Waterfalling: A New Hope

aripaparosellsiderThe Sell Sider” is a column written for the sell side of the digital media community.

Today’s column is written by Ari Paparo, CEO at Beeswax.

Why does waterfalling or “daisy chaining” exchanges produce additional yield for publishers?

I received a lot of theories following my last dispatch, which examined this mysterious behavior of publishers trying to draw out additional yield. The most compelling explanation came from my friends at Datacratic, who explained that floor prices are to blame because they are used to cut price reductions in the second-price auction.

I want to spend a little more time diving into the root causes and implications, since ultimately I think we all want a market that is more transparent, efficient and fair. We must resist the dark side.

The Sell-Side Cheats The Second-Price Auction

Obi-wan: A second bid is worth more than any floor price

If we accept that floor-price manipulation is to blame for waterfalling, we must also ask what a floor price is actually supposed to represent. Buyers and sellers tend to have dramatically different definitions and perceptions of floor prices.

A buyer thinks of a floor price as the minimum price at which sellers are willing to sell an impression. Presumably if this floor isn’t reached, sellers will keep the impression for themselves.

But that’s not the way sellers think about floor prices. Sellers think of a floor price as the minimum amount they are willing to receive from a specific auction before the impression competes with other forms of liquidity, somewhat like the “ask” price in financial markets. See the difference?

Buyers think of the sellers like Leia resisting Jabba the Hutt’s grasp, maintaining some dignity and willing to let the impression remain unsold rather than sell below a specific price.

Sellers think of themselves like denizens of Mos Eisley’s cantina, making whatever deal they have to so they can survive in the universe’s most wretched hive of scum and villainy.

To a buyer, the auction has rules, and as long as you work with the good guys, you should get good results. To a seller, each auction is another attempt to gain some yield before Han shoots first.

Second-Price Mechanics

Yoda: Only one bid the auction has, pay what price the bidder will?

If an auction only has a single bidder, what price is actually paid when we’re supposedly in a second-price auction? In most exchanges, the second price is the floor price. This sounds reasonable at first because what else would you use? But it actually gives publishers a profoundly perverse incentive. By increasing floor prices, they effectively reduce the difference between bid and price paid, thereby capturing additional revenue.

As Datacratic pointed out, this is the primary reason that publishers waterfall. If publishers use a single exchange and increase floor prices, they will lose yield from buyers offering bids that are lower but still acceptable. But by using multiple exchanges, each with a different floor price, sellers can cover more of the demand curve, not to capture demand with lower bid prices but to lower the amount each bid is reduced along the demand curve.

aripaparochart

Obi-Wan: I felt a great disturbance in the auction, as if millions of impressions suddenly cried out in terror and had only a single bid.

This discussion relies on a key assumption that auctions frequently do not have multiple bids and get reduced to the floor price. How often does that actually happen?

Amihai Ulman recently made a similar argument: The market for individual impressions is actually quite illiquid, even while the overall marketplace is awash in liquidity. He references a 2013 study showing a 50% difference between bids and prices paid for auctions. The implication: Bid reductions, which are highly subject to arbitrary manipulation by publishers, have a huge impact on clearing prices.

Perverse Incentives, Soft Floors

Darth: You underestimate the power of soft floors.

Some genius at some point thought of a solution to this whole bid-reduction problem. The conversation probably went something like this:

Publisher: What if, instead of reducing bids by a lot, we reduced them by less?

Vendor: Well, that wouldn’t really be fair. It’s a manipulation of the auction.

Publisher: It’s really just a different floor price.

Vendor: Oh, OK. Let’s do it!

Thus was born the “soft floor,” a level at which bids get reduced even in the absence of a second bid. How does this fit into the economics of second-price auctions, which are designed to give buyers incentives to bid honestly and accurately? It doesn’t. It is simply a setting that publisher-side platforms offer to allow sellers to get more money when there’s only one bid.

But wait, why stop there? If an exchange or sell-side platform is paid primarily by the seller and is not subject to any kind of audit, regulation or standard for the proper execution of the auction, how do we know what’s going on here at all?

I have seen SSP contracts that set a specific auction deduction (say 10%), then add on an additional “RTB fee” that is unspecified and at the discretion of the vendor. Normally, no rational businessperson would sign a contract like that, but when you can’t determine how much money you’re making anyway, why sweat the calculation at all?

Trust The Algorithms

Supply-side platform: I love you.

Demand-side platform: I know.

The evidence accumulated thus far suggests a straightforward solution to our problem: We, as an industry, should move to a first-price auction. The argument is simple: Second-price auctions work best when you have bid density, but in RTB, most auctions only have a single bid. In that environment, there is enormous opportunity for shenanigans.

When I’ve pressure-tested this proposed solution with thoughtful members of the community, they’ve made the counter-argument that sophisticated buyers know this is all happening and the algorithms adjust automatically. Thus, if a publisher is “cheating” in some way to artificially increase the price paid, the buy-side system will see that placement as less desirable and will, over time, bid less. Thus, the bid reduction is irrelevant since both sides are gaming it all the time.

I don’t buy that. First, implicit this argument is the assumption that it’s OK to screw over less sophisticated bidders as long as the cutting-edge algorithms can adapt. Second, the auction mechanics create real externalities for the ecosystem, including latency, an increased likelihood of fraud and data leakage from daisy chaining, complexity and opacity. Third, it isn’t just the buyers that are getting screwed here – the second-price reduction is a murky, abuse-prone part of the system that may hurt publishers as well. Finally, the complexity in pricing makes both buyers and sellers less empowered to make smart business decisions about things like pricing and channel conflict.

Another related argument in favor of second-price auctions: They remain useful for audience buys and retargeting since those buyers are willing to pay far higher prices. This argument also holds little water since it translates to “the unsophisticated buyers don’t care that they’re being ripped off.”

Why, exactly, wouldn’t a bidder prefer clear guidelines for what they will pay over the convenience of just bidding high all the time?

Wait For The Sequel

We started with talking about waterfalls, and now I’m basically saying everything is wrong and we need to start over. I’m not sure the situation is really that dire.

It’s time to develop some guidelines for how auctions should be conducted, a task that belongs on the plate of the IAB’s Quality Assurance Group. The programmatic business developed organically on the model of search, but it’s now time to straighten out some of the mess.

Buyers should ask their supply partners about first-price auctions, how floors are set, and whether supply is being waterfalled.

Sellers, ask your exchange about switching to first-price auctions and don’t use soft floors. Ask your supply-side platforms about fees they take on the buy side and squeeze out the hidden stuff.

Use the Force!

Follow Ari Paparo (@aripap), Beeswax (@BeeswaxIO) and AdExchanger (@adexchanger) on Twitter.


Popular On AdExchanger Right Now:

7 Comments

  1. Another insightful article on what's really happening in the auction world, Ari. Moving to a first price auction is an interesting idea, and one I've certainly explored with my own SSP partners before when I was working on the publisher side. What they've usually said to me though is that one, their contracts prohibit altering the auction methodology, and two, even if they could change it, unless the entire industry switched a seller using a first price auction would be penalized as DSP algorithms optimized away from their inventory. I wonder what your thoughts are there.

    I'd also suggest publishers take your advice at the end of the article - "Ask your supply-side platforms about fees they take on the buy side and squeeze out the hidden stuff." Those mysterious RTB fees buried in the contractual language are much larger than I think most realize, if to your point, they even read their contract close enough to realize they're present at all.

    Reply
  2. if in practice people are splitting their budget via bandit-like techniques and the daily budget for a campaign is fixed beforehand... setting a soft-floor or a hard-floor will not change anything... you always will end up buying X% from seller A and Y% from seller B... if they set whatsoever floor, that doesn't change anything... if they set floors they are only reducing the number of chances for you to observe conversions on their inventory and naturally you will end up spending less on them... on that view you can say that setting floor prices (even if it looks like advantaging for the seller, it is not), they will only win more money here and there when you are really targeting some given user but on general line, when you are spending a fixed budget just to get some reach, setting floors will make very little difference... the flaw in the mainstream argument is that people are buying a fixed quantity inventory and that's not reality... reality is that people are spening a fixed budget in advertising; the floor discussion is mostly paranoia

    Reply
  3. Ranjeet Tate

    Second price auctions seem amenable to gaming, and both buyers and sellers are probably gaming it. As a buyer, it is also very hard to answer the question: "How much do we value this impression?" Is it the bid CPM (is that really how much we were willing to pay for it or is it a fictitious value we assigned it in order to win it), is it the cost CPM (which is what someone else - the competition or the seller - valued it at)?

    Reply
  4. I feel a disturbance in the Force Ari.

    Discrepancies.

    Ben's point is strong. But ALSO not only should publishers look at fees, what about discrepancies? Each buyer adds loss, which publishers should be calculating into the inventory they send their buyers.

    It would be interesting to see what the difference is between the bids at the time of the auction, vs the ACTUAL revenue displayed to the publisher in their reporting UI at the time of their billing.

    All stats change as cookies match impression logs, or view ability audits are made of course, but discrepancies are underestimated.

    10-20% of this $120b industry is lost in discrepancies?

    Reply
  5. Two things.

    First, how can you write a lengthy piece with multiple Star Wars references and spell Princess Leia as LEAH. Travesty.

    Second, I've read your initial hypotheses about why waterfalls exist and your explanation through this and I think your conclusion falls short. Going to a first price auction model is deeply flawed for a publisher, somewhat positive for SSP and incredibly advantageous for buyers. You're basically advocating for front running. Unless that's what you're arguing for in-between the lines, first price would be as helpful to publishers as Jar Jar Binks was to our childhood dreams.

    To be clear, I think you've been very thoughtful and I don't want to troll up. Second price auction dynamics always maximize sellers' value but as you've stated, it doesn't solve the waterfall situation because it's a totally different problem. If you look at SSPs, they'll all bitch about being at the back of the waterfall but no one ever complains about being at the very, very top. This is part fad but it's part of the floor you've mentioned, along with the artificial floor created by sales teams, kickbacks and manufactured value created by the top waterfall to deter waterfalls to be broken down.

    Reply

Add a comment

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>