Bandeali oversees the technology, engineering, and data teams at Pixalate. He holds an MSE in Software Engineering & a BSc in Computer Science. He co-founded Pixalate in 2012.
Lazaris leads Pixalate's data science and research teams at Pixalate. He holds a PhD in Electrical Engineering and an MSc and BS in Electronic and Computer Engineering. He joined Pixalate in 2014.
Poovakottu develops data analysis framework models. He holds MS degrees in Information & Data Science and Engineering Management & Computer Management. He joined Pixalate in 2019.
The Mobile Publisher Trust Index provides ratings for various combinations of the following data points:
App Store Name: The App Store name characterizes where the publisher apps can be downloaded from. Currently only the “Apple App Store” and “Google Play Store” are supported.
Region Name: The name of the broader geographical region in which the end-user resides (i.e., this is essentially the location of the end-user generating impressions on a given publisher). The following regions are supported, and the countries that correspond to each region are defined in Appendix 1:
Category: This is the category of the publisher (e.g., News and Media, etc.). An “All” category is added to rank publishers across all the categories of a given app store. Categories are determined using the IAB taxonomy.
The main columns used to identify a publisher are the “appId'' and the “title”. AppID is unique for a given app store. For every appId, the following metrics are produced, characterizing its score compared to other publishers:
- Programmatic Reach Score: This score corresponds to a composite measure of programmatic popularity that is calculated in terms of both reach and inventory availability. It is a number between 0 to 99 indicating how extensive user coverage and inventory an app has compared to its competitors.
- IVT Score: This score corresponds to the percentage of IVT of the publisher compared to other publishers in a given category and GEO (i.e., other publishers in the same category).
- Ads.txt Score: This score corresponds to the detected presence of ads.txt for a given publisher, as well as the presence of a “balanced” mix of sellers and resellers compared to its competitors. For this, Pixalate has developed a methodology for building a scoring curve for ads.txt that assigns a score for each combination of total sellers and resellers for a given publisher. The main rationale is that too many resellers may increase the risk for IVT, and too few sellers or resellers may limit the availability of the app's inventory to the open programmatic marketplace (see, for example, Figure 1. grading curve). A balanced combination of the two may provide the best results.
- Brand Safety Score: This score captures brand unsafe keywords or brand unsafe advisories declared by a given app in their app store page.
- Permissions Score: This score corresponds to the increase in the likelihood of the presence of IVT when specific permissions are given to a given app that are not common for its category. Permissions that can be risky for a given category might be necessary for the operation of the app in another category, and this kind of information is taken into consideration in order to derive this score. Note: this score is currently only available for Android.
- Viewability Score: This score corresponds to the performance of the app in terms of its viewability.
- Rank: This corresponds to the rank order of the publisher compared to its competitors within a given app store, category, and region combination.
- Final Score: This score corresponds to the average of all the individual scores mentioned above.
Figure 1. An example of a bell grading curve. If “μ” corresponds to the best value for the x-axis (e.g., number of direct sellers), then As and Bs are going to be assigned to the publishers with values around μ+σ and μ-σ. On the other hand, if the best value on the x-axis is after 2σ (e.g. for the case of IVT), then the As will be assigned to the publishers in the right tail of the distribution.
- Programmatic Reach Score: This score corresponds to a composite measure of programmatic popularity that is calculated in terms of both reach and inventory availability. It is a number between 0 to 99 indicating how extensive user coverage and inventory an app has compared to its competitors.
- IVT Score: This score corresponds to the percentage of IVT of the publisher compared to other publishers in a given category and GEO (i.e., other publishers in the same category).
- Ads.txt Score: This score corresponds to the detected presence of ads.txt for a given publisher, as well as the presence of a “balanced” mix of sellers and resellers compared to its competitors. For this, Pixalate has developed a methodology for building a scoring curve for ads.txt that assigns a score for each combination of total sellers and resellers for a given publisher. The main rationale is that too many resellers may increase the risk for IVT, and too few sellers or resellers may limit the availability of the app's inventory to the open programmatic marketplace (see, for example, Figure 1. grading curve). A balanced combination of the two may provide the best results.
- Brand Safety Score: This score captures brand unsafe keywords or brand unsafe advisories declared by a given app in their app store page. Apps with fewer brand safety risks will achieve a higher score in this category.
- Viewability Score: This score corresponds to the performance of the app in terms of its viewability. Apps with higher viewability will achieve a higher score in this category.
- Rank: This rank corresponds to the rank order of the publisher compared to its competitors within a given Platform Name and Category combination.
- Rank: This corresponds to the rank order of the publisher compared to its competitors within a given app store, category, and region combination.
- Final Score: This score corresponds to the average of all the individual scores mentioned above.
Figure 1. An example of a bell grading curve. If “μ” corresponds to the best value for the x-axis (e.g., number of direct sellers), then As and Bs are going to be assigned to the publishers with values around μ+σ and μ-σ. On the other hand, if the best value on the x-axis is after 2σ (e.g. for the case of IVT), then the As will be assigned to the publishers in the right tail of the distribution.
The CTV Publisher Trust Indexes provides ratings for various combinations of the following data points:
Platform Name: the platform characterizing the mobile device (currently Roku and Fire TV are supported).
Region Name: the name of the broader geographical region that the end-user resides (i.e., essentially the location of the end-user generating impressions on a given publisher).
Primary Genre Name: This is the category of the publisher (e.g., News and Media, etc). An "All” category is added in order to rank publishers across all the categories of a given platform. Categories are defined using the app categories as provided by Roku and Amazon Fire TV, respectively.
The main columns used to identify a publisher are the “appId'' and the “title”. AppID is unique for a given platform. For every appId, the following metrics are produced, characterizing its score compared to other publishers:
Metrics Used on the Publisher Trust Index for Roku Channel Store apps- Programmatic Reach Score: This score corresponds to a composite measure of programmatic popularity that is calculated in terms of both reach and inventory availability. It is a number between 0 to 99 indicating how extensive user coverage and inventory an app has compared to its competitors. The score to grade assignment is shown in Table 1 below. Apps with more extensive user coverage compared to competitors will achieve a higher score in this category.
- IVT Score: This score corresponds to the percentage of IVT of the publisher compared to other publishers in a given category and GEO. Apps with less IVT will achieve a higher score in this category.
- Ad Density Score: This score characterizes the number of ads (i.e., ad density) per 10-minute video time. This can be seen as a tradeoff between maximization of revenue on the advertising ecosystem side while minimizing the user disturbance. Pixalate has developed a method for assigning scores to various ad density values. See Figure 2 in The Methodology of our Ratings & Metrics section for more.
- User Engagement Score: This score captures the user engagement with a given app in terms of time spent consuming content. Apps that have a higher average watch time will have a higher score in this category.
- Rank: This corresponds to the rank order of the publisher compared to its competitors within a given app store, category, and region combination.
- Final Score: This score corresponds to the average of all the individual scores mentioned above.
Metrics Used on the Publisher Trust Index for Amazon Fire TV apps- Programmatic Reach Score: This score corresponds to a composite measure of programmatic popularity that is calculated in terms of both reach and inventory availability. It is a number between 0 to 99 indicating how extensive user coverage and inventory an app has compared to its competitors. The score to grade assignment is shown in Table 1 below. Apps with more extensive user coverage compared to competitors will achieve a higher score in this category.
- IVT Score: This score corresponds to the percentage of IVT of the publisher compared to other publishers in a given category and GEO. Apps with less IVT will achieve a higher score in this category.
- Ad Density Score: This score characterizes the number of ads (i.e., ad density) per 10-minute video time. This can be seen as a tradeoff between maximization of revenue on the advertising ecosystem side while minimizing the user disturbance. Pixalate has developed a method for assigning scores to various ad density values. See Figure 2 in The Methodology of our Ratings & Metrics section for more.
- User Engagement Score: This score captures the user engagement with a given app in terms of time spent consuming content. Apps that have a higher average watch time will have a higher score in this category.
- Rank: This corresponds to the rank order of the publisher compared to its competitors within a given app store, category, and region combination.
- Final Score: This score corresponds to the average of all the individual scores mentioned above.
Table 2: Ad Density Scoring Table
|
#Ads/10min |
Normalized Score (1-4) |
|
< 1 |
2 |
|
1 |
3 |
|
2 |
3 |
|
3 |
4 |
|
4 |
4 |
|
5 |
4 |
|
6 |
3 |
|
7 |
2 |
|
8 |
1 |
|
>9 |
1 |
Figure 3.
The brand safety scoring metric of Publisher Trust Index indexes aims to capture apps with increased likelihood of adult, drug, violence, alcohol, hate speech, or gambling content. To do so, Pixalate relies on information from the two major app stores that characterize the app’s content advisory ratings, as well as text analysis on the description, title, or image captions. Examples of brand unsafe advisories include “Frequent/Intense Sexual Content or Nudity”, “Mature 17+”, “Adults only 18+”, etc.
IVT scoring corresponds to the percentage of invalid traffic (IVT) of the publisher compared to its competition in a given category and geography.
An app permission is a mechanism for controlling access to a system- or device-level function through the device’s operating system. Permissions can often have privacy implications (e.g., access to fine-grained location). App permissions help support user privacy by protecting access to the following:
- Restricted data, such as system state and a user's contact information.
- Restricted actions, such as connecting to a paired device and recording audio.
Monthly Active Users (MAU) is a metric commonly used to characterize a publisher in terms of the users that have engaged with an app over a period of a calendar month or 30 days. MAU can be used as a metric that captures the growth of a given publisher. From the advertising perspective, MAU can be used as an upper bound of the number of distinct users that are expected to generate impressions assuming that all the user’s ad requests have been filled. However, since this might not be always the case, Pixalate has followed a different approach and defined programmatic MAU (pMAU) as the monthly active users that have generated programmatic traffic.
In order to estimate the MAU (and its programmatic counterpart) for a given publisher, Pixalate has developed Machine Learning (ML) models that generate MAU predictions given various data points that characterize the publisher such as: a) app store information, b) user growth information, c) impression-level data, and d) various other composite metrics that are generated after removing any potential sampling biases in order to provide robust estimates. In the core of the MAU prediction system lies a modeling ensemble that combines predictions from many individual models and generates a single prediction for every given publisher.
User Engagement scoring captures the user engagement with a given app in terms of time spent consuming content.
Viewability scoring aims to score the performance of a given publisher in terms of the percentage of display impressions being viewed. For this score, Pixalate relies on its impression data and assigns scores in proportion to the viewability performance of a given publisher, after all the IVT has been removed (per MRC viewability guidelines).
