FEATURED

GREG HARRISON, MPIRE CO-FOUNDER AND CTO

Greg Harrison is the co-founder and CTO at Mpire, which created Adxpose: “the first and only transparent, page-level campaign verification and optimization technology in the online display advertising market.” Greg overlooks the network and the 10-15 terabytes of data which is processed each and every month.

Welcome Greg. You must be delighted that someone from the real British ‘em-pire’ has heard of you - spelling errors forgiven in the company name. And we’re grateful you’ve joined us here for some brain-picking on policing fraudulent traffic and preventing “black hat” techniques which is the ‘day to day’ running of Adxpose.

Goallover: So Greg, can you tell me about what Mpire used to do, how and why you ended up creating AdXpose?

Greg Harrison:Sure, Mpire operated a comScore-ranked top 30 ad network from 2007 until 2009. In the course of building that business, we developed a lot of technology regarding yield management, semantic evaluation and categorization, creative optimization, fraud detection and prevention, and a slew of other advanced reporting capabilities.

In late 2008 we updated the publisher ad code with the ability to identify page placement and user engagement events, among other instrumentation. This allowed us to tune our ad delivery methodology, including such things as only showing the ad content when the ad unit was within a viewable area of the browser and measuring when and what type of user engagement (read: specific types of mouse interactions) has occurred (for ad creative feedback), and more. This data is also used to enforce site/domain-level traffic control on approved sites only. We used this information to communicate with our publisher base regarding page placement, rotation and overall engagement.

In early 2009, we began to experiment in “off network” trafficking to specific partner networks and with exchanges. Within those experiments, we began to suspect fraudulent traffic, and subsequently enabled our container technology on those campaigns, where we were quickly able to identify sources of fraud. Most of the fraud was not occurring with the direct networks, but rather their respective down-stream traffic partners. When we communicated these issues to our direct partners, they were intrigued and ultimately became some of the first beta clients for what ultimately became the commercial offering now called AdXpose.

Goallover: That’s interesting, so are you telling us that you have been the “buster” more than the “busted”?

Greg Harrison: Yes, the AdXpose technology definitely helps ad networks police their own publishers, but it’s not just about “busting” bad actors. By far the most effective use of the technology has been in the identification of where the good and bad traffic resides throughout your campaign, and empowering optimization around that. On the topic of policing traffic, however, one of the first tests we ran was captured in the Analyst report and September 2009 White Paper entitled “Identifying and Combating Fraud to Optimize Ad Network Buys“.  We have seen as much as 95% fraudulent traffic on RON and Category-based exchange buys, which was a huge eye-opener.  There is an obvious systemic issue with how exchanges optimize for specific types of campaigns that, left unchecked, leads to an inordinate percentage of fraud traffic.  Most of the corrective actions we and our clients have taken has been to update I/O’s to ban specific domains and specific downstream networks.  We have developed a passion for identifying these fraudulent traffic sources – I actually have a “Most Wanted” list in my office of known notorious offenders that we hope to put out of business or at least assist in their efforts to deliver legitimate traffic.  However, once the bad traffic is eliminated and the good traffic is able to proverbially “shine through”, there is an extended process of evaluating the ongoing traffic for InView rates (a measurement of how well your traffic dollars are being spent) and performance optimizations – allocating dollars against the best placements is a win-win for advertisers and networks alike.

Goallover: Ok, so as a business, MPire has migrated away from its original model in which the likes of Kelkoo were the client, and if I get the gist of the above, you were managing the equivalent of a syndicated Overture search feed but in the field of comparison shopping. Now venture backed and remodeled, who are your target customers?

Yes, we have officially “sunsetted” the old feed syndication business.

We are now entirely focused on serving agencies, DSP’s, exchanges, networks, optimizers and ad servers with thousands of terabytes of digital advertising “event data”, empowering each of these customers to make better decisions about brand safety, performance, real time bidding and yield optimization. We work with everyone from Mediacom and MEC in the UK, to Zenith and Razorfish in the US, to global video ad networks, audience measurement firms, and publisher side yield optimizers. Anyone who relies on comprehensive structured ad data to optimize media should be using our data and our services.

Goallover: Let’s get into some technical specifics. You’re using code to determine the URL, the daisy-chain, how far down the page an ad appears. Can you expand on that for our readers?

Greg Harrison: Sure. First thing to be clear about is the model by which we instrument ads.  We apply proprietary JavaScript code to banner tags, which are then trafficked in the advertiser’s/agency’s respective ad serving platform.  As such, the client continues to use the same platforms and workflow for campaign management (delivery, retargeting, conversion tracking, etc.) that they use today.  We are providing additional reporting capabilities that are bound to your existing traffic reporting as a basis for ongoing campaign optimizations.

Now to the tech.  Although I am not going to give away the farm here, I will say that key to our model is the fact that we are NOT dropping a pixel or beacon for later “Spidering” techniques.  Based on how we evolved the ad container model for our publisher solutions, we ride alongside the banner as it is being requested and ultimately rendered on the site page. In this way, we can report on placement, engagement and detailed timelines of all site visitor-banner interactions, regardless of where the ad appears on the page and how the ad was ultimately delivered.  We report event data asynchronously to our servers in real-time as the banner is requested, rendered, scrolled/paged, engaged, etc.  We do this in such a way that it does not affect page nor banner render time.  In fact, all AdXpose-instrumented banners will continue to render even if the AdXpose global server footprint were to go offline – we are not critical path on ad delivery.  We also have techniques for evaluating iframe-based delivery chaining.

When it comes to delivery path, yes, we have developed techniques for evaluating the delivery “chain” that led to the delivery of a specific ad impression.  Whether it be directly to a publisher site or via a multi-layered (nested) path that includes multiple networks, we provide you the data needed to isolate the source of the domain traffic.  This allows the client to “surgically” alter their campaign trafficking to either eliminate (bad) traffic or to focus more (good) traffic on out-performing networks and sites.  We have seen clients save entire contracts by being able to pro-actively isolate and remove traffic before it becomes a problem.

Sure, there are always limits to technology, and although we have made some breakthroughs in breaking down some of these traditional barriers, there are two specific areas that should be addressed.  The first is the rampant use of iFraming.  By today’s standards and practices, there is really no reason for networks to generate iframe tags for distribution to publishers and other networks.  Not only does it limit the ability of contextual ad serving, but it one of the most common methods for obscuring (or attempting to obscure) fraudulent traffic.  Although AdXpose can extract usable information regardless of the level/depth of iFraming, if the banner is delivered via 3 or more levels of daisy-chaining via iframes, there are limits on the amount of detail we can provide. However, in these cases, there are still clear indicators of the validity of the traffic sufficient to warrant campaign trafficking changes.

The other limitation is simply a human one. That of wearing a “white hat” or a “black hat”.  There is no gray hat.  Either you are knowingly trafficking to fraudulent traffic sources or you are not.  As much as the technology can identify and isolate the techniques employed to mix good traffic with bad traffic, it still boils down to the traffickers themselves to take the appropriate actions to mitigate their risk and establish a process of eliminating the bad tools from their respective toolboxes.  We have seen first-hand where it is occurring and the simple measure that can be taken to dramatically reduce it (fraud), but it still comes down to better controls.

Goallover: What is a CTO’s response - albeit undoubtedly aided by your commercial team - to questions about where technology should end and good “buying terms” begin?

Greg Harrison: The reporting data that AdXpose provides is something I keep referring to as “Actionable campaign trafficking intelligence”.   As such, the technology only achieves true value when applied to active and future trafficking. As the client continues to gather knowledge regarding outposts of avoidable traffic, they should incorporate these learnings into all future I/O’s.   For example, for any campaigns we run involving third-party networks, we have very specific ban lists of domains and networks.  The sooner we incorporate the specific ban items in the I/O terms, the better.  This and the insistence that iframing is prohibited.

Goallover: Do you think there is more that can be done by leaders in adserving - DART and Atlas - and leaders in Exchanges - RightMedia - to police other aspects of the business?

Greg Harrison: Yes, I think that broad adoption by the main platforms of AdXpose would be a good first step! We are actually integrated natively within most of the top ten global ad serving platforms already today, and are working towards broad activation within those platforms. Not only would transparency quickly become a de-facto trafficking practice, but I think we would see a longer term benefit in bringing more offline advertisers to online. OK, so beyond the obvious plug for Mpire and AdXpose, I think that regardless of the enabling technology, the platforms and exchanges should do everything they can to promote transparency. The network effect of cleaner traffic and less hesitancy on the behalf of brand and performance advertisers to grow their budgets will cover the nearer-term impact of biting the proverbial bullet in removing the fraudulent element of most networks that make up considerable margin. For exchanges specifically, global banning of known fraudulent traffic (domains and networks trafficking in these domains) should become a priority – they are already facing a huge image problem, here is an opportunity to make hay out of the domains and networks they are kicking out. Right Media is certainly taking strong steps here, and we will be announcing a partnership with another large exchange soon.

Goallover: Have we learned anything from the Exxon Valdez model. Let me elaborate. There are apparently some 3000 oil tankers chugging around the planet, and every now and then they break apart. But, due to “flags of convenience” and other layers, it is often very difficult to peel away the onion and find out who really  owns the broken tanker.  Do you think it’s the same in online, and have you noticed any correlation between the incidence of  fraud and the incidence of “hidden identities” in the whois  information regarding site ownership? I guess my question is - are you telling me that  domains owned by known entities are  perpetrating fraud and nobody is  taking action against the owners?

Greg Harrison: Touchy area.  Although we have not gone so far as to research the domain owners of known fraudulent domains and other sources of botted traffic, what I can say is that I am shocked to continue to see the same networks actively/knowingly trafficking in these fraud domains.  In October alone, I saw no less than 10 different campaigns (i.e. different clients) trafficking to networks and exchanges where all of the campaigns had some level of intersection regarding fraud.  This intersection took place on the same domains and ad networks.  These are the same ad networks that have been repeatedly called out for fraudulent traffic.   For these (in some cases well known) networks, it seems that the default trafficking treatment is to include somewhere between 30% to 90% or more fraudulent traffic, unless and/or until specific bans are dictated by the client.  Bottom line, the answer is that there are networks and exchanges in a position to enforce global ban rules, but some choose not to.

Goallover: What is your response, or business intention, to overcome the suggestion that anti-fraud technology will fall foul of the law of diminishing returns? In other words, if you and your  counterparts at  AdSafe and DoubleVerify and ClickFacts et al. are  successful, then  presumably the 9 billion dollar fraud problem reduces to a 9 million dollar fraud problem. What say you, good  fellow?

Greg Harrison: If we were able to reduce fraud to a fraction of what it is today, then we would see (and all benefit from) massive increases in online spend and transitions of more budget from offline to online.  As more money moves to online, so would the efforts of fraudsters.  Another way of saying that security cameras can reduce, but do not eliminate crime - they simply make it easier to identify the perpetrators.

To the extent that it reduces the need for AdXpose to provide reporting data as a basis for policing efforts, I think that would be great, because it would allow the client to focus on the other facets of actionable campaign intelligence data AdXpose provides, such and placement and performance transparency for optimizing trafficking across good traffic.

Goallover: OK, so what are the next steps? Where are you at, and where do you intend to be in 12 months?
Greg Harrison: We currently provide all standard verification services, including content and keyword based ad serving prevention, as well as a host of engagement and performance data services, to more than 50 clients. Our next steps are going to be focused on additional channels such as video, and we’re really excited about starting to tie the engagement data to actual performance (brand lift and conversions) downstream – we think 2011 will be the year of performance attribution upstream from search, and we are on the leading edge of that trend.

© 2010 GOALLOVER.