Tuesday, February 19, 2008

SES London: Click Fraud Issues

I'm just out of the Click Fraud session where we had Shuman Ghosemajumder from Google, Jon Myers from MediaVest and Andrew Goodman from Page Zero Media presenting.

I scribbled notes on Google Docs. The whole conference centre is safely blanketed with wi-fi.

Let's start with Google. While running through the different click fraud strategies (attempts to do it, rather than prevent it) Shuman noted that there were regional fashions.

Automated approachs such as click bots and botnets are more common in Eastern Europe than elsewhere in the world.

Click farms (manual labour) are more popular in India and China.

Pay-per-click sites (in which people are paid to click on ads) are more common in United States and Canada than elsewhere in the world.

Shuman also noted that Google disqualifies just under %10 of all clicks as fraud. That's a huge amount of clicks for Google to refund. Although... Google doesn't actually refund most of those clicks. The vast majority of click fraud, Google says, is caught by their algorithms at the point of click and the advertiser is never charged in the first place.

Google does engage in reactive and manual investigations - these are the type that are set up with a significant client complains. Here they'll use human skill as well as algorithms to examine what had happened (rather than catching something live).

Less than 0.02% of click fraud is found via these reactive manual investigations.

Here's a breakdown of Google's anti-click fraud steps:
  • Proactive
    • Filters
      • Automated algorithms which filter out in real time
      • Analyze all clicks
      • Accounts for vast majority of invalid click detection
    • Offline Analysis
      • Automated algorithms and manual analysis
      • Focused on the adsense network
      • Accounts for a smaller percentage than the filters
  • Reactive
    • Investigations
      • All advisers inquiries are investigated by the traffic quality team
      • Invalid click percentage is a negligible slice compared to #1 and #2
      • Relatively rare (<0.02%)
Google also pointed out that estimates for the total click fraud volumes are tend to overestimate the problem. If you sign up for a £10 AdWords account - ie, a daily cap of £10 then someone could send £1,000,000 worth of invalid clicks then you'd still only pay £10 (if Google doesn't catch it - which at that volume they would). So, to you it's £10 fraud but some of the "total value of click fraud" estimates would quote the £1,000,000.

Rushing off to the next sessions; updates to follow.

4 comments:

seo london said...

Interesting stuff. Thanks for filling us all in.

Jon Myers said...

Thanks for the session coverage Andy, good to see you last week

Andrew Girdwood said...

Thanks. I'm behind in writing up all my other notes - like your bit of the Click Fraud session!

Neil Matthews said...

only 0.02 gets through that they know about. Google set great store in their filters, none of which are open to scrutinty, except for one review by an Academic a number of years ago.

In my experience, the figure depends on where you advertise i.e. content network, search partners or just on Google.

I have personally received refunds from very simple click fraud attacks so I doubt the efficacy of Google filters.