Click Fraud: a Non Issue

55 comments

Eric Schmidt recently swore off click fraud as a non issue:

"Believe me, as a computer scientist, we have the ability to detect the invalid clicks before they reach advertisers," says Mr. Schmidt, who has a PhD in computer science from the University of California, Berkeley.

But that same article started with

NEW DELHI -- Click-click. Click-click. On a side street in New Delhi, past the pushcarts and down the stairs, the familiar sound of a computer mouse signals trouble for Google Inc.

In a basement cubicle under bare fluorescent lights, Rajiv Kumar sells the names of Web sites that pay people to click on Internet ads. His price: 300 rupees ($7.82).

"There's nothing wrong with any of this," he tells a prospective customer. Clicking is easy, he says.

Whatever happened to Boser's click fraud test? Why did Google recently settle on a suit for $90 million if it is no big deal? Why did Mr Reyes, Google's CFO, think click fraud was such a huge issue about a year ago?

"I think something has to be done about this really, really quickly, because I think, potentially, it threatens our business model," Google Chief Financial Officer George Reyes said Wednesday.

Google Blogoscoped recently mentioned this video about a guy wanting others to click his ads.

Comments

It is a non-issue

In the same way that shoplifting is a non-issue. I don't hear Walmart talking heads yammering about shoplifting threatening their business model.

Cost of doing business. Why must everything Internet-related become a major crisis while similar issues facing traditional business models are simply dealt with?

what irritates me about it

what irritates me about it is that google seems to be doing what so many big companies do once they get dominant market share: stifle innovation. obviously, a world with click fraud protection is better than one without, and obviously, google has the ability to detect click fraud, as eric schmidt himself noted. by not doing something that is beneficial and that they clearly have the ability to do, they are stifling innovation.

ultimately, though, google's added an unprecedented amount of efficiency to advertising even with click fraud, so value is still being added, and so the customers will keep on coming. when this is no longer the case is when google might start looking at more aggressively combating click fraud and adsense-sponsored splogs, IMHO.

Shoplifting?

I don't think that's a valid comparison.

When someone shoplifts from Wal-Mart, they don't make MORE money.

When click fraud happens, Google makes more money (in the short Term).

But, if you think the comparison is valid, how long do you think Wal-Mart could last if 75% of their merchandise was being shoplifted?

shoplifting and click fraud

can't compare, shoplifting takes, click fraud enriches.

We may as well spin this thing real nice:

Click fraud is our way of giving back to those less fortunate, probably works better than the World Bank, just help yourself. Google saves the world, ends poverty, money available for all!

Maybe that is the motivation behind the $100.00 laptop ;)

Turn Handle, make money, everybody happy.

__

Okay, so shoplifting might not be a good analogy. I was thinking theft, and click fraud is stealing.

I think 75% is a little high though and I don't buy into the conspiracy that Google allows click fraud when they detect it. Makes no sense for a long-term plan.

I also happen to think the market will make the final decision. Advertisers seem to think the current level of fraud is acceptable, or they'd stop advertising. That's the incentive for Google to keep click fraud at acceptable levels.

It is a cost of doing

It is a cost of doing business. However, you can't compare it to shoplifting because the lack of security in a store hurts the store, not other advertisers. You'd have to correlate it to newspaper circulation and misrepresenting the numbers. Or a TV station misstating their actual viewership.

The reason it is a big issue is because Google can do so much more to prevent it from happening. They turn a blind eye to it and allow advertisers to get squashed in the name of a buck.

It's a shit model You could

It's a shit model

You could make the same if not more money on a pay per hour time slots. If you blanket the clicks that go through in a time slot you eliminate the need for fancy inaccurate fraud/low quality detection algos that hide behind a PHD.

I've seen freaking click fraud data that they can't catch. Oh they can grab the 2 clicks in a row from Bombay easy enough. It's the clicks that come in once a week starting at 10:00 am on the dot and last through a 15 minute window over a hand full of adsense sites that they cannot catch.

They milk the system subtly over a wide network of sites and they do it so quitely it doesn't sound any alarms.

be careful what you wish for

"Believe me, as a computer scientist, we have the ability to detect the invalid clicks before they reach advertisers," says Mr. Schmidt, who has a PhD in computer science from the University of California, Berkeley.

When the people who write viruses and malware get pissed at Google for and start writing programs generating fake clicks from a widely distributed user base against some high paying keywords I wonder what he'll say.

limits of the system

I haven't read the originating articles, but if Dr. Schmidt truly believes that 100% of invalid clicks can be detected and not counted against advertisers, I think he's way off base, as graywolf suggests.

Likewise, surely everyone here realizes that Google really is doing work to identify and not count clicks that are fraudulent. But, as we all know, there are some serious weaknesses endemic to the structure of the internet which make it impossible to identify a fraudulent click past a certain point. Sure, we can use heuristics, pattern recognition, and click user identification methods to identify a lot of simpler exploits, but beyond that there are systemic limitations which make it impossible to identify all of it.

One component of fraud is also trying to identify the intent of a person clicking the link. Although Google would like to be able to read our minds, none of the search engines have that ability yet. Until they can detect our motives, there will always be some level of fraudulent clicks that are counted.

Big Words

Mr. Schmidt says shareholders shouldn't worry: Google monitors every click and refunds advertisers for many of the phony ones

Define MANY

Refunds?

In my years of advertising I've seen one instance where they gave me a refund. The other times I had to bring it to their attention, pull some teeth, and then get maybe half of the money back.

Imagine

Just stop for a moment and imagine a world without Adsense. What would that first page of Google look like?

Wow, now there's an image that knocks you out of your chair.

regarding refunding...

Google is referring to pulling out clicks and charges against your account before billing you. In other words, after Google receives clicks, they run algorithms to try to identify the suspicious/fraudulent ones, and then the don't charge for those. That's what they mean by refunding.

So, your ad is getting more "clicks" than you are being charged for. A few of those non-charged clicks are possibly real, live, interested consumers, so you may be getting more than you are paying for. In other cases, they're going to miss flagging some clicks that are actually fraudulent, and charge you for those.

One would hope that the two instances cancel each other out, but, it's rarely an ideal world...

Click Fraud

It's not that hard to id clickfraud. The problem has always been the motivation to end it. It's always been a money issue.

http://tinyurl.com/jbwmx

click fraud easy? yes and no

flyboy, the methods mentioned in the article you linked are simplistic, and are already in use by all major players.

So, yes, it's simple to back out clicks that are committed by easily identifiable agents/users/ips.

As other savvy people in this thread mentioned, there are ways around those simplistic identification methods, unfortunately.

Do you have a method which can tell whether a request is coming to you from an IP address through the action of human vs. a script? No, you don't. Yes, it's easy to see that tons of clicks per second or minute are coming from one IP address, and filter all those out (because we know that a human can't physically commit all those clicks that rapidly). Not so easy if the requests are not coming quicker.

And, there's a whole world of more complex methods which are harder than that to detect...

Those of us in the industry would love to be able to ex out 100% of all fraudulent clicks, and we're all working to find ways to better identify the bad usage. But, it's an arms race, and it's not clear that any victory is soon in sight.

This is a lot like fixing spam. There's lots of companies that want to fix spam, and some are better than others at keeping spam away from your in-box. But, nobody has "solved" it.

Already in use?

Quote:
the methods mentioned in the article you linked are simplistic, and are already in use by all major players.

Are they? Who really knows?

I noted at WMW that I've clicked on competitor's ads to check out their landing pages. (You may not like it, but it's not 'fraud', so shut your pie hole already). And given the hysteria that comment created, it's plain that either 3/4's of WMW participants are simpletons who are unable to figure out that Google should be automatically flagging my clicks as not valid, or they honestly believe that Google doesn't track my static IP address, know what words I advertise on, and discount any clicks made from my IP address. That's the kind of mindset we're dealing with when it comes to click fraud - it's all a big speculative mystery.

And if Google doesn't even have that level of click protection in place, then their systems are easier than a drunk virgin on prom night. But again, who really knows? Why we put up with this level of secrecy is beyond me.

(For me, I put up with it because I refuse to advertise in the crap content networks and I don't believe that fraud coming from their search results is a big deal at all - the campaigns are profitable and the odd dollar or two to copmetitive research isn't a concern)

The actual cost of click

The actual cost of click fraud on the content side will be and already is evidenced in the bid prices. Now, the search side ads are a whole 'nother story in the meantime. At the end of the day, the smart search marketer bids based on ROI and the overall level of search-side PPC bids would meet some sort of equilibrium with the amount of click fraud inherent to the system. Google makes more money on the value of the keyword economy than on the fraudulent activity, so they should be trying to clean this up more than anyone (while publicly denouncing the problem so that everyone stays engaged).

so you may be getting more than you are paying for

damn it Silver - you made me spit coffee on my keyboard

Stats

Quote:
And, there's a whole world of more complex methods which are harder than that to detect...

I'm not certain that this is the case. If you're 'thinking about it' then yes, the clickers can always out think you. However if you're a phd in stats and have mountains of information I submit that it's entirely possible that they can handily detect click fraud within certain ranges.

to detect or not to detect...

wheel wrote:

...within certain ranges...

what's that saying?

There are lies, damned lies and statistics.

When you start talking about using statistics, you're basically referring to various methods which could be used to estimate the likelihood of a click being fraudulent.

Everyone's already using statistical methods to identify the likelihood of a click being fraudulent or not. Those methods are already in use.

The point is, most of those methods do not give you an ironclad method of detecting all fraudulent clicks, which is what people are complaining about. An estimate of one's trustworthiness of a click is never going to be a certainty, and there's always a percentage of error involved with any sort of estimation method.

The blackhats know ways to avoid the simpler detection methods. I can't detail all the detection methodology without revealing proprietary secrets, but everyone I know inside the industry is using the best possible methods to screen out fraudulent clicks. The point is, there are technical barriers to being able to identify bad clicks, simply due to the way the internet is structured.

Let's say you have one single click from one IP address in your logs for this week. What statistical method are you proposing to detect if that click was initiated by a human clicking their mouse pointer on a link, versus a bot making the request? There isn't any statistical profiling that can be done to accomplish that, without further click patterns from that address.

Just one of the problems you have here is trying to detect if a human is on the end of the click or not. It's like a Turing Test, only we don't have nearly as much info as assessing a question/answer conversation to decide if the agent is human or machine -- you're trying to base your assessment off of "clicks".

This puzzle is not easily solved.

Yeah well

I don't know why you're dismissing statistics so decisively. google's got enough data to estimate and model just about anything to do with clicks, with as darn close to 100% accuracy as you'd like. It's not 'lies, damn likes, and statistics'. You're welcome to say that if you don't understand statistics and how it's applied, which certainly doesn't describe Google.

Who gives a rats ass about one click from someone once a week. That's hardly click fraud. If they're doing it in any sort of volume, scattered, 'random' or whatever, I have little doubt that Google's got the experts on board to easily detect that. Sheet, they used to be able to *accurately* detect the number of deaths in the French army from being kicked in the head by a horse, down to just a few deaths per year.

And the blackhats aren't getting around statistics. Not and do it in any sort of volume (or they are, but only because it's being allowed to happen). If they can algorithmically detect networks - any doubts here that they can do that handily? - they can easily detect click fraud.

(want another example? there's a statistical analysis that could be run on click charges that would probably give some good indicators of click fraud. Insurers use it to watch the amounts they pay in claims checks. Too many claims checks starting with the digit 3, and bingo, we've got fraud. It's known that the first digit should follow a trend, logorithmic IIRC. As soon as the first digits don't trend that way, there are external influences at work.)

And that's just off the top of my head from someone who hasn't looked at stats in years, and never knew much about it to begin with.

Personally if I was Google, I'd be showing two numbers - clicks, and 'paid clicks'. Who cares if they show that click fraud is X% if they're not charging for it anyway? That'd keep everyone happy IMO.

You make it

sound so hard Silver. When I see a "sandbox" on new publishers maybe I'll believe that G is applying the wiz bang PhD smarts evenly. Reliance on rhetoric when all that brain power is available is a bit weak.

I also have seen a number of

I also have seen a number of new sites come online and had higher revenue and average CPC the first month than the second. I believe they almost place an anti sandbox on new publishers, giving them a bit of a boost.

I have also known some who have transfered ownership of certain sites and noticed the same patterns - where the new publisher seems to get a bit more.

I also remember a friend showing me one day he got 13 clicks on 7 page impressions for about $15 earned on a virtually contentless single page AdSense spam blog.

piling on..

a bit here but if google can detect hidden text in a site, dispatch an email and process a re-inclusion request don't you think they could spot an MFA? Zero quality standards for publishers, most of the stuff doesn't even make the published webmaster guidelines. The last campaign I turned off ran 7 or 8 thousand terms, got tap tapped for a few days, a few thousand "vistors", zip conversions on a site that converts 10% normal serps and billed for every click that came through, my tracking was pretty much dead nuts with theirs, no allowance for click fraud although I believe 90% of the traffic was fraud. Screw requesting a credit, the time spent playing with the interface, KW uploading, creative tweaking, etc. made that thing a major loss, a credit would probably just piss me off.

First month

Quote:
I also have seen a number of new sites come online and had higher revenue and average CPC the first month than the second. I believe they almost place an anti sandbox on new publishers, giving them a bit of a boost.

Possibly this is because Google has to assign some sort of average CTR to new accounts, which over time is slowly replaced by the advertisers own - maybe lower - CTR.

For most of my accounts, the first month is the toughest.

statistical modelling

I'm not trying to make anyone upset. But. Statistical analysis is one area that is used by all the major players, and despite your insistence, it has limitations.

If you talk to a statistician, you will find that they have great ways of computing percentage error of various statistics and for trending, and they can compute percentage error of the percentage error. And even %age error of the %age error of the %age error.

Why would you need to compute percentage error for the glorified statistics if there weren't times when the statistics were wrong?

Insurance is a bad example, because it's estimating overall percentages of risk of things happening. Say, there's a general percentage of homes which will catch fire in a year's time. That's great for figuring out how to spread costs, but those statistics cannot tell you which house will be the one to burn. If those of us in the industry know general statistical ranges for the likely percentage of fraudulent clicks, we still wouldn't know which clicks are the fraudulent ones with that type of information. (It's a bad example, though, because in Insurance stats, it's clear WHICH houses burned last year -- in PPC ads, we can't tell across the board which clicks were bad with 100% certainty, so what would you base your statistics upon to begin with?!?)

What this boils down to is that the industry IS using statistics of clicks to predict the likelihood of a click source being fraudulent or not. I already mentioned heuristics and pattern recognition.

I cited the case of the one click in a week period because it is completely salient. The easy clicks to discount are the thousands of clicks per minute/hour from a single source. The harder to detect schemes could distribute clicks from numerous points on the internet in various ways. It's extremely naive to think that detecting fraud from a clever scheme is easy!

Certainly, identifying patterns from fraudulent behaviour is very useful. But, this is a way of estimating when a click should be counted as fraudulent, when you have no other evidence otherwise.

I won't bore you all with posting the definition of "estimate". But, an estimate is simply not the same as reality. It could unnecessarily flag a click as fraud when it is really from a good user, or it can miss flagging clicks that are bad.

If you still don't agree with me, please solve these two problems:

- Using loads of historical data, one should be able to predict how the stock market will function. Write a program to track the statistical trends and accurately predict them so that you can become rich by basing stock purchases off of your trend/pattern identification.

- Using statistics from a website's usage over time, predict how much traffic the site will have every day next year. If the site is being used by humans, make sure that you predict exactly how many visits and pageviews that site will have.

Why am I being seemingly snarky in posing these problems? Because each of these problems and click fraud have similar basic factors which mean that you can estimate and predict, but not accurately know the future. There are too many variables involved which cannot be contained in your calculations.

One variable I cited earlier is the inability to identify if a particular click is actually from a human clicking a mouse cursor over an ad link. If you can't create a way to accomplish this (and you can't), then you immediately start out with incomplete data for assessing whether a click is fraudulent or not.

Because of complexity, and because of the fact that many of the variables cannot be measured in calculating trustworthiness of a click, you can only estimate whether a click should be discounted as fraud with less than 100% certainty in a great many cases. And, an estimate will always have a percentage error, leading some folx to be unhappy at the prospect that they might not be paying for the interested human user they were hoping for.

It isn't that the industry players do not want to solve click fraud, it's that there are limitations to internet technology which make it so that it can't be fully addressed. Current methods can assess and discount clicks which are profiled as suspicious, and this is being done. But it should be understood that there is likely a percentage of error involved.

Your reasonings are nonsense.

You know even less about statistics than I do.

The stock market is a random walk. You CAN model a random walk. If the stock market is at 10,000 today - is it likely to be somewhere between 9K and 11K tomorrow? Sure it is. Is it likely to be 2K tomorrow? Not likely. Could happen, but not likely. There. I just modelled the stock market for you.

Fraudulent clicks are not a random walk. Suggesting that because error in a random walk model grows large means you can't accurately model a completely seperate and unrelated model is ridiculous.

Quote:
Insurance is a bad example, because it's estimating overall percentages of risk of things happening.

wtf? And I've done some research into using statistics to track fraud. For your information you can use statistics to point at a specific data point and say with *reasonable certainty* that it's fraudulent.

If you don't believe statistics, it's because you don't understand the math. Nobody every claimed that a model is a 100% accurate representation of a process, it's a model after all. But you can use statistics to claim that the model fits the process with enough accuracy to keep everyone happy. So what if it misses one click out of a 100K.

As for 'percentage of error', a sample of only 1000 will give you 95% accuracy, 19 times out of 20. I have no idea what a sample of millions of clicks will give you, but I'm sure the accuracy is good enough for anyone here.

typical...

It's typical to attack the person when you can't effectively discount their argument. It's weak to declare that I'm spouting nonsense if you cannot logically and knowledgeably refute the logic.

Your example disproves itself almost immediately. Sure, you can model the patterns of such things as the stock market, but there's still a huge possibility for error. Whose model effectively could foresee Black Monday, for instance? No one's.

I'm not stupid, and I can speak authoritatively on the subject because I *do* know what samples of millions of clicks can provide. I know what I'm talking about from direct experience because I implemented and manage the analysis systems for a Fortune 10 company website. I manage teams of PhDs who work on these systems.

With *some* clicks, one can say with certainty that it's highly likely to be fraudulent. (Do I have to repeat my example of thousands of clicks per second from a single IP?!?) With other clicks, it's not that easy.

And, as I said before, estimating is not precise. Modelling is estimating, and it has a percentage of error involved. 95% accuracy may be okay from your viewpoint, but what if the 5% error all falls on the clicks for a particular advertiser -- I bet it wouldn't be acceptable from his viewpoint.

There's no way to guage the accuracy in the first place, because we don't have any way to *know* which clicks were valid to begin with. Without being able to guage that, you have a lot higher error rate, inbuilt, from the very outset. We can only estimate, and we're pretty confident that we estimate well, but that still leaves the percentage of uncertainty that cannot be accurately measured.

With the insurance example I cited earlier, we all *know* precisely how many houses burned, and can frame some projections off of that *actual* data. With clicks, there's no accurate guage at the beginning as to which ones were absolutely fraudulent and which were absolutely good. Without that, all projections/models/forecasts are off by a higher percentage. By the nature of the beast.

It's circular logic to say that we can tell with certainty how much is fraudulent because we can estimate it based off estimates. To have a good estimate you have to have *actuals* measured at the beginning of the cycle! There's little purpose in continuing to argue on this subject if you cannot logically refute this one point.

motivation is the issue...

Google makes more money short-term on higher volumes (allowing some click fraud).

But in the long term, they put their business model at risk (enough click fraud some advertisers will bail).

Unfortunately, click fraud or not, people will continue to use Google ads (both search and content). So Google isn't motivated enough to stop this.

Given their inability to conclusively clean up the SERPs, getting rid of the AdSense and other spammers (after years and years of working at it), I would guess they are not able to clean up click fraud either.

I think wheel is being optimistic about Google's efforts/abilities here, although in the abstract he's right. All of this is fixable.

Probably there has to be a tighter tracing of identity. As SEO's and SEM's, if we knew that if one of our sites got banned for subdomain spam, all of our client sites would undergo human examination and possible banning. Or that all of our sites would be just dropped for three months (first offence), six months (second offence), life (third offence), we'd be a hell of a lot more careful with how we choose to optimise and rank.

Same thing with those committing click-fraud (either as agents of advertisers or the advertisers themeslves). They'd be out of business very quickly.

Gradually identity standards will be tightened in the virtual world (as in the real world) and a lot of these problems (with the addition of accountability) will just go away.

In the meantime, I do hope they improve their math.

I'm not stupid, and I can speak authoritatively ...

according to the poster.

The mark of a true expert is one who can make his area of expertise understandable in plain engish. That ain't happened yet. As a matter of fact, my eyes have been glazing over at an unprecedented rate.

Any oldtime stockbroker can tell you what the market is going to do tomorrow with absolute certainty. Fluctuate.

His argument is extremely

His argument is extremely clear, The summation of his argument ( in my opinion feeel free to correct me Silver) is:

You can not apply a macro model to tell you what occurs at individual instances.

I think

what Silver is saying is:

We can't detect click fraud.

Load a proxy list, send bot, cash check.

What no one is saying is that there are two ends to the transaction.

Not so clear

Quote:
You can not apply a macro model to tell you what occurs at individual instances.

That's what he's saying. Except it's done all the time. Which is what I'm saying.

Wheel I keep reading your

Wheel I keep reading your posts and all I see is you want a 100% fool proof click fraud detection and you believe its possible. Am I incorrect ? I just don't beleive thats possible.

Is it that you want just better ? If so how much better? How could you even demonstrate "better" or "good enough" ?

Spits

your words out of my mouth :).

I'm mostly disputing some cockamemee statistics concepts. Nobody expects 100%.

Demonstrating 'better' or 'good enough' in statistics is straightfoward. Ever see political polls where they say 'plus or minus 2.5%?' (It's actually 'plus or minus 2.5%, 19 times out of 20'. The other one time in 20, you expect your estimates to be further out than 2.5%) Good enough for every media outlet I've ever seen. And worth noting they get that level of accuracy with only about 1000 data points.

Thank you for clarifying

Thank you for clarifying wheel. I have to disagree though, I don't think even thats possible. There will always be someone who can identify and demonstrate a new way to game the system, then claim that everyone is doing it. These claims are nearly impossible to refute because everyone *could* be doing it. So your +/- suggestion goes out the window, as it doesn't build trust and is easily argued.

Sure

Don't take this personally, but 'i don't think that's possible' doesn't refute anything I've said. You'd 'think' that just by looking at a bunch of dollar amounts of insurance claims that there's no way to start pinpointing fraudulent claims. You'd be wrong.

Here's a non-intuitive example: Benford's Law. Who knew that in claims (or click costs) that the first digit of the click charge isn't randomly distributed between 1 and 9? There's even a course on it and how to use it to detect fraud.

Waving your hand Dogbert style and saying 'baaah' at statistics is really odd IMO. I don't know why I'm even arguing the point.

I am not arguing that

I am not arguing that statistics are not relevent Wheel, I beleive they are. I am probably not being explicit enough, my apologies.

What I am trying to say is that people who do not believe that big companies are doing the best they can now will never beleive it. I assert that if they did say +/- 2.5% that the naysayers reaction would be one of the following:

1) They are full of shit
2) No its higher, here's why....
3) Thats not good enough.

The Back door

I guess the SE's like everyone focused on the front door but take a cursory look at the Back Door, no doubt one can go into "click intent" while the jets are leaving the back door with checks. Run some algos on the back end stupid!

All you money go to China ;)

I have been saying for a

I have been saying for a long time that click fraud is a huge issue and that companies like google are lying when it comes to having a REAL MODEL to detect what they have deemed "invalid clicks". They have a financial incentive to let fraudulent clicks happen. We all know this so why is it such a freakin shock to our system?

There is a lot of corporate double talk and speculation but no PROOF that the engines even have software to detect click fraud. Want to know what they really have? It's a mash up of proxy checking and geo-ip tagging, plus a group of people looking through traffic logs for "patterns of click activity". No real science or software is implemented here. It's simply a group of guys and gals sitting in cubicles looking through the server's traffic logs. Trying to spot (by eye mind you) suspicious patterns. It's ineffective and highly subjective. Yes, guys and gals, the do indeed have people looking at the issue but the real issue is whether or not to turn a blid eye for the sake of revenue.

Click fraud can be detected. However only the ad network itself can gather the necessary data points to determine the validity of a click.

I helped Anthony Effinger with a lot of the background for the article he just wrote on click fraud. I even referred him over to the gang at ClickFacts. What I told him is the same thing I'll tell you guys. The bottom line is that advertiser's don't trust the companies they are advertising with. So why do they advetrtise? Because to a big degree the PPC model is still productive even though there are staggering losses. The loss is worth the gain. That in my book that is unacceptable. Engines need to address the click fraud issue with a set of standards and practices that the advertiser can trust and verify. Verification means if the advertiser chooses to track click fraud with their own system that they can do so and be taken seriously for it.

This whole, "we need to keep our proprietary click fraud detection processes secret so click fraudsters can't work around it" idea is nothing but a line of crap needs to get flushed. Click fraudsters are going to try and game the system whether or not the public at large knows what "methods" these engines have to detect click fraud. The only question here is can a system be built to detect the vast majority of click fraud, and identify other suspicios activity on the fly. I guarantee you that that system can be built and will be. When it is the other engines will be forced to do something about it or risk losing thousands of advertisers over it.

I suggest...

If your worried about click fraud pay less for the clicks.

Why pay less

When you should REALLY not have to pay for bad clicks at all? That's part of this whole problem in my book. PPC as an "acceptable loss". It should not be that way. The right methods can be put in place to combat it to a great degree. The engines simply don't want to do it.

right.. there is a huge

right.. there is a huge conspiracy to screw you out of your money involving hundreds if not thousands of engineers at 3 different search engines and a bunch of other companies. The upshot of it all is they really don't think that anyone will find out and would rather pay progressively larger settlements if people did.

In related news the pentagon is hiding aliens, jfk was shot by 2 gunmen, and the black helocopters over my house recently got a paint job.

Wow professor

If you truly want to believe that the engines don't know they are profiting from click fraud you're kidding yourself.

You are insinuating that

You are insinuating that thousands of people are corrupt, working together, and trying to defraud you. If it were a handful of people I might agree. Its not a handful though its a ton, with the number of people involved and the consequences of them not trying to prevent fraud I see no way there could be a conspiracy.

Further more where are the whistle blowers ? Why has no one on the inside come forward ? Surely there is at least ONE honest and upstanding person working there.

WTF, Do you think its the mob running those places ?

I will accept the arguement that they are doing a bad job as fundamentally sound, but I can not accept the idea of wide spread collusion & fraud as it is illogical.

It is a mob

that has access to all them CCs.. thats easy pickins my friend, a few hundred bucks for scripts and hosting and you can zap advertisers all day long. Google can blame others (as usual) and checks get cashed, it won't let up.

There is the concept of "fiduciary responsibility" that seems to be largely overlooked. Do you really think the computer scientist Eric Schmidt is looking out for YOU? Don't forget the "Chinese wall" between sales and geeks, the twain shall never meet.

You forget

Professor I worked for engines. I came out on this issue while I was the Senior VP over at BlowSearch after working for Kanoodle for 2 years. I openly talked about click fraud on my web log as well as a number of forums and stood my ground. I know what takes place inside the engines and how many of them deal with the click fraud issue. I'm not shouting "conspiracy" all I am saying is that the engines DO KNOW who on their network is commiting click fraud (for the most part) but they turn a blind eye to it until enough advertisers complain about a particular traffic source. It about managing revenues from the search engines perspective. Not about creating quality traffic 9x out of 10.

Professor I'm not new to this issue and I do have a great deal of insight as to how engines manage it. Remember that PPC is an incestous business. What goes on at one engine is mimicked by another because of the downlines created in the traffic stream by publishers. The publisher newtorks carrying PPC ads revolve around each other. Many of them recycle the same traffic.

The bottom line with PPC fraud in this industry (from the engine side) is not eliminating it but how to effectively manage the revenue and partners generating the fraudulent clicks. Believe what you like professor but I have first hand knowledge and I do know that standard practice in the space is never to drop a partner unless either a) you get caught with your pants down by an advertiser or b) you can replace the click revenue generated by a shady publisher with some OTHER publisher.

The engines know exactly who, what, why, when, and where the traffic comes from in most cases. They simply choose not to do anything about bad traffic until it gets to an extreme that warrants attention. Think what you like. I know what I have seen and heard as an executive in the PPC business.

proprietary click fraud detection ...

This whole, "we need to keep our proprietary click fraud detection processes secret so click fraudsters can't work around it" idea is nothing but a line of crap needs to get flushed.

Is this not a reversal of the position you took some months ago on TW about your technology? I, and several others asked about technical details and got no real answer. Does that mean it will be published now?

The only question here is can a system be built to detect the vast majority of click fraud, and identify other suspicios activity on the fly. I guarantee you that that system can be built and will be. When it is the other engines will be forced to do something about it or risk losing thousands of advertisers over it.

I thought that your claim at the time of your earlier posts was that it had already been done. Are you now saying it hasn't been done?

the thread was: New Anti-Clickfraud Tools from Blowsearch

Plum, you're right

It is a reversal. I've thought long and hard about my stance since then and I have come to the conclusion that someone needs to set an open industry standard. However I'm no longer at BlowSearch and have not been for some time. Yes, it was built but because the company fell apart over management squabbles amongst the 4 controlling partners it never got finished. So at this point what they had is pretty much dead. It was indeed real.

The processes I have knowledge of has nothing to do with thier technology. Rich Kahn, BlowSearch's COO at the time wrote it without any help from me. I simply helped to steer him in a bit of a direction on how to see the issue with my knowledge. However the system he built was a bit extreme because it actually invalidated A LOT of valid traffic. It assumed that every click was a fraudulent click.

I never told BlowSearch everything I know about click fraud or other technologies. I didn't even do that for Kanoodle. Yeah i made them both assloads of money but I kept a lot to myself. Hell I (and a few others) even have a pending patent through Kanoodle so it's not like I didn't contribute. I held back because I knew that someday the ideas and knowledge I had would be useful for my own company. At this time I'm headed in that direction slowly. All will become clear when the company is launched and the ad system is up and running. I do not want to promote my personal agenda on TW though. I appreciate the community and the open discussion involved here too much to do that. All I can say plum is that yes it will all be out there for everyone to see in due time.

OH,

And you can hold me to it.

uh oh

someone get the geeks a dictionary:

due diligence

Click Fraud Detection

I really, really, really doubt there's anything to detect click fraud 100%. It's a never-ending battle. The best the white hats can come up with is something like captcha, and guess what, given enough time and effort, you can break that, too.

Anyway, we're going to open source our algorithm for detection of fraudulent clicks (http://blog.sheergoodness.com/?cat=5) just because no one in the SE industry really cares enough about it since it's the end advertiser that pays for it.

I was wondering:

Do you guyz/gals* think that pay-for-search projects like Blingo and - more recently - Searchchips are a form of (potential!) click fraud? Cuz since I don't do ads - I might be "biased".

* because most of you are - no doubt - using ads and stuff to promote your respective sites

I just think stuff like

I just think stuff like Blingo is too gimicky and attracts the low end of the market.

>> Whatever happened to

>> Whatever happened to Boser's click fraud test?

yeah whatever happened? One month from december is months ago.. did the click bot get built or what?

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.