikTok on 2022-05-10.
• Linear
  • Lookback Window: 30 Days
  • Credit:
    • Google: 33%
    • Facebook: 33%
    • TikTok: 33%
  • Explanation: Divides evenly between all three touch points in the last 30 days.
• Linear
  • Lookback Window: 7 Days
  • Credit:
    • Facebook: 50%
    • TikTok: 50%
  • Explanation: Divides evenly between the two touch points in the last 7 days.
• J-Shaped
  • Lookback Window: 30 Days
  • Credit:
    • Google: 20%
    • Facebook: 20%
    • TikTok: 60%
  • Explanation: In the last 30 days, the first touch gets 20%, middle touches 20%, and last touch 60%.
• J-Shaped
  • Lookback Window: 7 Days
  • Credit:
    • Facebook: 30%
    • TikTok: 70%
  • Explanation: There is no middle touch, so the 20% is split across the first and last touches.
• Custom; 5% - 20% - 75%
  • Lookback Window: 30 Days
  • Credit:
    • Google: 5%
    • Facebook: 20%
    • TikTok: 75%
  • Explanation: In the last 30 days, the first touch gets 5%, middle touches 20%, and last touch 75%.