Scoring Engine
The scoring engine leverages a machine learning approach using the above attributes, behavioral, and community data. It employs algorithms to automatically learn to recognize complex patterns in data. By discovering hidden patterns in a very large cross-sectional data set, the engine can accurately score each individual click according to its associated quality. Scores are generated by applying heuristic rules to the data set and deriving quality information based on performance against these rules. An example of a highly simplistic rule might be: "did the click occur from the same visitor during the same session within a 2 second period?" A positive response would indicate a simple repeat click, invalid but not necessarily fraudulent. The hundreds of rules employed by the scoring engine range from simple to highly complex, and take into account data from the visitor, the ad, the site, and the community data. 
Adaptive Intelligence
To accommodate varying traffic profiles among different online advertisers, publishers, and ad networks, the scoring engine will automatically adjust rules and thresholds to produce traffic quality scores that reflect propensity of conversion for each specific customer. The scoring engine continuously tracks the correlation of scores with a large community data set that includes conversion data. Because the shape of the traffic quality curve can differ across clients, the scoring system adapts to unique behavior patterns in traffic. For example, an ad network that caters to retailers in Asia might have very different traffic patterns than a B2B ad network in North America. Community Data
The power of the scoring engine is derived from the massive data set that has been assembled over the past three years from online advertisers, publishers, and ad networks. The community data set consists of billions of paid clicks aggregated across Click Forensics' diverse customer base as well as from a subset of the Click Fraud Network community. This large community data set includes all the attributes and behavioral data, including conversion information, for a wide variety of online advertising traffic. By comparing new traffic patterns with known valid and invalid traffic, Click Forensics can quickly identify anomalies in click behavior and correlate a meaningful estimate of conversion propensity to produce an accurate ClickScore. By scoring clicks across such a large set of online advertisers, ad networks, and publishers, Click Forensics has built a strong "reputation database." This reputation database is updated on a daily basis. The predictive quality scoring and the fraud engine are continually updated as new fraud patterns arise. Block Lists
The community data set also provides insight into a wide variety of invalid traffic that Click Forensics clients can access for immediate benefit. Known offenders, bots, and other sources of invalid traffic are updated constantly and provided to Click Forensics customers daily. Click Forensics recommends simply blocking these zero-value "visitors" from CPC campaigns wherever possible. - Known Offenders List - community wide
- Visitor Block List - specific to client's click-stream
- Ad Provider Bot List - Spiders
- Ad Provider Test List - Testing Spiders
- Known Bot List - IAB times 10
- Colo List - Colo Server Bot Fraud
- Hackers List - Hacker fraud
Clearly, simple block lists are not the solution to invalid traffic problems, but it is an easy first step that can deliver immediate value, stop wasted spend by advertisers, and reduce billing discrepancies for ad networks and publishers.
|
