In the world of digital advertising and anti-fraud, one of the biggest challenges faced by advertisers and partners is distinguishing between fraud and non-fraud in equivalent traffic. Understanding where traffic is high-quality and where it is hidden is no easy task. However, there is a solution that makes this much easier—cohort analysis.
Cohort analysis is like a bright flash that illuminates user behavior across different groups. And when we talk about fraud, these flashes allow us to see where the deception is hiding. With cohort analysis, we can accurately determine which traffic is high-quality and where fraud is hiding.
What is Cohort Analysis?
Cohort analysis is the process of dividing data into groups (cohorts) based on common characteristics or behaviors. In advertising, this can be, for example, dividing users by the time of their first visit to the site, traffic source, or devices. By analyzing the behavior within these groups, we can spot anomalies that might otherwise go unnoticed using a standard approach.
Imagine that you see data about users but don't know where they came from or how they behave. Cohort analysis helps to draw a clear picture. We can divide users by traffic sources, and this is where the fun begins. As practice shows, some of these groups may exhibit abnormal patterns that indicate fraud.
Why is Cohort Analysis So Important?
When traffic comes through different channels, and you don’t know exactly where fraud may be hiding, it is crucial to find a way to accurately separate "good" traffic from "bad" traffic. This is where cohort analysis comes in. It helps not only to analyze user behavior but also to identify the groups where fraud might be lurking.
Fraudsters, like all other users, pass through the funnel, but their behavior is different. They may initially act like regular users, but at a certain point, they start showing anomalies—too many clicks from the same device, constant returns to the page, illogical behavior.
How Does Cohort Analysis Work to Identify Fraud?
Group Division
First, it is important to properly divide traffic into groups. For example, users can be grouped by the time of their first visit, traffic sources, or devices. Dividing into such cohort groups gives us the opportunity to look at them through different lenses and track changes in their behavior.
Behavior Tracking
Each group needs to be carefully studied. Here we begin to see patterns. Some groups, for example, those coming through an ad network, may show strange patterns—clicks from different IPs but the same device. This could indicate fraud. Similar anomalies can be found in groups coming from push notifications, where redirects occur in hidden tabs.
Identifying Anomalies
The next step is to analyze anomalies. If users from the same device but different IP addresses begin to behave suspiciously, this is a reason for deeper investigation. It is very important to identify such groups before they cause losses.
Creating Reports and Analysis
After creating cohort groups and analyzing their behavior, we can generate reports that show exactly where failures in the system occur. This allows us to precisely track where and when fraudulent actions are happening.
Example from Kaminari Click's Practice
One of our clients encountered fraudulent traffic in RTB campaigns. The traffic came through several ad networks, and the conversion rates were inflated. Cohort analysis allowed us to divide users by traffic sources and their activity times. And here’s what we found: one group, coming through a particular network, showed strange results—users with the same devices but different IP addresses. This was a clear indicator anomaly pointing to fraud.
As a result, after we excluded these groups from the ad campaigns, the client reduced the share of fraudulent traffic and improved overall conversion.
Cohort Groups in Kaminari
At Kaminari Click, we identify several cohorts for more precise traffic analysis and filtering. Here’s what we use:
Cohort by Time
- First-time users: These users visit the site for the first time, and we track how their behavior develops afterward.
- Returning users: We monitor what happens with users who return to the site.
Cohort by Traffic Source
- Traffic from RTB campaigns: We analyze users coming through ad networks.
- Traffic from direct buys: We separate traffic coming through direct links and without network involvement.
Cohort by Devices
- Mobile devices: We track users coming through mobile devices.
- Desktop devices: We analyze the behavior of users on PCs.
Cohort by Geographical Location
- Geo-analysis: We divide users by regions, which helps identify anomalies, such as multi-accounting.
Cohort by Device Type
- Desktop browsers: We analyze the behavior of users coming through desktops.
- Mobile browsers: We track users using mobile browsers.
Cohort by User Action
- Clicks and conversions: We monitor user behavior before and after conversion.
- Repeated actions: We identify repeated actions that may indicate fraud.
Cohort by Time Spent on Site
- Short sessions: We analyze users who spend minimal time on the site.
- Long sessions: We observe users who stay on the site for too long.
How Kaminari Implements Cohort Work for Clients, Especially Ad Networks
Kaminari Click actively implements cohort analysis in working with clients, particularly ad networks, to improve traffic quality and protect against fraud. This approach helps not only minimize fraudulent activities but also optimize ad campaigns, providing a higher ROI for our clients. Here’s how we apply cohort analysis in practice:
Creating Custom Cohorts for Ad Networks
For each client, depending on their business goals and the specifics of their ad campaigns, we create unique cohort groups. This can be especially useful in working with ad networks, as different networks may generate different types of traffic. Cohort analysis allows us to segment traffic based on many factors, including traffic sources, devices, and geographical location.
Identifying Fraud through Cohort Behavior
Cohort analysis allows us to track user behavior in real-time and identify anomalies. We analyze metrics such as the number of clicks from one device, suspicious actions from the same IP addresses, as well as session duration on the site. This allows us to timely detect fraud and exclude it from ad campaigns.
Constant Adjustment and Filter Optimization
We configure filters for each traffic source and continuously update them based on the data obtained from the cohorts. This helps us minimize fraud and improve traffic quality.
Feedback and Traffic Quality Improvement
After we identify fraudulent patterns in cohort groups, we share this data with partners and clients. This allows ad networks to improve traffic filtering and increase the effectiveness of their campaigns.
Conclusion
Cohort analysis is a key tool for those who want to understand exactly where fraud is hiding in their traffic. It helps to divide users into groups and track their behavior, identifying anomalies and deviations from normal patterns. This approach not only helps minimize fraud but also improves traffic quality, increasing ROI.
Now that you know how cohort analysis works, use this method to improve the quality of your ad campaigns. Protect your budgets from fraudsters and achieve great results.
