AMC Model Gallery

Author: Robin Jia，Tommy Lin

Last Update：2025/01/20

1. Function Introduction

Xnurta's AMC Model Gallery provides a series of standardized analytical models based on AMC data. Each model focuses on a specific marketing analysis scenario and presents insights through visualized reports, allowing users to flexibly select based on different business objectives.

1.1 Menu

The model gallery page mainly consists of the following parts:

• Model selection area at the top: Displays all analytical models in the form of cards. Clicking on a model card will enter the corresponding analysis report page. This part of the data is predefined.

• Custom model filtering area at the bottom::
  • Displays custom analytical models created by users (generated using user-defined parameters based on standard model templates)
  • Provides multi-dimensional filtering options, such as model name, creator, template type, update status, etc., to help quickly retrieve target models
  • Click the "Edit" button or model name in the model list to enter the details page of the custom model
  • Click the "Cancel" button to cancel a queued query
  • Click the "Delete" button to delete custom models that are no longer needed

• Area in the upper right corner:
  • Question bank: Lists common AMC analysis question scenarios. Clicking on a question card quickly enters the corresponding analytical model, guiding users to conduct analysis
  • Instance name: Displays the currently selected AMC instance. Clicking on the instance name will enter the instance management page; clicking on the drop-down arrow on the right side will quickly switch to other instances, and the model gallery will refresh to the predefined and custom model data under the new instance.

1.2 Model Data

Model data is divided into predefined data and custom data.

Predefined data means that Xnurta regularly extracts model data for the previous month based on system-preset conditions every month. According to the different attribution methods of each model, the extraction time of each model varies and is divided into the 8th and 15th of each month.

Unique Reach, Audience Label, Cross-Product Association, Time to Conversion, Search Term Analysis: Data from the previous month is queried on the 8th of each month, and the data query duration may take up to 24 hours.

Path To Conversion, Overlap Analysis, Geographic Analysis, Multi-Touch Attribution, Customer Lifetime Value: Data from the previous month is queried on the 15th of each month, and the data query duration may take up to 24 hours.

Custom data is model data that users extract using the custom model conditions function in the model library. Each model provides different customizable conditions (see the specific model sections for details), and users can flexibly select and extract customized data reports.

Each model's custom models are divided into three extraction methods: preset, instant, and loop.

Preset models are models that users save when using predefined models. Instant models are models that users extract once instantly after defining model conditions. Recurring models are models that users set to extract at a fixed frequency within a certain time range.

Regardless of the method chosen, AMC data only stores data for the past year, so the maximum time range for each custom model's data does not exceed 365 days.

Based on the user-defined time range, data scope, and other conditions, there will be a certain waiting time after a custom model is extracted before it can be completed. When extracting multiple models at the same time, they will be queued in the order of creation time, and the model in the first position will enter the querying state. This time does not exceed 24 hours.

Based on this feature, custom models will have multiple statuses after creation.

1.3 Model Details Page

Clicking on a predefined model card or a custom model in the custom list will enter the details page of the corresponding model.

Each model consists of the following parts:

1. Model Introduction: Provides an overview of the analysis scenarios applicable to the model, data sources, and analysis logic.
2. Filtering Parameter Settings: Based on the standardized model, users can further adjust analysis parameters to filter out specific data views of interest. Filtering parameters typically include date range, display metrics, etc.
3. Analysis Report Presentation: Presents the analysis insights generated by the model in the form of visualized charts. Each model contains multiple analysis perspectives from different dimensions.
4. Custom Model Creation: For some analytical models that support customization, users can click "Compose Model Query" to enter the custom parameter configuration page.

The following is a simple introduction using the Geographic Analysis model as an example:

In the model menu at the top, each model includes the display of data scope and time range, and has different filtering options for adjustment.

The chart data at the bottom can be interacted with and clicked using the mouse, depending on the specific model template.

Some models include an edit button. Clicking it will open a sidebar for more detailed data selection and filtering.

After filtering the model data, you can click the button in the upper right corner to save it as a custom model, which will save the model to the custom model list. When entering again, it will save the current filtering options.

Clicking the Compose Model Query button in the upper right corner of the model details page will enter the page for creating the custom model.

In the drawer, each model needs to define the name of the custom model, the time range of the data, and the extraction method of the model (see the Model Data section).

Specific models may have special definable conditions. Please refer to the introduction section of each model in this document for this part.

After clicking the Save as New Model button, it will return to the main menu. In the custom model list, you can find the newly created model. At this point, the model still needs to wait for data extraction, but you can enter the model page at any time to view the previously defined model conditions.

The conditions of created models cannot be modified, but you can click to enter the details page, modify, and save them as another custom model.

Once the model status becomes Completed, you can click on the model name to enter the page and view the data.

 

2. Model Navigation

AMC data models analyze different variables that affect advertising performance, such as users, products, time, location, and ad types. Each model may cover multiple dimensions, but also has a focused theme. The following is a classification based on the different purposes of the models.

Advertising Effectiveness Verification	Through the following models, you can verify the delivery effects of various ads and provide multi-dimensional data references for marketing strategies.
		Unique Reach	Analyzes the number of unique audiences covered by ad campaigns, helping you understand audience reach trends and related ad campaign spending trends. Supports defining custom models based on specified time range, ad campaign scope, time interval, and other conditions.
		Overlap Analysis	From a quantitative perspective, analyzes the coverage and overlap of audiences reached by multiple ad types or different delivery strategies under the same ad, as well as the corresponding conversion rates, to identify better ad type combinations or strategy groups and guide the next step of delivery. Supports defining custom models based on specified time range, ad campaign scope, product scope, etc.
		Path To Conversion	Analyzes the sequence from the first touchpoint to the last touchpoint before consumers make a purchase, deeply understands the path that audiences are influenced by ads before making a purchase, and improves ad effectiveness through multiple touchpoints. Supports defining custom models based on specified time range, touchpoint ad campaigns, purchased products, etc.
		Multi-Touch Attribution	Utilizes multiple attribution models to re-attribute conversion goals brought by different ads, re-evaluates the role of each ad touchpoint in conversion, and optimizes budget allocation based on this.
		Customer Lifetime Value	The principle of the customer lifetime value analysis model is to evaluate the total value brought by a customer over the entire period of their contact with the brand/product. By referring to this data, brands hope that the value provided by customers (CLV) is far higher than the cost of acquiring a customer (CAC).
Brand-Level Insights	Through the following models, you can gain insights into brand-dimensional data, including various aspects of users, products, and brands.
		Audience Labels	Through dimensions such as number of people covered, number of purchasers, DPV, DPVR, and purchase conversion efficiency, judge the performance of each audience label in DSP delivery, thereby optimizing label combinations and discovering more potential labels to cover more valuable audiences. Supports defining custom models based on specified time range, ad campaign scope, etc.
		Cross-Product Association	Understand which products are more likely to attract consumers to purchase other products in the store, helping sellers discover potential products to assist in bundled promotion campaign design. Supports defining custom models based on specified time range, product scope, etc.
		Geographic Analysis	Analyzes ad performance by different countries/regions. Helps discover differences in ad performance among users in different regions, thereby more accurately defining the geographic scope of ad delivery and adopting different ad strategies.
		Customer Lifetime Value	The principle of the customer lifetime value analysis model is to evaluate the total value brought by a customer over the entire period of their contact with the brand/product. By referring to this data, brands hope that the value provided by customers (CLV) is far higher than the cost of acquiring a customer (CAC).
Campaign Optimization Guidance	Through the following models, you can provide optimization suggestions for advertising strategies, including clear actionable optimization directions for ad accounts.
		Audience Labels	Through dimensions such as number of people covered, number of purchasers, DPV, DPVR, and purchase conversion efficiency, judge the performance of each audience label in DSP delivery, thereby optimizing label combinations and discovering more potential labels to cover more valuable audiences. Supports defining custom models based on specified time range, ad campaign scope, etc.
		Time to Conversion	Analyzes the main time range in which audiences complete the purchase of products from behaviors such as exposure to ads, browsing detail pages, adding to cart, searching, etc., applied to more reasonably planning and adjusting media plans and advertising budgets in the future. Supports defining custom models based on specified time range, ad campaign scope, product scope, keywords, etc.
		Cross-Product Association	Understand which products are more likely to attract consumers to purchase other products in the store, helping sellers discover potential products to assist in bundled promotion campaign design. Supports defining custom models based on specified time range, product scope, etc.
		Geographic Analysis	Analyzes ad performance by different countries/regions. Helps discover differences in ad performance among users in different regions, thereby more accurately defining the geographic scope of ad delivery and adopting different ad strategies.
		Multi-Touch Attribution	Utilizes multiple attribution models to re-attribute conversion goals brought by different ads, re-evaluates the role of each ad touchpoint in conversion, and optimizes budget allocation based on this.

3. Data Privacy and Filtering

Amazon Marketing Cloud (AMC) implements data privacy safeguards to protect customer information while enabling advertisers to perform meaningful analytics. These safeguards automatically filter data that could potentially expose individual user information, ensuring both data utility and user privacy.

AMC uses aggregation thresholds to determine when data should be filtered. Different types of data have different threshold requirements based on their sensitivity level. For example, metrics like impressions have no restrictions, while user-specific data requires higher levels of aggregation. When a query returns data, rows that don't meet the minimum user count requirements will have sensitive fields replaced with NULL values.

Below is the data filtering requirements table for each analytics model in Xnurta: