Key Driver Analysis

Overview

Before a user completes a key business conversion, they go through a complex sequence of behaviors and decisions. This sequence often contains several factor events that are highly correlated with the target conversion. Businesses need to determine how much each factor event contributed to the conversion. The Key Driver Analysis feature includes several common attribution models. You can choose a suitable model for your business to assign contribution to each factor event.

In Quick Tracking, the "target conversion event" is the conversion behavior mentioned above. It is usually a user behavior that achieves a business goal in an application. The related factor behaviors are "attributable events". These are events that may have led to the target conversion event. For example, in an E-commerce purchase scenario, a user might first search for a product. Then, they might click on a promotion for that product. Finally, they complete the purchase after receiving a promotional text message. The purchase is the "target conversion event". Searching for the product, clicking the ad, and clicking the promotional text message are the "attributable events" in this scenario.

Key driver analysis helps answer questions such as:

• In an E-commerce scenario, a user makes a purchase. How much did internal resources such as the Taobao home page search, event venues, and "Guess You Like" contribute to this sale?

• In an E-commerce scenario, how much did behaviors such as using spend-and-save promotion coupons or participating in promotional activities contribute to the total order amount?

• In a real estate application, a user leaves their phone number as a lead. How much did behaviors such as viewing a property, inquiring about a price, subscribing to price updates, or receiving a property recommendation contribute to the user leaving their phone number?

• A user subscribes to a video streaming membership. How much did behaviors such as previewing paid content, receiving a membership discount, adding an upcoming paid title to their favorites, viewing the membership benefits page, or receiving a membership expiration reminder contribute to the final subscription?

Function Introduction

Page layout

The Key Driver Analysis model page has three main parts: the Configuration area, the Analysis Results area, and the Query History list.

• In the Configuration area, you can specify the analysis entity, target conversion event, and attributable events. You can also set linking events, forward-linking properties, filter conditions, attribution models, and the attribution window.

• The Analysis Results area displays the query results in a table.

• The Query History list lets you save, update, view, and export data from historical reports.

Operations

Select an analysis entity (Required)

You can select an analysis entity: Device ID, Account ID, or Entity ID. Calculations are based on the selected entity.

Select a target conversion event (Required)

• Select a target conversion event

All event types, including virtual events, can be target conversion events. They are typically events that generate revenue.

• Select a contribution metric for the target conversion event

You can calculate contribution based on the number of occurrences of the target conversion event, the sum of its numeric event properties, or the sum of its numeric user properties. You can set multiple metrics to query the contribution of each attributable event to multiple metrics of the target conversion event. This supports comparative analysis.

• Set a forward-linking event (Optional)

Setting a forward-linking event and its properties can improve the accuracy of the attribution model. When you set an event as a forward-linking event, a user must trigger it after an attributable event and before the target conversion event for the contribution to be counted. The forward-linking event acts as a rule to split the behavior flow. Only attributable events that occur before a forward-linking event are considered to have contributed to the target conversion event. You can set multiple forward-linking events. Triggering any one of them splits the flow.

Consider this example:

A. After splitting the sequence by the target event, the model determines a valid lookback window for each target event sequence. It only calculates attributable events within this window. In this case, attributable event A_0 and forward-linking event F_4 are not included in the calculation.

B. After splitting by the forward-linking event, the model only calculates attributable events within each forward-linking event sequence. Attributable event A_4 is not counted because it has no preceding forward-linking event.

C. The forward-linking event for attributable events A_2 and A_3 is F_1. The forward-linking event for attributable event A_1 is F_2. Because no attributable event precedes forward-linking event F_3, it is considered a direct conversion without any attributable events. If you use the first-touch attribution model, this conversion is not attributed to any attributable event. It is counted as a conversion contributed by "Other" behaviors.

Setting a forward-linking event is useful in scenarios such as the following:

Consider the purchase behavior in an E-commerce scenario. Submitting an order is the target conversion event, while clicking a product promotion and searching for a product are attributable events. Regardless of the attributable event, the user must view the product page before converting. If you set "view product page" as the forward-linking event, only attributable events followed by a "view product page" event are considered to have contributed to the conversion. If a user triggers some attributable events but then makes a purchase directly without viewing the product page, such as during a flash sale in a live channel, the attributable events in this phase are not considered to have contributed to the target event and are not included in the calculation.

• Set associated properties (Optional)

Similar to setting forward-linking events, setting associated properties improves the accuracy of the attribution model. By setting associated properties, you can require that a property value of the target conversion event matches a property value of the forward-linking event for the contribution to be counted.

For example, in the E-commerce scenario previously described, you can use associated properties to add a restriction. In addition to the forward-linking event logic, you can require that the product ID property value of the target conversion event (submit order) matches that of the forward-linking event (view product page). This ensures that the user viewed the page for the same product they ordered. If a user views the page for product A but orders product B, the attributable events before the linking event are not considered to have contributed to the target event. These attributable events are not included in the calculation.

Associated properties only check if the property values are identical. To handle cases where a property has the same meaning but different names for different events, you can associate different properties. For example:

In a "view product page" event, the unique ID of a product might be reported with the property "product_ID". In a "submit order" event, the same unique ID might be reported with the property "submitted_order_product_code". Both correspond to the unique product ID "AA123456". In this case, you can associate the "product_ID" event property of the forward-linking event "view product page" with the "submitted_order_product_code" event property of the target conversion event "submit order". The forward-linking event is successfully associated when these property values are equal.

Select attributable events (Required)

• Set attributable events

Attributable events cannot be the same as the target conversion event. A virtual event that includes the target conversion event is also considered the same and is not supported. Otherwise, all event types can be set as attributable events.

• Set filter conditions and an alias for an attributable event (Optional)

Key Driver Analysis uses the same logic as Event Analysis to set filter conditions and add aliases.

• Set associated properties (Optional)

You can associate properties between an attributable event and the target conversion event. This means you specify that a property value of the attributable event must match a property value of the target event for the attributable event to contribute to the conversion. For example:

You can require that the order ID from a "submit order" event must be associated with the order ID from a "use coupon" event. Only then is the use of the coupon considered a real contribution to the order submission.

• Set groups, global filters, and user segments (Optional)

The logic for setting groups, global filters, and user segments in Key Driver Analysis is the same as in Event Analysis.

• Select an attribution model

This service supports common attribution models, such as First-touch, Last-touch, Linear, Position-based, and Time decay attribution. You can choose the analytical model that best suits your business needs.

A. First-touch attribution

When multiple attributable events contribute to the same target conversion event, the first valid attributable event receives 100% of the credit.

First-touch attribution is suitable for promoting new brands or products, or when a business is focused on acquiring new users and expanding its market. In these cases, the business wants to attract more users into its customer pool. First-touch attribution can show which channels are most effective for user acquisition and what initial event drives user awareness for the target conversion.

Note that first-touch attribution is less suitable for businesses with long conversion cycles. Because of period and window constraints, it may not capture the true first behavior.

B. Last-touch attribution

When multiple attributable events contribute to the same target conversion event, the last valid attributable event receives 100% of the credit.

Last-touch attribution data is reliable, making it suitable for business scenarios such as running short-term campaigns to quickly improve performance, or for companies focused on understanding which behaviors best drive final sales conversions.

C. Linear attribution

When multiple attributable events contribute to the same target conversion event, each valid attributable event receives equal credit.

Linear attribution distributes credit evenly across all stages of the customer journey. It is suitable for companies that want to maintain contact with customers throughout the sales cycle and sustain brand awareness. It assumes all attributable events play an equal role before conversion.

The linear model can dilute the impact of high-value channels. For example, in an E-commerce scenario, a user first visits a product page through a friend's share. They then visit the product page three more times from their favorites before making a purchase. Linear attribution would assign 25% of the credit to the friend's share and 75% to the favorites. The weight of the friend's share is too low, and the weight of the favorites is too high.

D. Position-based attribution: When multiple attributable events contribute to the same target conversion event, the first and last attributable events each receive 40% of the credit. The remaining attributable events share the remaining 20% of the credit.

Position-based attribution is also known as the U-shaped attribution model. It is a multi-touch attribution model suitable for companies that place high value on both lead generation and closing sales. This model de-emphasizes the marketing touchpoints in the middle of the journey and is suitable for companies that focus on the lead generation phase.

E. Time-decay attribution

When multiple attributable events contribute to the same target conversion event, the attributable events that occur closer in time to the target conversion event receive more credit.

6. Set the attribution window

The Key Driver Analysis model has two time-related settings: the time range and the attribution window.

• Time range: Similar to Event Analysis, the time range is the period during which the target conversion event occurs.

• Attribution window: Key Driver Analysis looks back for attributable events from the client time of the target conversion event. Only attributable events that occur within this window for a target conversion event are considered to have contributed to that conversion.

Consider the following case:

Without any linking events or associated properties, the target event G_1 occurs within the set time range. The attribution window is set to 1 day. The model will look back 24 hours from the time of the target event. All attributable events that occur after the lookback start time and before the target conversion event are included in the calculation. In the diagram, attributable events A_2, A_3, and A_4 are included in the model calculation. A_1 is outside the lookback period and is not included.

You can set the attribution window to "Today" or a custom period.

• Today: Looks back for attributable events from the time of the target conversion event to 00:00 on the same day.

• Custom: You can select days, hours, or minutes to look back for n days, n hours, or n minutes before the target event occurred.

View results

After you configure the model, click Start Analysis to run the query. The results of the key driver analysis are displayed in a data table.

The chart shows the following metrics for each attributable event: Attributable event clicks - Total clicks, Attributable event clicks - Valid conversion clicks, Valid conversion CTR (Click-Through Rate), Target conversion event - Trigger count, and Target conversion event contribution.

• Attributable event clicks - Total clicks: The number of times this attributable event occurred within the time range and attribution window. This is related to the target event but not the attribution model.

Example: If the analysis time range is from the 2nd to the 8th and the attribution window is "Today", the total clicks are the total number of times the attributable event occurred from the 2nd to the 8th. If the attribution window is 1 day, the total clicks are the total number of times the attributable event occurred from the 1st to the 8th.

• Attributable event clicks - Valid conversion clicks: The total number of valid attributable events that occurred before each target conversion event within the time range and attribution window.

Example: If the analysis time range is from the 2nd to the 8th and the attribution window is "Today", the valid conversion clicks are the number of attributable event clicks that occurred on the same day as each target conversion event and meet the forward-linking event or property conditions.

• Effective conversion click-through rate is the ratio of effective conversion clicks to total conversions.

• Target conversion event - Trigger count: The number of times a conversion can be successfully attributed to this attributable event based on the selected attribution model. For example, the number of first-touch conversions.

• Target conversion event contribution: The total trigger count for this target conversion event / The sum of total trigger counts for all target conversion events.

• Other: Unattributed events. For example, you want to analyze the attribution for event A and select B, C, D, and E as attributable events. If event F also occurred but was not selected, then F is an "Other" attribution event.

Calculation example

The analysis time range is from the 2nd to the 8th. The attribution window is "Today". The user behavior sequence is: (1st) A_3, (2nd) A_1, A_3, (3rd) A_1, (8th) A_1, G_1, A_1, A_3, A_1, A_2, G_2, A_2, (9th) A_3, G_3. The model is Position-based attribution.

• A_1 - Total clicks = 5. A_2 - Total clicks = 2. A_3 - Total clicks = 3.

• A_1 - Valid conversion clicks = 4. A_2 - Valid conversion clicks = 1. A_3 - Valid conversion clicks = 1.

• A_1 - Valid conversion CTR = 80%. A_2 - Valid conversion CTR = 50%. A_3 - Valid conversion CTR = 33.3%.

For G_1, the attribution is G_1 -> {A_1(100%)}. For G_2, the attribution is G_2 -> {A_1(40%), A_1(6.6%), A_3(6.6%), A_1(6.6%), A_2(40%)}. The A_2 event that occurred after G_2 on the 8th cannot be linked to any target conversion event. The A_3 and G_3 events on the 9th are outside the time range and are not calculated.

• Target conversion event trigger count for A_1 = 1 × 100 × (1.0 + 0.4 + 0.067 + 0.067) / 100 = 1.534. Target conversion event trigger count for A_2 = 1 × 100 × (0.4) / 100 = 0.4. Target conversion event trigger count for A_3 = 1 × 100 × (0.067) / 100 = 0.067.

• A_1 contribution = 1.534 / 2 = 76.7%. A_2 contribution = 0.4 / 2 = 20%. A_3 contribution = 0.067 / 2 = 3.35%