« Ad stramineum hominem | Home | The garden of cosmic horror an... »

Project Wonderful, 2006-2018

Thu 14 Jun 2018 by mskala Tags used: ,

Project Wonderful, a banner-ad network popular among Web comics authors, suddenly announced a few days ago that they would be shutting down. I've used them off and on almost from the very beginning of the system in 2006, at that time to bring in traffic for my own Web comic and later for some other projects, both buying and selling ads. There was a Project Wonderful ad box on this very Web site for several years, now replaced by a permanent ad for North Coast Synthesis Ltd.

My account number on Project Wonderful is 356, out of what looks like it will be a final count of about 120,000 accounts. I've already killed all my own publisher-side boxes as part of winding up my involvement, and maybe others have as well, but looking at the list of remaining ad boxes it looks like there are only 15 remaining publishers senior to me.

Although the timing is sudden, I can't say I am surprised that they eventually found it not commercially viable to go on. I've written about the economic issues with Project Wonderful before; my writings on that never really got the attention they deserved; and this being possibly the last chance, I'd like to comment once more.

Truthful Auctions

Project Wonderful's ad auction was not a truthful auction. I don't mean it was run dishonestly, like Google's; I mean it did not have a specific mathematical property that in auction theory is called "truthfulness." And lacking this property had a huge impact on the economics of using Project Wonderful. I don't know that this is what killed them - they blame instead the death of personal Web sites, another topic that interests me - but I think if they'd fixed the truthfulness problem many years ago when I discussed it at some length with Ryan North the founder, they could have done much better.

The core of the problem is that when I'm buying something, I usually want to pay as little as possible. There is a maximum amount I'm willing to pay, but I would really like to pay less than that. Sellers, similarly, may have a minimum they're willing to accept, but would prefer to receive more. So if you ask me, "How much will you pay?", I truthfully name my maximum price, and you say, "Okay, deal!" then I'm left thinking that you would probably have accepted a lower price after all. Among deals we could both accept, and assuming my maximum is at least equal to your minimum (so that there does exist some deal we could both accept), settling at my maximum price is the best possible deal for you and the worst possible for me. I would be better off not telling you the truth immediately about my maximum price; instead, I maybe should make a low offer first, and then try to negotiate without revealing my maximum, or maybe I should encourage you to be the first to name a price.

A truthful auction is one structured so that the best game-theoretic strategy for buyers is to tell the truth about their maximum prices. It removes the incentive for lying. The classic example, usually called a "Vickrey auction" (although he also studied several other types) is where the seller is selling, one time, one indivisible good, buyers all commit to secret bids, and then the sale is made to the highest bidder at the price of the second-highest bid. If Alice, Bob, and Carol bid $12, $15, and $10 respectively, then Bob wins the auction and the good is sold to him for $12. In case of a tie for highest bid, the tied highest value is considered to also be the second-highest, and some rule not specified here is needed to choose who wins, but the details of that rule are not important to the truthfulness property.

It is important here that given Bob wins the auction, bidding higher would not cause him to pay more. If he bid $20 instead of $15, and the other bidding did not change, he would still pay only $12. Bob shouldn't bid less than the maximum price at which the deal would be acceptable to him (or he could miss out on making a deal that is still better than no deal); he shouldn't bid more than his own maximum price (or he could be stuck with a deal he considers unacceptable); but he is assured that he gets the lowest price the seller could reasonably accept, given making a deal with Bob at all (because at any lower price, Alice would be making a better offer than his own).

Project Wonderful's auction system, at least through most of its history, worked almost like that except not quite. There was a difference which seems minor but was actually catastrophic, throwing away the truthfulness property and forcing the whole system into a race to the bottom. In Project Wonderful, advertisers (buyers) bid on a given advertising opportunity. Bids were quantized: whole US cents per day from zero up to $0.10/day, then in $0.10/day increments. (There were special limitations on zero bids, but they aren't relevant to the truthfulness property.) The sale was awarded to whoever made the highest bid. Ties, which frequently occurred in practice because of the bid quantization, were broken by timing: if two or more advertisers made bids at the same price, the one who bid first won. But the sale price was not the second-highest bid value. In the case of a tie, the winner paid their own bid value (which was equal to the second-highest bid also, because of the tie), but if the highest bid was strictly more than the second-highest, then the highest bidder paid the second-highest bid plus one bid increment.

If Alice bids $0.20/day, first, and Bob bids $0.10/day, then Alice wins and pays $0.20/day. But she would have been better off by a factor of two by bidding only $0.10/day; then she would still win under tiebreaking, and would only pay $0.10/day. If Alice can make reasonable guesses about Bob's bidding, she is better off by a factor of two not to honestly bid her maximum.

And it was even worse because Project Wonderful re-evaluated the auctions continuously in real time and bid amounts could be changed without losing priority. Alice could start out bidding $0.20/day, see that she was awarded the win at full price, then try reducing her bid to $0.10/day. If it didn't work, she could always bounce back up to $0.20/day. Alice could easily extract from the system the information she needed to optimally make untruthful bids. So, under reasonable assumptions, she would.

Ryan's point of view last time I talked to him about this, years ago, seemed to be that the difference between $0.20/day and $0.10/day was only ten cents, and people won't put a lot of effort into chasing that. (All the more so at the $0.01/day to $0.02/day boundary, which may actually have been an even more important gap in practice.) He also thought it was important that as often as possible, losing bidders should be able to see that someone else was willing to pay more than their own bid, not just the same amount winning on tiebreaking; this reduced the need to convince losing bidders that the tiebreaking rule was implemented fairly. My own point of view was that a difference of one bid increment wasn't ten cents or one cent, it was (in the worst, and frequently occurring, cases) a factor of two; and people whose actions matter absolutely will put effort into getting a factor-of-two improvement in prices.

I think my prediction was borne out by the pathetically low prices at which ads were actually sold throughout Project Wonderful's history, compared to the prices that prevailed on other advertising systems. Project Wonderful ads routinely ran at something like US$0.10 per thousand impressions, where real, professional advertising with comparable placement on sites of equivalent quality might cost the buyer US$1 or even US$10 per thousand impressions. (The seller would not actually get any of these prices; all the networks take large cuts between what the buyer pays and what the seller receives.) There was a factor of ten to one hundred between prices on Project Wonderful and on other systems. Enough people were making untruthful bids on Project Wonderful (less than they would be willing to pay, just enough to win the auctions by tiebreaking, and modified in real time to stay on top) that it had a depressing effect on overall prices. Then low prices partly caused by the untruthful auctions, supported feedback loops through other effects which further kept prices low.

I should add that it's possible the rules I describe here are (or were, on shutdown announcement day) no longer in force. They certainly were in force for most of Project Wonderful's history, but I tried some test cases with the actual system (which is still running, in a phasing-out mode) just now and I saw bizarre results I couldn't explain. In particular, I saw a case where increasing a bid that was already winning apparently caused it to start losing! I was expecting to still win but be charged more; the observed behaviour is inexplicable without a serious bug in the software, new priority rules I don't know about that heavily penalize realtime bid changes, or someone placing a new higher bid just within the few seconds I was busy changing mine. There was another case where increasing a bid that had been winning at $0.02/day apparently caused it to start winning at zero (coincidental cancellation by another bidder?). I couldn't reproduce a definitive example of being charged more for increasing a winning bid while still winning with no other bids changing; but my bids at $0.10/day are now being outbid at what is reported as $0.20/day, so at least some kind of plus-one-increment rule still exists. Best case would be that the current rules are you pay exactly the lowest price at which you could bid and win, taking into account tiebreaking, even if your actual bid is higher; I think that would restore the truthfulness property, although the effect of the realtime recalculation is hard to analyse. Such a rule is what they ought to have had right from the start, and if it's in force now it could explain everything I saw in my current tests if I also assume there were some bids placed or changed by others at coincidental times while I was testing. So... maybe Project Wonderful did fix the untruthful auction issue at some point in the past without my knowing about it. If so, good for them; surprising they didn't publicize the change; and too bad it evidently didn't save them.

The Quantity Premium

Early in the history of Project Wonderful I wrote about an effect I thought I'd observed that bigger sites were getting paid more per unit of advertising: I might get paid $0.01/day for an ad on my 1,000 hit/day site while someone else with a 100,000 hit/day site would get $5/day. I was roundly flamed by idiots who didn't understand that I understood the concept of a per unit price, and they scornfully explained that concept to me while completely missing the point! I never said it was surprising for the site with 100 times as many visitors to get paid 100 times as much money... but the effect I wrote about was that the large site would get even more than that by a significant factor. It was a superlinear relationship between size and price where at most a linear relationship would seem to make sense.

I am no longer convinced that that effect was real, or if it was real when I thought I saw it, that it remained real past an initial period of price irrationality in the first couple of years of Project Wonderful's operation. In later years, as I collected more data and got a better understanding of how to account for differences in site quality as perceived by buyers, and as the prices on the entire system settled down into a more stable pattern, it looked to me like large sites and small sites were getting about the same prices per unit, maybe even with a small premium in the opposite direction (smaller sites getting higher prices). It's very hard to nail down all the confounding factors, especially because as I describe in the next section, to a large extent participants were not really buying and selling with real money.

But there's a different, somewhat related, effect I am sure is real: the quantity premium on the buyer's side. These days, with more years of Web comic writing behind me and now as I'm trying to run a business and make money on the Net myself, I am generally more a buyer than a seller of Internet advertising, and there are many people like me. How much will I pay for a click? At a given fixed level of quality, paying for ad space or paying for "conversions" are all equivalent to paying for clicks because they're all related by constant factors, but I'm going to use click count both because it's easy to count and because it's how I actually do think of these matters, and my psychology in this case is pretty important.

If you offer me 1 click when I advertise this Web log: I won't pay for that at all. I can get just one reader for anything I write by sticking the link on Twitter. (Hi, Mom!) One click in a day won't make a difference to me; my Web log will still be an obscure backwater of the Net.

If you offer me 10 clicks in one day: I might pay $0.01 each for them. That is about the best I could do on Project Wonderful with careful use of zero and low bids. With 10 clicks in one day it'd be enough to put a blip in my visitor stats. For a Web log entry I can hope, but not expect, that out of 10 new viewers someone might re-share it, and for a product on my commercial site, I can again hope but not expect that it could just maybe result in a sale.

If you offer me 100 clicks in a day: I might be happy to pay $0.03 each, and willing to consider paying $0.10 each. This is the level at which re-shares or sales become a real possibility instead of a hope, though they are still pretty unlikely.

Nobody is offering 1000 clicks in a day without catastrophic compromises on quality, but if I could have 1000 real clicks of uncompromised quality all in one day, that's the level at which I could realistically expect to see some re-sharing that could lead to viral propagation, or a commercial sale, and I'd consider $0.10 per click to be a fair price for that.

Look at the pattern of prices here: as number of advertising clicks increases, the price per unit goes up, in a pattern I call the quantity premium. Note that although I wrote about it in terms of what I'm willing to pay, it's not just me, and it's not just buyer psychology. It is an observed fact that the actual sales of Internet advertising that do occur, do exhibit a quantity premium. When I've tried to buy clicks both on Project Wonderful and elsewhere, I really have ended up paying more per unit when I buy them in larger quantity. That is totally insane! Other goods and services do not sell that way at all!

For example, I use a lot of 1N4148 switching diodes in my electronics manufacturing business. The current prices from Mouser, a fairly reputable supplier for these, in Canadian dollars are $0.138 each if you buy 1; $0.084 each if you buy 10; $0.030 each if you buy 100; $0.021 each if you buy 1,000; and continuing to decrease per unit as the quantities go up. Last year I bought one reel of 10,000 of these diodes, current price $0.014 per diode, which is almost a factor of 10 less than the price for just one; and the quoted per-unit price keeps decreasing beyond that to as low as $0.010 per diode at the 100,000 diode level. If you were buying millions of them, you could deal with ON/Fairchild, who run the factory, directly instead of going through Mouser, and get an even lower price yet. If you need many billions of diodes, you'd consider building your own diode factory and going into competition with ON/Fairchild. This kind of quantity discount is pretty much universal for electronic parts and most other goods.

It's easy to understand how the quantity discount would come about. If you buy just one 1N4148 diode from Mouser, somebody is going to have to cut it off the reel and put it in a little plastic zipper bag for you and slap a label on it and the label and bag and the employee's time all cost money. (Probably far more than $0.14, actually; that price is a loss leader for getting you onto the Mouser Web site to buy other things.) If you buy the whole reel, someone has to slap a label onto the reel and throw it in the courier bag... which costs about the same as selling you the one diode, therefore 0.01% as much per unit. The diodes themselves are literally made from sand. The cost of the materials is not significant in small quantities. Until you buy them in really huge quantities, it's stuff like the labour of putting labels on the package, more proportional to the number of orders than to the size of the orders, that makes up a large proportion of the overall price.

Although I think it's especially noticeable with electronic components, it's the usual case for most goods to have some amount of quantity discount. So why does Internet advertising do just the opposite?

Other times that I've written about the quantity premium on Internet advertising, my best guess was that it had to do with some kind of risk seeking. Posting an article here and getting one or ten readers for it is basically worth no money to me. I want to be famous. I want 10,000 readers for what I write, minimum. And actually getting that is qualitatively different from more of the same. There are things I can do with a 10,000-reader platform that I just can't do at all (not, do to a smaller degree) with a 10-reader platform. So having 10,000 real readers for a posting of my choice is worth more to me than 10,000 times the value of getting just one reader. "Risk seeking" is usually thought of in terms of betting - it would make sense to me, as a rational bet, to pay more for a 1 in 10,000 chance of getting 10,000 readers than for getting just one reader guaranteed. But mathematically, saying that I'm risk-seeking in that sense is pretty much equivalent to saying that I'm willing to pay more per reader if I can have 10,000 of them. Risk-seeking utility of Web comic (or Web log) readership could explain my willingness to pay a quantity premium.

One problem with that analysis, which has become clear to me as I've moved from advertising a Web comic to advertising a for-profit business, is that I don't think for-profit businesses (who are the main buyers of Internet advertising) operate the same way I did when I was a Web comic artist. Mine certainly doesn't. If I could have two sales, that's twice as good as having one, but not really any more than twice; the money I'm willing to pay out for advertising to make sales is in proportion to the money coming in and that is only linear. And I wouldn't even want to have one million sales, because there's no way I could fill that many orders. Web comics seem to be a weird special case with risk-seeking utility functions, but the quantity premium on advertising does not only apply to Web comics.

My current thinking is that the quantity premium mostly comes about because of simple supply and demand. More so than with other goods, Internet advertisers often want to buy a significant fraction of the entire supply of suitable advertising that exists, or maybe even more than all of it. Just by placing such a large order, the advertiser has meaningfully increased the total demand in the world. Then with fixed supply, it is straightforward that prices will go up.

ON/Fairchild will not run out of sand and can make as many diodes as companies like North Coast care to pay for. But new advertising opportunities in a network like Project Wonderful cannot be created on demand. At best we can hope that if prices go up more Web sites will be tempted to join as publishers because of the higher prices, and there are limits to that, too. The whole world isn't really so big. Project Wonderful in its heyday was only boasting of maybe 10 million ad impressions served per day. I could only reliably get about a 0.1% clickthrough rate on my ads and that only from the publishers who were the best matches for the ads I wanted to run. A 0.1% click-through rate on 10,000,000 impressions makes 10,000 clicks in the entire system. Across the entire system, maybe 10% of those potential clicks really exist and are acceptable quality for me in particular. The rest are either people who won't really read my site after clicking the ad, or won't actually click on my ad after all (driving down the 0.1% click-through estimate) because my ad is a lousy match to what that publisher's visitors are interested in clicking on.

So it's reasonable to estimate that there existed at the absolute most about 1,000 clicks per day for sale on Project Wonderful that I could really buy and would not consider seriously compromised as to quality. If I wanted to buy 100 of those, it would be enough to make a noticeable dent in the entire supply (and was, the few times I tried to do it). If I wanted 1,000, that was the entire supply and I'd be at the mercy of whatever minimum-bid prices the sellers wanted to charge. I actually wanted 10,000, and too bad - that many clicks in a day without quality compromise could not be had at any price, because they just didn't exist.

The fact of real, serious single orders being big enough fractions of the entire market to distort the supply-demand balance seems like it explains most of the quantity premium on Project Wonderful. It suggests that to reduce or eliminate the effect, Project Wonderful really needed to grow a lot bigger, so that the supply would be big enough to absorb orders people really wanted to place without creating price distortion. Never reaching that critical mass may have been another factor leading to their doom.

But even with an increased supply I think there'd be reasons that some bidders who could afford it would always want to buy a significant fraction of all the advertising opportunities that exist. One reason is to shut out competition. It's worth more to me to have you see my ad and not see my competitor's, than to just have you see my ad. If there is a reasonably hard limit on the channels of communication to potential customers, and I can realistically expect to buy up a large enough fraction of them that other parties can be completely excluded, that's worth a premium to me. It's fragile, though, and may be difficult in this highly connected era. I might plausibly buy out all of Project Wonderful, but not all of the Internet. And it only works if I buy all or nearly all of the supply. Fifty percent will not do.

Here's a more significant reason to pay a premium for a significant fraction of the supply but not a large majority: the effect that Kevin Simler calls cultural imprinting. Nearly everything human beings ever do, and nearly all the money human beings ever spend, we do for the purpose of proving - signalling - to other humans what kind of people we are. See also David Chapman's Buddhist take on this fact.

If you buy a Leapfrog VCF, you don't really want the best-sounding VCF available. (Too bad! That's what you get!) Really, you almost certainly buy a Leapfrog VCF because you want to prove to others that you are the kind of person who buys a Leapfrog VCF: you are someone who values hand-made electronics, analog technology, and top-quality documentation; someone who has the necessary disposable income to afford an expensive, complicated toy, and the technical skills to play with it properly; and so on. And you only get your money's worth in signalling your identity to others, if they know what the signal means. If the response is just "VCF, WTF?" then you didn't get your money's worth. I want saying "I bought a Leapfrog VCF!" to mean more to someone whose opinion you care about, than saying "I bought a Zorgla Frobnugget!"

So I don't just need you to see my ad. I need you to know that the people you want to impress have also seen my ad. I need you to see it in public where you know many others see it too.

Right there we've got a reason for a quantity premium. The value of my product to you increases with the number of others who (as far as you know) have seen the ad for it. If you see my ad multiple times, it doesn't just mean linearly more chances for you to click; it means more reason for you to click each time, because you estimate more people to have seen my ad too and so my product has more relevance for signalling to them. My ad needs to reach a significant fraction of the population you're part of, which means I need to buy advertising in an amount that is proportional to the size of that population. I don't need to buy all or almost all of the market, but I need to buy what in computer science we would call some constant fraction of it.

Cultural imprinting would also explain the premium (though as I've said, I'm no longer convinced this is large or necessarily even exists) for advertising on large Web sites. When you see my ad on some little podunk Web site that you obviously visited but you don't think your friends visit, you can't expect them to have seen my ad. But if you see it on a major, popular site, you can expect others in your peer group did too, and that makes my value proposition of "Buy the widget that sends a message to others!" more valuable. Very public impressions are more valuable than the same number of impressions on smaller sites.

Not Real Money

Bids on Project Wonderful typically ran in the range of US$0.01/day, maybe up to US$0.05/day. I never saw a sustained take greater than that on my own site, even when my site had more traffic than it does now. A tiny number of super-star sites (Oglaf the porn comic being one of them) actually got tens of dollars per day.

If I'm making minimum wage - supposedly rising to Ca$15.00/hr here in Ontario next year, but it's anyone's guess whether that will survive the new government - the network's 30% share basically eats the US/Canadian dollar conversion and my entire day's take of $0.01 as an ad publisher comes to 2.4 seconds of my time. At minimum wage. Actually, I have an expensive PhD and don't feel entirely complimented to be offered work at minimum wage. So how much of my time am I willing to spend dealing with a financial transaction the size of a Project Wonderful bid?

In a very real sense, financial transactions in amounts like $0.01/day are not transactions in real money. The way we deal with and understand such transactions, both as buyers and sellers, is different from the way we deal with larger transactions.

Ryan and Project Wonderful seemed to be hoping that the apparent insignificance of the payments would lead market participants to spend money more freely. "They're only $0.01 each? I'll take a thousand!" His expectation that people wouldn't care about the 100% price penalty of the untruthful auction, seemed to fit into that analysis. Then for the serious buyers whom we all always hoped would show up, the people who would spend real money amounts (dollars per day, at least), there were "campaign bidding" and other tools to help them easily manage the very large numbers of very small bids it took to actually spend that much.

I think that in practice, it worked the opposite way. Both because of the small amounts and for other reasons, I valued "money" in my Project Wonderful account much more than the same number of supposed dollars in other, more real spaces (like my wallet). And I think the observed behaviour of other users was consistent with their doing the same. Again: money in Project Wonderful wasn't real money. It was something else that happened to have a dollar sign attached.

First of all, there's a transaction cost associated with making or processing a bid. Just like with bagging and labelling diodes one at a time, there's some fixed amount of work that goes into buying or selling anything no matter how small. That transaction cost sets the minimum possible scale of the differences in prices. It takes a lot more than 2.4 seconds to place a bid on a Project Wonderful ad box; and having done so I'm not going to say "Oh, then I don't care whether it ends up being $0.20 or $0.10"; once I've geared myself up to care about buying something so small at all, I'm damn well going to get the $0.10 price instead of the $0.20, and the $0.01 price instead of either of them if I can. I can remember getting into vicious multi-box tactical bidding fights with other users on Project Wonderful over amounts of money that ended up being less than I would think nothing of spending on a pint of beer.

Next, there are significant barriers to transferring money in, or even more so out. Project Wonderful transfers only through PayPal, PayPal may charge fees and Project Wonderful adds fees of its own, there are minimum amounts, so on. Spending a dollar that is in my Project Wonderful account, and subsequently replacing it, both is and feels like a much bigger deal than spending a dollar out of my pocket. This is another reason for me to be very cautious about even a penny-sized transaction on Project Wonderful, and that drives the nominal prices down.

Since the prices are so low, basically nobody has a hope of making a living on Project Wonderful advertising. Computing market size and the 30% cut suggests that even Ryan himself, even if he had no costs or employees and pocketed all the money, couldn't really have lived off of it at the height of the system's popularity. But it's worse than that: ordinary people on the Net with ordinary-person-sized audiences couldn't even cover Web hosting by selling Project Wonderful advertising. As very rough orders of magnitude:

That's just insulting. Why bother if it's about the money? The answer is, it's not about the money. Or at least, this path will only be chosen by people for whom it's not about the money.

And so Project Wonderful started and permanently remained the domain of non-profit Web comics and a few other kinds of marginal characters, such as the online community of persons who have the sexual fetish of being restrained with glue in particular. (I think they call themselves Stickies.) These non-profit participants couldn't afford to put real amounts of money into advertising. Instead, they were basically bartering ad swaps with each other through the medium of Project Wonderful. Web comics advertising on the sites of other Web comics makes sense in a way that we would not expect for most commercial businesses.

I think most Web comics on Project Wonderful were operating as both buyers and sellers, and most had the expectation that they would simply spend as buyers the money they received as sellers, never taking it out of the system and not putting in money from outside either. That was certainly how I intended to operate when I was running my Web comic.

Obviously, someone had to pay for the 30% cut the network took at every step. I think some of the Web comics crowd threw small amounts of money in, and I occasionally did too, especially at the start before I became somewhat disillusioned by it all, but most of the deposits probably originated with the few serious commercial businesses that were actually trying to buy serious commercial advertising. I'm sure much more than 30% of the money paid in eventually ended up in the network cut instead of coming out in someone's publishing payout, because of the constant recycling of publishers' "earnings." Very little of the Project Wonderful credit that ever passed through my own account, ever came from real money I put in from the outside world, and I'm not sure whether I ever actually made a withdrawal. It is reasonable to guess that others operated as I did.

With only a tenuous link between the value of credits inside the system and the value of money outside the system, there was little reason for prices to have any connection to the prices of "real" advertising paid for in a way more closely linked to money in the outside world. I could imagine the effect going in either direction. If I didn't convert real money into my Project Wonderful credit to begin with and I can't convert credit back to real money either, why not spend thousands of dollars of credit per click when everyone else is doing so? The numbers are meaningless. But in fact, the effect went in the other direction: high value on even small amounts of credit, driving nominal prices again to the absolute rock bottom. And, yet again, it was a vicious circle: with no hope of making real money nobody who actually needed the money would come online as a seller, and with sellers so let us charitably say specialized, nobody who could spend real money would come online as a buyer.

And I think that is the bottom line: what killed Project Wonderful was that nobody could afford such low prices.


(optional field)
(optional field)
Answer "bonobo" here to fight spam. ここに「bonobo」を答えてください。SPAMを退治しましょう!
I reserve the right to delete or edit comments in any way and for any reason. New comments are held for a period of time before being shown to other users.