Video completion rate

How Video completion rate Works

+---------------------+      +----------------------+      +---------------------+
|   Video Ad Request  |----->|  Ad Server & Tracker |----->|  User Device Plays  |
+---------------------+      +----------------------+      +---------------------+
           |                             |                             |
           |                             |                             ▼
           |                             |                  +---------------------+
           |                             |                  |   Viewing Data Sent |
           |                             |                  | (Start, 25%, 50%...) |
           |                             |                  +---------------------+
           |                             |                             |
           |                             |                             ▼
           |                             |                  +---------------------+
           └-----------------------------'----------------->|  Fraud Detection    |
                                                              |  System (VCR Analysis) |
                                                              +---------------------+
                                                                         |
                                                                         ▼
                                                     +---------------------------------------+
                                                     |    Pattern & Anomaly Detection        |
                                                     | (e.g., 100% VCR from one IP block)    |
                                                     +---------------------------------------+
                                                                         |
                                                                         ▼
                                                           +--------------------------+
                                                           |   Flag/Block Fraud       |
                                                           +--------------------------+

Video Completion Rate (VCR) is a critical metric for gauging video ad engagement, but its true power in security lies in anomaly detection. By analyzing VCR data, fraud detection systems can identify patterns inconsistent with human behavior, acting as a primary filter against invalid traffic and ensuring ad spend is directed at genuine viewers.

Data Collection and Tracking

When a user’s device requests and plays a video ad, tracking pixels fire at key milestones. These typically include the start of the video, and often quartile-based checkpoints (25%, 50%, 75%, and 100% completion). This granular data is sent from the user’s device back to the ad server or a third-party tracking service. Each event is logged with associated metadata, such as the viewer’s IP address, user agent, device ID, and timestamps, which are essential for subsequent fraud analysis.

Behavioral Pattern Analysis

A fraud detection system ingests this viewing data and aggregates it to analyze behavioral patterns. Human viewers exhibit diverse and varied completion rates; some drop off early, while others watch to the end. In contrast, bot traffic often displays unnatural, uniform behavior. For example, a large volume of views originating from a single data center and all showing a 100% completion rate is a strong indicator of non-human traffic designed to fraudulently maximize ad revenue.

Anomaly Detection and Mitigation

The system flags significant deviations from established human-behavior benchmarks. Anomalies that trigger alerts include impossibly high VCRs across a specific IP range, unnaturally low completion rates that suggest automated “view-and-drop” bots, or completion times that are faster than the video’s actual length. Once a pattern is identified as fraudulent, the system can automatically block the responsible IP addresses or device IDs from receiving future ads, thereby protecting the advertiser’s budget and purifying the data pool.

Diagram Element Breakdown

Video Ad Request & User Device

This represents the initial step where a user’s browser or app requests an ad. The user’s device is where the video is ultimately played and where the viewing data originates. Its characteristics (IP, user agent) are key data points.

Ad Server & Tracker

This is the central hub that serves the video ad and the associated tracking code. It’s responsible for logging the impressions and receiving the completion data (e.g., quartiles) sent back from the user’s device.

Fraud Detection System

This is the core analytical engine. It aggregates viewing data from countless devices, analyzes VCR metrics in conjunction with other signals (IP, geo-location, device type), and runs algorithms to spot suspicious patterns indicative of bot activity.

Flag/Block Fraud

This is the final, actionable outcome. Based on the analysis, the system flags suspicious traffic for review or automatically adds the source (IP, device ID) to a blocklist to prevent it from consuming future ad impressions, thereby protecting the campaign budget.

🧠 Core Detection Logic

Example 1: VCR by IP Address Monitoring

This logic tracks the average Video Completion Rate for each IP address. It is used to identify non-human traffic from sources like data centers or botnets, which often exhibit unnaturally uniform viewing patterns to maximize fraudulent ad revenue.

FUNCTION analyze_ip_vcr(ip_address, video_views):
    // Define thresholds
    MIN_VIEWS_THRESHOLD = 50
    HIGH_VCR_THRESHOLD = 98.0 // in percent

    // Check if the IP has enough views for a reliable analysis
    IF video_views.count(ip=ip_address) < MIN_VIEWS_THRESHOLD:
        RETURN "Insufficient data"

    // Calculate the average VCR for the given IP
    completed_views = video_views.filter(ip=ip_address, completed=True).count()
    total_views = video_views.filter(ip=ip_address).count()
    average_vcr = (completed_views / total_views) * 100

    // Flag IPs with abnormally high completion rates
    IF average_vcr >= HIGH_VCR_THRESHOLD:
        FLAG ip_address AS "Suspicious: Unnaturally high VCR"
        ADD ip_address TO blocklist
    ELSE:
        MARK ip_address AS "Normal"

    RETURN average_vcr

Example 2: Session Heuristics with Quartile Analysis

This logic examines not just full completions, but how quickly users drop off. Bots may be programmed to simply start a video and leave, leading to high “start” numbers but extremely low 25% or 50% quartile completions. This helps catch inefficient bots that don’t mimic full engagement.

FUNCTION check_quartile_dropoff(session_data):
    // Define drop-off thresholds
    QUARTILE_25_MIN_RATE = 20.0 // Minimum percent of views reaching 25%
    QUARTILE_50_MIN_RATE = 10.0 // Minimum percent of views reaching 50%

    // Calculate completion rates for each quartile
    total_starts = session_data.count(event='start')
    q1_completes = session_data.count(event='quartile_25')
    q2_completes = session_data.count(event='quartile_50')

    q1_rate = (q1_completes / total_starts) * 100
    q2_rate = (q2_completes / total_starts) * 100

    // Flag sessions with a severe drop-off after starting
    IF q1_rate < QUARTILE_25_MIN_RATE AND q2_rate < QUARTILE_50_MIN_RATE:
        FLAG session AS "Suspicious: High initial drop-off rate"
        SCORE session.user AS high_risk
    ELSE:
        MARK session AS "Normal engagement pattern"

    RETURN "Analysis complete"

Example 3: Behavioral Rule for Impossible Completions

This logic detects a clear form of fraud where completion events are fired faster than the video's duration allows. This can only be achieved by automated scripts, not human viewers, making it a definitive rule for identifying and blocking sophisticated invalid traffic (SIVT).

FUNCTION detect_impossible_completion(view_event):
    // Get video metadata
    VIDEO_DURATION = getVideoDuration(view_event.video_id) // e.g., 30 seconds

    // Get timestamps from the view event
    start_time = view_event.timestamp_start
    completion_time = view_event.timestamp_complete
    actual_watch_time = completion_time - start_time

    // Check if the completion time is plausible
    IF actual_watch_time < VIDEO_DURATION:
        FLAG view_event.source_ip AS "Fraudulent: Impossible completion time"
        // Immediately add the source IP to a real-time blocklist
        BLOCK_IP(view_event.source_ip)
        RETURN "Fraud Detected"

    RETURN "Legitimate View"

📈 Practical Use Cases for Businesses

• Campaign Budget Shielding – By identifying and blocking IP addresses with impossibly high VCRs (e.g., 100% from a data center), businesses prevent ad spend from being wasted on automated bots that are not real potential customers.
• Data Integrity for Analytics – Filtering out traffic with abnormal VCR patterns (e.g., massive views with 0% completion) ensures that engagement metrics are based on real human interactions, leading to more accurate marketing decisions and performance analysis.
• Improving Return on Ad Spend (ROAS) – Focusing budget on sources that demonstrate human-like VCR distributions ensures ads are served to genuinely engaged audiences, which increases the likelihood of conversions and improves overall campaign profitability.
• Publisher Quality Vetting – Advertisers can use VCR and quartile metrics to evaluate the quality of traffic from different publishers, prioritizing partners who deliver engaged human audiences over those with suspicious, bot-like viewing patterns.

Example 1: IP Blocklist Rule for High VCR

// This rule is used in a server-side firewall or fraud-filtering service.
// It identifies and blocks IPs that exhibit non-human viewing behavior.

RULE "HighVCR_IP_Block"
WHEN
  // Aggregate VCR data over the last hour
  let ip_stats = Analytics.query(
    "SELECT AVG(VideoCompletionRate) as avg_vcr, COUNT(views) as view_count
     FROM video_views
     WHERE timestamp > NOW() - 1h
     GROUP BY source_ip"
  );

  // Find IPs with high VCR and sufficient volume
  let suspicious_ip = ip_stats.find(ip -> ip.avg_vcr > 0.98 AND ip.view_count > 100);

THEN
  // If a suspicious IP is found, add it to the blocklist
  IF suspicious_ip IS NOT NULL:
    Firewall.add_to_blocklist(suspicious_ip.source_ip, "High VCR Anomaly");
    Log.alert("Blocked IP " + suspicious_ip.source_ip + " for high VCR.");
  END

Example 2: Session Scoring Based on Engagement Drop-off

// This logic is used within a user session analysis tool to score traffic quality.
// It penalizes sessions that show signs of being automated 'view-and-drop' bots.

FUNCTION calculate_session_score(session_events):
  let score = 100; // Start with a perfect score
  let video_starts = session_events.filter(e -> e.type == "video_start").count();
  let quartile_25_views = session_events.filter(e -> e.type == "video_quartile_25").count();

  IF video_starts > 5: // Only apply to sessions with multiple video views
    let q1_completion_ratio = quartile_25_views / video_starts;

    // Penalize if there's a huge drop-off after the video starts
    IF q1_completion_ratio < 0.1: // Less than 10% reach the first quartile
      score = score - 50;
      Log.info("Penalized session for severe engagement drop-off.");
    END
  END

  RETURN score;

🐍 Python Code Examples

This function simulates analyzing a list of video view events. It calculates the Video Completion Rate (VCR) for a specific IP address and flags it if the rate exceeds a threshold, a common method for detecting bot-like behavior from a single source.

def analyze_ip_completion_rate(events, ip_address, threshold=95.0, min_views=20):
    """Analyzes video completion rate for a given IP to detect anomalies."""
    ip_events = [e for e in events if e['ip'] == ip_address]

    if len(ip_events) < min_views:
        return f"Insufficient data for {ip_address}"

    completed = sum(1 for e in ip_events if e['status'] == 'completed')
    total = len(ip_events)
    vcr = (completed / total) * 100

    if vcr > threshold:
        print(f"ALERT: IP {ip_address} has an abnormally high VCR of {vcr:.2f}%")
        return "Suspicious"
    else:
        print(f"IP {ip_address} has a normal VCR of {vcr:.2f}%")
        return "Normal"

# Example Usage
video_views = [
    {'ip': '192.168.1.1', 'status': 'completed'},
    {'ip': '203.0.113.10', 'status': 'completed'},
    {'ip': '203.0.113.10', 'status': 'completed'},
    # ... Assume 98 more 'completed' views from 203.0.113.10
]
video_views.extend([{'ip': '203.0.113.10', 'status': 'completed'}] * 98)
analyze_ip_completion_rate(video_views, '203.0.113.10')

This code identifies invalid view data by checking for impossible scenarios, such as a video being marked as 'completed' in less time than its actual duration. This is a definitive way to catch fraudulent events generated by automated scripts.

def validate_view_duration(view_event):
    """Flags view events that are impossibly short."""
    video_duration = view_event.get('video_length_sec', 30)
    time_watched = view_event.get('time_watched_sec', 0)

    if view_event.get('status') == 'completed' and time_watched < video_duration:
        print(f"FRAUD DETECTED: View from {view_event['ip']} has impossible duration.")
        return False
    return True

# Example Usage
legitimate_view = {'ip': '198.51.100.5', 'status': 'completed', 'video_length_sec': 60, 'time_watched_sec': 61}
fraudulent_view = {'ip': '198.51.100.6', 'status': 'completed', 'video_length_sec': 60, 'time_watched_sec': 15}

validate_view_duration(legitimate_view)
validate_view_duration(fraudulent_view)

Types of Video completion rate

• Quartile Completion Rates - This involves tracking the percentage of viewers who reach the 25%, 50%, and 75% marks of a video. It helps identify bots that are programmed to watch for only a few seconds before dropping off, which reveals a steep decline between the "start" event and the first quartile.
• Absolute Video Completion Rate (VCR) - This is the standard metric showing the percentage of total impressions that resulted in a fully completed view. In fraud detection, an unnaturally high VCR (e.g., 99-100% across a large volume of traffic) from a single source is a strong indicator of non-human traffic.
• Audible and Visible on Completion (AVOC) - This measures what percentage of completed views were both audible and fully visible on the screen when they finished. A low AVOC score for completed videos can indicate ad stacking or backgrounded tabs, common tactics used in sophisticated impression fraud schemes.
• User-Level VCR - This method analyzes the average VCR for individual users or devices over time. A user account that consistently exhibits a 100% completion rate across hundreds of different video ads is likely a bot, as this pattern is inconsistent with natural human browsing and viewing behavior.

🛡️ Common Detection Techniques

• Behavioral Analysis - This technique examines VCR patterns in aggregate to define a baseline for normal human behavior. It then flags outliers, such as IP addresses or device IDs that exhibit impossibly high completion rates (e.g., 100% VCR over thousands of views), which strongly indicates automation.
• Quartile Drop-Off Analysis - This method tracks how many viewers proceed from one quartile of a video to the next (e.g., from 25% to 50% complete). A sharp, unnatural drop-off after the first few seconds can indicate low-quality bot traffic designed only to register an impression.
• Session Scoring - This technique analyzes the VCR across a single user session. If a user has an abnormally high completion rate across dozens of videos in a short time, the session is flagged as suspicious, as this behavior is not typical for humans.
• IP Reputation and Geolocation - VCR data is correlated with IP information to identify anomalies. For example, if a batch of views with 100% VCR originates from a known data center instead of a residential ISP, it is almost certainly fraudulent traffic.
• Timestamp Anomaly Detection - This technique validates the time elapsed between the start of a video and its completion event. If a completion is registered faster than the video's actual duration, it is definitively marked as fraudulent because it's technically impossible for a human viewer.

🧰 Popular Tools & Services

Tool	Description	Pros	Cons
Ad Verification Platform	Offers a suite of tools that measure viewability, brand safety, and invalid traffic (IVT) for video ads. It analyzes VCR data alongside other metrics to provide a comprehensive fraud score.	MRC accreditation, detailed reporting, pre-bid and post-bid protection, global coverage.	Can be expensive for smaller businesses, may require technical integration.
Click Fraud Protection Service	Specializes in real-time detection and blocking of fraudulent clicks and impressions on PPC and social ad campaigns. Uses VCR data as one of many signals for behavioral analysis to identify bots.	Real-time blocking, automated IP exclusion, easy integration with major ad platforms, session video recordings.	Primarily focused on click-based campaigns; may have less emphasis on sophisticated video-specific fraud.
Programmatic Analytics Suite	An omnichannel analytics platform for programmatic advertising that monitors traffic quality across CTV, mobile, and web. It uses VCR metrics to identify invalid traffic sources within the supply chain.	Covers a wide range of devices including CTV, detects over 40 types of invalid traffic, provides supply path transparency.	Can be complex to navigate, insights may require dedicated analyst interpretation.
Anti-Fraud Cloud Phone System	A specialized service that uses real cloud-based mobile devices to test and optimize ad creatives by simulating organic user behavior. It helps identify how VCR is affected by different user personas and networks.	Provides hardware-level simulation for accurate A/B testing, good for audience segmentation analysis.	Niche use case focused on testing rather than large-scale, real-time campaign protection.

📊 KPI & Metrics

When deploying fraud detection systems based on Video Completion Rate, it is vital to track metrics that measure both the accuracy of the detection and its impact on business goals. Monitoring technical KPIs ensures the system is correctly identifying fraud, while business metrics confirm that these actions are positively affecting campaign performance and budget efficiency.

Metric Name	Description	Business Relevance
Invalid Traffic (IVT) Rate	The percentage of video views identified as fraudulent based on VCR anomalies and other signals.	Directly measures the volume of fraud being caught, justifying the need for a protection solution.
False Positive Rate	The percentage of legitimate views that were incorrectly flagged as fraudulent by the system.	A high rate indicates over-blocking, which can harm publisher relationships and limit campaign scale.
Wasted Ad Spend Reduction	The amount of advertising budget saved by blocking fraudulent views identified through VCR analysis.	Provides a clear return on investment (ROI) for the fraud detection tool or service.
Clean Traffic VCR	The average Video Completion Rate for traffic that has been filtered and deemed legitimate.	Offers a true benchmark for creative performance and audience engagement after removing bot interference.

These metrics are typically monitored through real-time dashboards provided by the ad fraud detection service. Alerts can be configured to notify teams of sudden spikes in fraudulent activity or unusual VCR patterns. This feedback loop allows for continuous optimization, such as refining detection rules, updating IP blocklists, or pausing low-quality traffic sources to protect the campaign's integrity and budget.

🆚 Comparison with Other Detection Methods

Detection Accuracy

VCR analysis is highly effective at spotting certain types of bot behavior, especially unsophisticated bots that play videos to 100% completion from data centers. However, it can be less accurate against advanced bots that mimic human-like drop-off rates. In contrast, signature-based detection is excellent at identifying known bots but fails against new or mutated threats. Behavioral analytics offer a more robust approach by creating holistic user profiles but require more data and processing power.

Processing Speed

Analyzing VCR is computationally moderate. It can be done in near real-time for post-bid analysis but is generally slower than signature-based filtering, which involves a simple lookup against a list of known bad actors. CAPTCHA challenges are much slower, as they require direct user interaction and interrupt the user experience, making them unsuitable for passive video ad verification.

Scalability and Suitability

VCR analysis scales well for post-bid and reporting use cases, where large datasets of view events can be aggregated and analyzed. It is less suitable for pre-bid environments where decisions must be made in milliseconds. Signature-based filtering is highly scalable and ideal for pre-bid blocking due to its speed. Behavioral analytics are highly scalable but are typically used for post-bid analysis and scoring due to their complexity.

⚠️ Limitations & Drawbacks

While analyzing Video Completion Rate is a valuable technique in fraud detection, it is not a complete solution. Its effectiveness can be limited in certain contexts, and it can be bypassed by sophisticated fraudsters, making it essential to understand its inherent drawbacks in traffic filtering.

• Sophisticated Bot Evasion – Advanced bots can mimic human behavior by randomizing completion rates, making them difficult to distinguish from legitimate users based on VCR alone.
• False Positives – Overly strict rules, such as flagging all views with high completion rates, may incorrectly block highly engaged human users, leading to lost opportunities.
• Lack of Context – VCR doesn't explain why a viewer dropped off. A low VCR could indicate a disengaged user or simply a poorly placed, intrusive ad, not necessarily fraud.
• Inapplicability to Short-Form Content – For very short videos (e.g., 6-second ads), completion rates are naturally high for all users, making VCR a less effective indicator for differentiating bots from humans.
• Post-Bid Limitation – VCR is a post-view metric, meaning the ad impression has already been served and paid for before the data can be analyzed. It helps in future blocking but not in preventing the initial fraudulent view.

In scenarios with advanced threats or where pre-bid prevention is critical, hybrid strategies combining VCR analysis with IP filtering, device fingerprinting, and machine learning are more suitable.

❓ Frequently Asked Questions

Can a high Video Completion Rate ever be a bad sign?

Yes, an unnaturally high VCR (e.g., 99-100%) across a large volume of traffic from a single source is a strong indicator of fraud. Bots are often programmed to watch ads to completion to maximize revenue for fraudulent publishers, a pattern inconsistent with diverse human viewing behavior.

How does VCR analysis work for short ads, like 6-second bumpers?

For very short ads, VCR is less reliable as a standalone fraud indicator because even humans have very high completion rates. In these cases, fraud detection systems rely more heavily on other signals like IP reputation, device fingerprinting, and checking for impossible completion times to identify invalid traffic.

Does VCR analysis stop fraud in real-time?

Generally, no. VCR is a post-view metric, meaning the data is analyzed after the impression has already occurred. Its primary function is to identify fraudulent sources (like bot-infested websites or suspicious IP addresses) so they can be added to a blocklist to prevent future wasted ad spend.

Is a low VCR always an indicator of poor ad creative?

Not necessarily. While a low VCR can signal unengaging content, it can also be a sign of fraud. For example, some bots are programmed to generate a high volume of impressions by starting a video and immediately leaving, which results in a very low completion rate and can be flagged as invalid activity.

What is the difference between VCR and View-Through Rate (VTR)?

The terms are often used interchangeably, but they can have slightly different meanings depending on the context. VCR typically refers to the percentage of video *starts* that were played to completion, while VTR often refers to the percentage of total *impressions* that were completed. Both are used to measure engagement and detect anomalies.

🧾 Summary

Video Completion Rate (VCR) measures the percentage of viewers who finish watching a video ad. Within digital ad fraud prevention, it serves as a key behavioral metric to identify non-human traffic. Abnormally high or uniform completion rates across a traffic source often indicate automated bots designed to defraud advertisers, making VCR analysis crucial for protecting ad budgets and ensuring campaign data integrity.

🔗 Related Articles

• What is video completion rate (VCR) | AppsFlyer mobile glossary – https://www.appsflyer.com/glossary/video-completion-rate-vcr/
• Once Again, About Video Completion Rate - DecenterAds – https://decenterads.com/blog/once-again-about-video-completion-rate/
• 'Fully on-screen' CTV can help reduce ad fraud by 83% - study - StreamTV Insider – https://www.streamtvinsider.com/video/fully-screen-ctv-can-help-reduce-ad-fraud-83-study
• Video Ad Report Reveals Completion Rates at New Highs, Fraud Rates at New Lows – https://www.extremereach.com/blog/video-ad-report-reveals-completion-rates-at-new-highs-fraud-rates-at-new-lows/
• Mobile video ads achieve up to 88% in viewability rates, but ad fraud still an issue – https://www.mobilemarketingmagazine.com/mobile-video-ads-achieve-up-to-88-in-viewability-rates-but-ad-fraud-still-an-issue