Scenario

Vicky, a vegetarian blogger, is a very controversial figure who is only avidly read by a small community of professionals and activists. Occasionally, though, she publishes a tasty recipe that gets widely syndicated around the Web.

Ralph, a cook, only blogs about interesting recipes. He has a much broader audience because he test-cooks and reviews recipes of all sorts, from all kinds of diets.

Lou, a nutritionist, primarily advocates low-carb diets and only occasionally considers vegetarian recipies, and then mainly to mock them.

Tony’s Tomatoes, an advertiser, wants to entice readers to his site. He’s willing to pay a commission on his sales. To approximate that effect, he’s willing to pay (at most) P cents per click-through.

PotatoHead, another advertiser, has an informational site advocating the health-benefits of carbohydrates in tuber form. It’s an ‘astroturf’ blog run by a lobbying group, Potato Farmers Underground. This organization is willing to pay-per-impression to promote their brand.

Keyword Advertising

Keyword-driven “Sponsored Search Auctions” are the norm today. These permit centralized ad networks to dynamically insert ads alongside publishers’ contents (or, as one satirical site would have it, replace the content entirely).

The primary (publicly-acknowledged) criteria for insertion are:

1. Keyword matching against previously-crawled & indexed text of the page. (n.b. the ad keywords need not occur in the text of the ad);
2. Click-Through Rate (CTR), as a proxy for how effective/helpful the ad copy is; and
3. Prices, typically according to a second-price sealed-bid auction.

Obviously, this does not account for taste: the model does not treat advertising as an endorsement by the content-publisher. While in traditional media, this sort of “church/state separation” between Editors and Publishers is a hallowed goal, it can be counterproductive when taken to extremes. Ultimately, editorial content cannot be separated from advertising content.

Today, systems include black-list and white-list systems for publishers to limit access to acceptable advertisers, but these mechanisms are quite clumsy. They fundamental “flaw” is that sponsored search auctions seek only to maximize revenue, not relevance.

Some Problems with Sponsored Search Auctions (SSA)

Ralph is pretty much the only happy player: he panders to the broadest possible audience and gets paid roughly in proportion to traffic he aggregates.

1. PotatoHead is out of luck. Their boring brand advertising doesn’t get clicked-through on very much, so they have to set a cost-per-click-through that’s dramatically high to compensate for that.
2. Lou is upset. Unfortunately, setting such a high price for potato ads means that even the most marginal attack on tubers from Lou’s articles will summon up a “Tasty Spuds!” offer to his readers.
3. Vicki is paid too little. She doesn’t make as much money as she ‘deserves’ to: her recipes get read far and wide, but when her tomato-and-potato-pie shows up on Ralph’s site, she gets nothing.
4. Tony’s Tomatoes pays too much. When tomato ads show up on Lou’s site, his carnivorous readers just click through for fun — they know they’re running up Tony’s bills without any intention of buying.

Some Opportunities for Sell-Side Ads (SSA)

This scenario is contrived to highight a few underlying issues: clickfraud, keyword mismatch, transactional vs. brand advertising…but most of all, the lack of informed consent between publishers and advertisers.

Now, the virtue of Web Advertising today is its scalability; it’s absurd to expect an independent publisher to review millions of ads, nor for an independent advertiser to select from millions of publishers. However, there are other techniques for harnessing the ‘wisdom of crowds’ — a workable decentralized solution suggests sharing reputation and tags.

Even before tackling scalability, though, we should consider the “how society works” part of our motto to “build software that works the way society works.” In real life, Vicki is a very important person and ought to be well-compensated for her effort. That’s because many readers like Ralph are so moved by her work that it informs their own intellectual efforts: Vicki has influence.

Transferring Power to Sellers

Placement. In an ad auction, the process of matching ads to content is largely automated. The choice of which ad will be placed is only partially determined by a publisher’s choice of keywords (itself a weak proxy for intent). To move the power to the player that’s selling ad space, publishers ought to directly control which advertisements are placed on their sites.

Pricing. In an ad auction, the actual clearing price of a given ad is determined by the competitive actions of other advertisers. By calculating payoffs using a fixed limit and dividing by the ultimate number of hits, the price of an ad is determined by the competitive actions of other publishers. This is reminiscent of pari-mutuel betting (in its strengths and weaknesses).

The problem that emerges from these two power shifts, however, is that such a scheme is completely impractical at scale.

Instead, what if we could decentralize the task of checking which ads are reputable for each publisher — what if, rather than wading through the vast universe of possible ads for ingredients in all of Ralph’s recipes, he could just wait for Vicki to sort them out and recommend just the right ad to run when he syndicates her recipe?… and of course, pay Vicki for her trouble, say 20% of his revenue?

Our Demo Storyline

• Tony’s Tomatoes would like to limit their financial exposure to a fixed amount over a fixed period of time. This implies the price depends on the supply of competing publishers’ ad space (which is by no means a homogenous supply curve — each publisher has different opportunity costs of runing other ad campaigns).
• Tony offers Vicki first crack at running this ad. If she agrees, she adds the ad to her feed.
• Ralph quotes Vicki’s blog and also decides he can sell more tomatoes, too, and copies the ad to his site as well. Now, he’s getting a share of the ultimate payout for the ad — but Vicki gets credit, too. The referral ‘tax’ encourages Vicki to get others to run the ad for her.
• Lou, who does not want to endorse vegetables, can simply run veal ads instead.
• PotatoHead, who wants brand advertising to a broad audience, can directly approach Ralph (or vice versa) and propose an effective-CPM price.

Obvious flaws? [Note: this is not strictly true with a fixed budget: a dishonest Vicki would run the ad precisely once and hide it from the world… Vicki needs to be in competition with other “roots,” too]

What the server does

Unlike content that gets cut-and-pasted and spindled-and-mutilated, an ad must be served up identically each time. Thus, an advertiser’s first preference would be to run their ‘own’ ad server. The server also has to provide a browseable catalog of current ad campaigns to join. This is an admittedly centralized component, and could only be eliminated by a widely-trusted peer-to-peer web hosting service with hit tracking.

Our server needs to credit each ‘hit’ for serving an ad (and potentially, for each click-through, too) to the proper publisher’s account. It also needs to create new “downline” ads when new publishers join the network to trace their provenance.

Finally, the server needs to implement a payoff algorithm(s) to allocate the pot amongst the participating publishers. At a 0% referral fee, this becomes a cost-per-click/hit model; at 100% you have Ross Mayfield’s extreme CPI model where only bloggers that get quoted get paid. In between, we look forward to experimenting with varying ratios; another proposal of Ross’s is to use 1/PageRank as the rate…