How Facebook works

I am sharing the attached document, "Understanding How Facebook Works," which details Facebook's ad measurement, filtration, and reporting methodologies for ad impressions.

It provides key information on:

- Accreditation Status: Facebook's adherence to IAB and MRC guidelines, noting that Feed and Right-Hand Column display ad impressions are currently accredited.

- Ad Placement Types: Details on the main placements under consideration for MRC accreditation (Feed, Right-Hand Column, In-Stream Reels, and Video Feeds).

- Ad Metrics and Reporting: How Facebook counts and reports billable impressions and how advertisers can access this data via Ads Manager or the Marketing API.

Please let me know if you have any questions after reviewing the information.

I am sharing the details regarding Meta's ad impression and video measurement methodologies for your review.

Ad Impression Measurement

An ad impression is counted immediately when any part of the ad (more than zero pixels) becomes visible on the screen for more than zero seconds. A JavaScript check is performed every 100 milliseconds to confirm visibility. If the check passes, an impression is counted. This check is not available for all devices (e.g., some feature phones and Stories display), which use a "rendered" measurement methodology instead.

Video Ad Impression Measurement

Meta's display impression metric counts an impression if any pixel of the ad is visible in the viewport.

To align with MRC guidelines, Meta also offers a video play metric, which only counts an impression when the video begins playing.

2-Second Continuous Video Play Measurement

A 2-second continuous video play is counted when 50% of the video player's pixels are visible on screen and continuous playback exceeds 2 seconds. This is available in the Ads Manager and the 1P viewability report (labeled "Viewable Video Plays").

Other Measurement Details

- Compound Tracking: Not used; each impression is evaluated individually.

- Offline Measurement: Offline impressions are stored and sent only if the user comes back online within one day.

- Auto-Refresh: Facebook auto-refreshes ad impressions in the Right Hand Column (RHC) on desktop unless a user action indicates interest (like hovering). These account for half of all RHC impressions.

- Continuous Play: Video ads may be counted even if displayed without user interaction as the video feed auto-advances.

Please let me know if you have any questions regarding these standards.

I am sharing the key details regarding Facebook's ad measurement methodology.

Key Ad Impression Measurement Protocols:

- Inactivity Rules: To ensure accurate measurement, story auto-advance pauses if the screen is off or if a user is inactive after opening a video. Impressions are not counted when the screen is off.

- Cache-Busting: Facebook uses the Cache-Control HTTP header and a client impression timestamp (cts parameter) to prevent browsers from caching and reusing the same ad.

- Data Logging: All raw impressions are logged, processed through a stream architecture, and then aggregated into the data warehouse. Non-billable events are flagged and removed during this process.

Known Limitations and Edge Cases:

- Negligible Limitations: Facebook considers the effects of caching and abandonment to be negligible due to its viewport-based measurement methodology and cache-busting techniques.

- Ad Blocking/Disabled Features (Desktop Only): Measures are in place to prevent ads blocked by AdBlock/AdBlock Plus from being measured as billable. However, disabling image rendering or JavaScript can result in impressions being measured when the ad was not actually viewed.

- Server-Client Connection Break: Impressions may not be captured if a user closes the app or browser very quickly after an ad impression is generated.

- Browser/Device Issues: Viewability is not available for mobile web on Safari and Chrome. Video autoplay does not work on Safari due to missing connection type data. Missing impression events may occur on older iPads due to hardware limitations, but customers are not billed for them.

- External Factors (Desktop Only): Google Chrome’s pre-rendering function may inflate impressions, and Facebook does not measure impressions that are covered by other windows or moved off-screen.

Facebook's ad impression measurement methodologies, partner qualification controls, key business relationships, and filtering methodologies.

The document covers several specific topics, including:

- Measurement methodologies for feature phones and mobile devices with inconsistent connectivity.

- How ad impression measurement is maintained during user interruptions, such as pulling down the notification tray on Android or opening the chevron menu.

- The logging duration when a minimized video overlays an ad.

- Facebook’s role as both a "publisher" and "ad server" and its lack of significant business relationships in the transactional chain.

- The requirements for advertisers and content creators, including valid accounts, compliance reviews, and adherence to various policies.

- The use of a forced duration for certain non-skippable video ads.

- Filtering methodology for identifying and removing invalid activity, including the use of robots exclusion files.

I am sharing the details of the processes Facebook employs to identify and filter out non-human, invalid, and fraudulent activity from advertising reporting.

1. Identification of Non-Human Activity

Facebook uses the IAB Spider/Robot Blacklist and the TAG Data Center IP list, reviewed at least quarterly, to update its user agent blacklist. Requests with empty user agents are blocked, and ad impressions not generated by known browsers (via the IAB whitelist) are filtered. Bots are prevented from logging billable impressions.

2. Internal Traffic Filtering

Impressions from Facebook's internal locations and production system testing are filtered out by comparing IP addresses against a comprehensive list managed by the operations team. Test data is completely segregated from customer data.

3. Activity-Based Filtering

Events are analyzed within a defined time window relative to the user source. If events from a single user source exceed a set threshold, subsequent events are logged but marked as non-billable and excluded from reporting.

4. Identification and Filtering of Invalid Impressions and Accounts

Facebook employs multiple systems and automated checks for common invalid traffic (e.g., bots, spiders, adware, malware, fake accounts).

- Filters and thresholds are continuously monitored and updated.

- Methods are applied at both the impression and user levels (i.e., an impression can be marked non-billable, or a user can be removed).

- "Ad impressions from accounts later deemed invalid" refers to delivered impressions that were later invalidated due to fraudulent user activity. These impressions are not counted.

- For in-stream ads on Reels, additional methods are used to prevent monetization from invalid partner accounts.

5. Machine Learning Processes

Facebook uses machine learning models with over a hundred features to predict Invalid Traffic (IVT) based on score thresholds.

- Data sources include account characteristics, behavior logs, and ad interactions.

- Monitoring processes on training data trigger alerts for human intervention upon anomaly detection (e.g., a decrease in accuracy).

- Models are continuously retrained and evaluated.

- Machine learning training and evaluation data are maintained with minimal bias; IVT processes are not approved if bias is detected.

Thank you for providing the detailed information regarding Facebook's invalid traffic (IVT) detection and data reporting practices.

This documentation clearly outlines:

- IVT Detection: The combined use of machine learning and human review, the appeal process for false positives and negatives, and the dynamic optimization of the human-to-ML ratio.

- Decision Rate: Facebook's ability to collect sufficient information for IVT determination on 100% of Facebook and Instagram traffic.

- Data Consistency and Filtering: The time check and maximum activity frequency limits used to filter invalid impressions.

- Data Reporting: The availability of reporting tools in the Ads Manager, standard viewability export requests, and quality assurance processes.

- Data Reissuance/Refunds: The formal investigation process for spend issues, with credit or refund decisions made on a case-by-case basis.

- Data Retention: The varying retention policies with minimum standards, noting that customers are responsible for storing older reporting data.

This information is very helpful for understanding your platform's commitment to data integrity and ad quality.

#mynewupdate #facebook #socialmedia