Advertising Technology

Whatever Did Happen to Contextual Targeting? A Chat With Twelvefold

 Admonster's Gavin Dunaway This column was written by Gavin Dunaway, U.S. Editor at AdMonsters, the global community of ad operations and technology leaders. 

“When client and agency teams look at content, they have a very good idea of where they want to run. But up to this point it’s been difficult to find those exact pages,” says Twelvefold CEO Dave Hills. “And that’s what we’re allowing them to do… It’s the ability to analyze and act.”

Before the release of version 3.0 of its real-time contextualtargeting Spectrum, I’m being treated to a demo by Hills, an ad industry veteran whose resume boast tours as the top executive at LookSmart; President of Media Solutions for 24/7 Real Media and Chief Operating Officer and President of Sales for About. The company once known as BuzzLogic switched its name to Twelvefold in 2010 and launched the first version of Spectrum in late 2011 – which is about the last time I can remember writing about contextual targeting. Even then, there wasn’t much to report.

Real-time audience targeting has been the chant of the media-buying crowd for years now, but suddenly “cookie-less” targeting is a hot ad tech marketing term. Turns out, cookies are corruptible – advertisers are increasingly awakening to the scourge of botnet traffic, where hordes of zombie computers pick up high-value cookies across legitimate sites and then are targeted on scab sites on networks and exchanges. The results are targeted impressions viewed by no one, while the site owners (and bot runners) rake in the cash a fraction of a penny at a time.

Then there’s mobile, where the app ecosystem andprivacy-focused browsers render cookies moot. Speaking of privacy, major desktop browsers are enabling machinations to at least reject third-party cookies by default.

Device IDs – OS-enabled or probabilistic – are alternatives to cookies, but what ever happened to good ol’ contextual targeting? In an age where digital audience-buying methods are increasingly compromised, it would seem there’s an opening for an innovative contextual targeting solution.

And innovation is the key word (see what I did there?), considering how much contextual targeting has lost its luster. First off, audience is the language of TV buyers, as context can often be an afterthought to demographic. In trying to win over TV budgets, digital sales needed to speak a familiar tongue.

Then there’s the question of overall effectiveness as language-learning systems never seemed to perform as well as advertised. Semantic targeting also has its fraud issues – scab sites drowning in high-value keywords (chasing keywords can be as wasteful as chasing cookies). Finally, with buyers flocking to RTB-powered exchanges, contextual systems didn’t seem to have the speed to keep up.

Twelvefold bypasses that last hurdle by being built on top of AppNexus (though it is not an AppNexus app, Hills notes) and other exchanges. Hills claims the ability to analyze 150,000 URLs a second; in total, the company has an index of 17 billion scanned URLs.

A contextual targeting solution has to “provide practical big data in an RTB/programmatic environment,” Hills argues. “Many good solutions are out there for expanding keywords, but they don’t take advantage of real-time trading. The system has to work at the impression level, has to have as much data around it as possible and has to work quick.”

A Page’s Purpose

A great deal of contextual targeting is based on systems that attempt to categorize pages based on their subject matter. TwelveFold eschews such taxonomies and categorization because of their inflexibility; instead the company organizes around content targets, which Hills compares to audience segments. In effect, each company writes a new taxonomy for each campaign – one that is easily optimizable, which is part of the point of the version 3.0 launch.

“Our algorithms tell us both what the page is about, which is price of entry, and why the page was written,” he says. “What was the mindset the author intended to create?”

Can a machine determine writer intent? Basically, is there a program out there that could have better written all my college essays on Victorian literature? Well, I doubt there’s one that could have added so much passive-aggressive disdain toward Dickens, and there isn’t a whole lot of money in the advertising around literary criticism. However, writer intent is the pathway to user intent in consumption – understanding that is a boon to direct-response targeting (Hills notes clients tend to use them for spot buys and always-on in-market).

Through calculating phrases and how words come into them, TwelveFold breaks mindsets into 12 segments like “information seeking,” “comparison seeking,” etc. Algorithms weigh term relationships to narrow the mindset. A buyer or trading desks then removes irrelevant terms (“Turkish Embassy” is not a great term for potential turkey hunters) and append blacklisted sites. In walking me through a demo for a hunting segment, Hills found 14,000 matching sites with a million possible page targets. Twelvefold’s proprietary forecasting algorithm (backed up by 18 months) estimated 230 million potential ad impressions monthly – per size.

And it gets easier – inside Version 3.0 is a self-serve platform, that includes a smart tool that populates targeted inventory with pages within TwelveFold’s index of 17 billion that match the content target segments of client-preferred URLs – in a matter of seconds, just a few URLs will generate hundreds of thousands of like-minded pages, each with a relevance score next to it.

Next, clients hit buy and the bidding begins. Twelvefold boasts a 1.5 millisecond turnaround time for making a decision on a bid – its tech sees a specific URL in an auction, compares it to all the other versions of the page it has seen going back 18 months, then updates score and bid price. For trading desks, Twelvefold backs up its systems to the unique bidding rules and gives them the page scores. Any time a URL works its way through the exchange, the Spectrum platform re-evaluates and re-scores it. Yes, that’s a lotta processing, and TwelveFold reportedly has an innovative Hadoop setup to assist in the number crunching.

Here’s where there could be some difficulty – Twelvefold requires full URL transparency, which not every publisher is willing to hand over on an open exchange. No surprise then that Hills is an advocate for programmatic transparency.

Audience and Contextual, Working Together?

I’ve long wondered when audience and contextual targeting were going to truly work together in harmony. Both have their strengths, but the weakness of buying cookies without context has been on glaring display of late. Private exchanges can offer site-level, section-level or URL-level transparency, but add technology that can actually determine page purpose and CPMs look set to rise. Accordingly Hills says Twelvefold is looking at the tech stack a few years down the road – namely how to enable client layering of first-party data

“When we take a look at the coming industry ecosystem, we get excited… The ability to do everything in real time, with real attribution against it – which pieces of data work and which don’t,” Hills comments. “We think we can supply the answer for the content side of things.”

Well, certainly one answer. With all the cookie calamity of late, it probably wouldn’t hurt advertisers and agencies to re-examine the contextual targeting scene – it seems interesting things are afoot.

Namely.com_Free_White_Paper_Performance_Reviews

Sponsor

Sponsor