Who doesn’t love a targeted ad? Users prefer them, and content networks and supply-side advertising players are okay with them, too – that is, until they get a little too narrowly targeted. Why? While CPMs go up, the inventory volume is so low that less money gets made. On top of that, smarter targeting eats more CPU cycles in the data center, so it’s costing more money to make less.

That’s the picture Tim Musgrove, chief scientist at Federated Media Publishing, laid out for Semantic Tech & Business conference attendees at a session during the event yesterday. Then he explained how Federated Media’s Conversation Targeting semantic-powered ad targeting platform has changed things.

“It’s different for Federated Media because we have an entirely different way to approach this that avoids this decline of revenue that you see,” he said. That different way includes the tech of Conversation Targeting, of course, but also the business side of advertising negotiations.

Talking About Conversation Targeting

Conversation Targeting, which Federated Media debuted last year (see this story), crawls its network of premier, independent bloggers, using a bottom-up semantic approach to discover, index and analyze the raw, unstructured prose in content comments and discussions. It bubbles to the top the topics the community of a site tells it the conversations taking place there are about, and determines how topics are related to other topics within these communities of conversation.

Federated Media’s approach is key to realizing the one case “when narrow targeting doesn’t reduce inventory but expands the inventory they purchase,” Musgrove said. That one case is “when it makes the advertiser comfortable moving outside their usual comfort zone.” For instance, the instincts of a tech vendor with a cloud computing campaign would be to target tech and maybe finance content. But with Conversation Targeting, Federated Media can show them that that approach might lead to missed opportunities – that, for example, cloud computing conversations bubbled up in sports blogs, too, thanks to discussions about how a major online sports network turned to the cloud so that its web site wouldn”t go down again during the World Cup.

“Why not let us target the cloud computing conversation no matter where it is,” Musgrove said. “That ends up getting advertisers to be happy about buying more inventory from us.”

It’s important to focus on quality content in your networks rather than trying to sell all your inventory to have success with such an approach, Musgrove noted. Results, he said, across hundreds of campaigns to date are “overall almost 50 percent better click-through rate on Conversation Targeting than what our company usually does on broad targeting.” And nearly every client that has gone this route with Federated Media, he added, has re-upped and renewed.

Conversation Targeting, he pointed out, is dynamic and reacts well to change, too. Google Plus, for instance, isn’t currently appearing as a topic in communities where SMB conversationslive, “but if that changes in our system, we will automatically detect that and bubble it up and add it to SMB.”

More Insight Into Conversations On The Way

Other changes on the way are enhancements to Conversation Targeting that work on other metadata characteristics for detecting ideology, optimistic or pessimistic tone, cultural dimension and the seriousness factor of a site, as well as what reading level it is written on.

With one site, it already has a year of experience proving on a fully automated basis that it can separate pessimistic from optimistic news stories, Musgrove told The Semantic Web Blog after his presentation. “It’s the same technology underlying that that we use to separate conservative form liberal and other attitudinal distinctions in text,” he said. The roadmap for the second half of 2012 focuses on internal testing with live production for 2013.

Its semantic detection of reading grade levels is more oriented to other conversations it encourages advertisers to engage in, such as sponsored blog posts. “We know amplification is affected by reading level,” he says. Some social networks respond better to amplifying content written on a lower-grade level and some to content written on a higher-grade level. “We have measured this and we have unmistakable and unquestionable correlations,” he says, noting that more details will be forthcoming next month.

The Business Side of Semantics

When it comes to the business side of making targeting work to the content network’s benefit, it’s time to apply the same model almost every other industry does when it’s got a limited number of valuable items that everyone wants: Don’t let buyers cherry pick the most narrowly targeted stuff and buy only that. “So we are adding a semantic layer of super-exact, smart, targeted stuff, but on the business side you have to sell that as icing on the cake,” he said. “This is the sweetener that entices them in.”

In The Semantic Web Blog’s follow-up conversation with Musgrove after his presentation, he discussed how selling such structured packages is a new way of thinking for online sales teams, which have been put through a feverish pace of new Internet sales techniques and technology over the last few years. At the same time, he’s seen that their involvement in understanding the impact of semantics on their sales has been restricted in his past experience at another company, where refined semantic targeting was turned on only as the rescue firehose when results were overpromised and traditional methods were under-delivering — without the salesperson even knowing it at all.

At Federated Media, it’s a different environment, where its senior vp of sales understands exactly what semantics offers, how to get clients’ attention on the opportunity, and how to disseminate that to the rest of the team so they can directly articulate it to prospects, he says.

“It’s like a big breath of fresh air to me because salespeople are learning what it is and presenting it with PowerPoint slides, and they are getting Levis and L’Oreal and Amex to understand this is semantics, and they even use the word…. They understand they are buying a special thing and the advertisers trust that no human being is manually editing which pages this goes on, that this actually is an AI tool.”