If you can read this, I just earned $0.00001

So. We’re ramping up advertising ’round here in MyxerLand. It started, like all buzz-kills, with the suits. Somebody invited a finance guy into the office the other day, and he spent the whole time he was here looking at one of those funny pieces of paper with all the numbers in a grid, and muttering something about zero being “an awfully small number.”

“No, no,” I tried to reassure him. “Look at all those really big numbers — the pretty red ones in parenthesis! That’s good, right?”

* * *

If you’re a regular visitor to MyxerTones, you’ve probably noticed that lately it looks like vandals have broken in and scattered Flash-formatted graffiti randomly on our pages. Hopefully you weren’t on the site during The Smiley Horror Trilogy (starting with the somewhat humorous but unwelcome “The Smiley Incident”, followed by the obligatory “Son of Smiley”, and rounding out with “Smiley’s Revenge”). I’ll have more to say about the Smiley people later, but for now let me just say that it is absolutely amazing how much money is poured into Smileys advertisements. It’s like, more then…umm…ringtone advertisements.

The good news is that we’re actually earning some money now. Weird. The bad news is that, even with the Smiley Guys banished, the ad quality is, well, terrible. There is apparently an irrational amount of venture capital being poured into companies with “spokeschickens,” and they seem to really like hanging around our website. Funny thing: we’re currently working on implementing a new UI for MyxerTones, and we’re working off of Photoshop mockups from the graphic designer. For the ad placements, he used really cool iTunes banners with black backgrounds and emotive subject matter, which really make the layout look sweet. With the chickens in the ad placements…not so much.

(Incidentally, I’ve heard that the Smiley people are so keen for you to download their “free smileys” because the download actually infects your machine with spyware, and the spyware shows you ads even when you’re not online (which presumably also infect your machine with their own spyware, and so on and so forth). I’m not sure if that’s true, and I’m also curious about the vulnerability of bloggers to libel lawsuits for repeating stuff they’ve overheard, especially given the obvious resources Smiley Inc. could bring to bear against me if they wanted…)

* * *

So we have to start this whole advertising bit up to get our revenue number to be, well, less small than zero. Just in case you’re not familiar with the way internet advertising works, allow me to present a really quick summary:

1. Get a web site. If you don’t have a good original idea, don’t worry. Just pick a name that sounds like another successful website our there, and hire a college kid to register “your” domain name and put a simple home page up.
2. Generate a lot of page views. You can do this by having a really useful website that people like to use and visit, or you can just rent a botnet and aim it at your web site. Either way, be sure to collect the demographics or your “users,” because that will be Really Valuable Information when you need to…
3. “Partner” with the Ad Networks. The obvious first step is to sign up with Google so that they can deliver crappy “contextual” ads on your pages. There are other companies that can get you different – but equally crappy – ads, and you should sign up with as many of them as possible. The most important criteria to consider is SPM (Smileys Per thousand impressions), which is a rough but effective measure of the throughput that a given ad network has between your site and Smiley Central. Believe it or not, some of these networks can actually exceed SPM’s of 1,000, displaying more than one Smiley per ad impression.
4. Retire.

OK, well, that’s what I was led to believe anyhow. A funny thing happened on the way to step (4)…

* * *

It turns out that when you use these ad networks to serve ads on your site, you generally don’t earn That Much Money. It’s really hard to give hard numbers for what you can earn (the networks are intentionally vague about it, and many even contractually forbid publishers from discussing their earnings!), but on a general purpose site with reasonable ad placements I would be fairly surprised to see an effective CPM of anything over $0.50. (Some types of sites actually can do better than this, and many do worse, so there’s a lot of fudge in that factor).
(I used the term effective CPM above. Effective CPM (eCPM) is a common way of measuring advertising earnings as a function of ad impressions. It’s useful to discuss earnings in eCPM because often times the ad networks pay publishers (websites) based on how many clicks ads get (“CPC”), or how many users the advertisers acquire (“CPA”), as opposed to how many times the banner is actually displayed (“CPM”). Regardless of the actual payment trigger, all of these other metrics can be converted into an eCPM value, which makes it easier to compare and optimize.)

So the first thing you will do after you hook up these ad networks and start looking at the stats is…Try to figure out what’s Plan B. ‘Cos you’re not going to retire on $0.00001 a page view if you’ve got 20 mouths to feed. See, these ad networks are generally only good for dealing with what’s called remnant inventory. Remnant inventory is a fancy way of saying “ad slots we couldn’t sell.” Ad networks are in the same basic category as the ad serving platform (like DoubleClick, Atlas, 24/7, etc.) — they’re essentially required infrastructure for an advertising-supported business, but a business they do not make. The eCPM you get when you fill with these networks is generally only going to make you profitable if you’re a part time operation.

To increase eCPM, there are several things you can do. First and foremost, you need to sell directly to advertisers to cut out all of the middlemen (sorry, Googuys!). But what happens at this point is that you realize that, umm, you have to talk to people. And you need to sell to them. So you have to hire people to do those things for you, because you are, after all, a technogeek who is much more comfortable behind a keyboard than in a sales call and has a notorious weakness for telling the truth. And so you hire people that know how to communicate with these alien advertisers. And then, according to my experience so far, the next thing that will happen is that you will be asked to quantitatively assert that:

1. Yes, we have a huge population of 18-24 year old males on our site, and
2. No, we have no potentially offensive content on our site.

Now, I’m not going to get back into the whole potentially offensive content discussion at this point, but I will reprise my earlier observation that an advertiser with no stomach for possibly having their banners on the same page as “potentially offensive” content in the form of butts-in-thongs or whatever has no absolutely business trying to reach 18-24 year old males. OK, I’m done.

See, people who sell internet advertising think that it’s really important to have a good handle on a site’s Demographics, because certain advertisers are interested in reaching certain groups of people, and they will Pay a Premium for the Privilege (read: higher eCPMs). You need to know, they say, the age, gender, and geographic location of every user of your site, and if you can extract information about income, marital status, bank account balances, key vices, etc., that would be swell, too. But the most important information is summed up by the handy ZAG acronym (“zipcode, age, gender”) that is apparently part of the parlance.

The easiest way to get this data from users, of course, is to simply require that every user enter the information when registering for your site. Well, that’s true assuming what you really want to do is:

1. piss off those of your visitors that are protective of their personal information;
2. lose another slice of users because they don’t want to fill out a long form;
3. get fake information from the rest of them.

But, hey, fake information is better than no information, right?

* * *

What’s really funny to me is that this whole ZAG thing, while apparently universally accepted as a great way to allow Targeting (and, ultimately, elevated eCPMs), seems terribly inefficient. Start at the beginning: a company has a product or service, and wants to reach people who are likely to be interested in said product or service. So, they hire some consulting firm to do a study to find out what are the Key Demographics of the users who are likely to be interested in the product. (Obviously, this step comes before product development in a lot of cases). The consulting firm comes back and breaks things down to, effectively, this ZAG group and that ZAG group is likely to respond well to products of this type, based on sampling and statistical analysis, etc.

Thence follows the demand for ZAG statistics from web publishers, which leads to required personal information fields on web registrations, which lead to pissed off users, reduced registration rates, and decreased accuracy of data.

But what if instead of asking people to supply us with their personal ZAG information, and using that to target campaigns for advertisers, we just cut right to the chase and simply ask “what types of products and services are you interested in?” Just come clean and say, hey, this is an advertising supported site, and we will indeed subject you to advertising. If you’d be so kind as to tick the boxes next to the topics you find most interesting, we will try to deliver ads that are most relevant to you. I mean, if my company sells jewelry, wouldn’t it be great to be able to buy targeted advertising that reaches users who have proactively indicated that, yes, they are interested in jewelry? That’s a first order correlation, unlike the significantly more removed ZAG data.

I call this approach the ZIG (“Zero personal Information Gathered”) approach. Instead of requiring personal information such as gender and age from users, we allow them to tell us which class of advertisements they would like to see. The visitor is happier because they don’t give out personal information (other than interests), and advertisers have the potential to reach a hyper-targeted audience that has professed to be receptive to their products and services. So, advertisers get better results, which justifies higher retail pricing of advertising inventory on our site, which (gasp!) increases eCPM.

* * *

So we’re going to be working on this advertising problem for a long time, trying to increase our eCPM as much and as quickly as possible while always maintaining a deep respect for our users’ personal information, and looking for ways that we can actually provide better options for advertisers than they might know exist. All the while, slaying Smileys and enduring spokeschickens and the rest of their ilk.

Internal (Myxer) email on content filtering

As a follow-up to my post of the other day, here’s an email I sent internally regarding our procedures for dealing with potentially offensive content.

From: Myk Willis
Sent: Wednesday, August 01, 2007 4:42PM
To: mVisible Technologies
Subject: Dealing with Potentially Offensive Content

I’ve written about five drafts of this email, with each one turning into a book that I can never find the time to finish. So while this actually turned out to be fairly lengthly, please understand that there is a lot (I mean a lot) of caveats, special cases, apologies, and admissions of deficiency swirling in the background that I wasn’t able to get into any readable form.

Filtering Potentially Offensive Content

By default, every user account (and anonymous web visitors) has a setting called “content filtering” that is turned on. Registered users may disable this content filtering in their account settings if they so choose.

The “content filtering” switch is intended to prevent potentially offensive user-submitted content from being displayed to users who have it enabled. When content filtering is on, three things happen:

(1) User-submitted text is run through a profanity filter before being displayed;

(2) Content from the catalog marked as “potentially offensive” is removed from all browsable or searchable areas.

(3) User-submitted images (note: this does not include wallpapers, which are covered by #2 above) that have been marked as “potenially offensive” will be filtered.

Mechanics of Filtering Profanity

All user-submitted text, including user profile names, item titles, descriptions, biographies, tags/keywords, comments, and probably some others I forgot, are run through a “sterilization” process before being written to an output page. If the content filter is on, any word or phrase in the text that matches a known ‘dirty word’ will be substituted with grawlixes so that, e.g., “shit” becomes “$%^!”.

The known dirty words are maintained in a file that is part of the project, and we can adjust them if need be. I would caution, however, that my advice for patrons of Taco Bell (“never, ever, open the burrito”) applies fairly well to this text file as well.

Note: Tags (keywords) that match a profanity will not be displayed at all.

Mechanics of Filtering Images

Users may submitted images to Myxer in three ways: (1) as user profile images, (2) as images associated with ringtones or songs, (3) as actual content items (wallpapers and videos).

Every new image uploaded to the site will be queued for review by a lucky member of our staff. We will commit to reviewing new images within 24 hours of their submittal, though we will likely “work the queue” during the day to keep up with new stuff.

There is no delay between when an image is submitted and when it may be displayed to other users, so there is a 24 hour window during which potentially offensive content may be in general circulation.

We currently do not have a method of knowing when images of type (1) and (2) have been changed, so we will have to do a little work to insure changed items are re-queued for review. (Wallpapers and videos are immutable, so we can get away with just reviewing them upon creation).

Deciding Which Images are Potentially Offensive

Please understand this is one of the points at which I could go on for volumes describing the impossibility and subjectiveness of any attempt at this. But I won’t. Instead, I give you my version of “you know it when you see it”:

Potentially Offensive Content is: any image containing
enough exposed flesh to embarrass you if your children and/or parents found
you looking at it.

This definition is, I believe, much more conservative than what we’ve previously applied to the site. In particular, an image need not contain full or partial nudity in order to be considered offensive. Extremely scantily-clad persons in suggestive poses would fall in this category as well.

Beyond Potentially Offensive

Full nudity is not allowed in the MyxerTones catalog. Any item containing nudity will be excluded from the catalog, such that it will not be visible in search/browse even if content filtering is off.

Anything beyond “simple” nudity will be deleted upon detection, and may result in termination of the originating user account.

Common Sense

Each of us, as employees, members of the Myxer community, and human beings has the right and responsibility to mark content as “potentially offensive” that we believe could bring harm to members of our community, whether because of nudity, hate speech, or whatever. Don’t be a slave to any particular rule, and do what you feel is Right at the time. We can and will revisit our decisions and guidelines often.

In Summary…

It isn’t our intention to be the moral cops of the mobile universe. As I’m sure most of you understand, the classification of something into the category of “offensive” depends in an impossibly complex way on the individual and the circumstance. And the importance of self-expression to our user community is self-evident. We have an important mission, and now a mandate from the masses, to be a force that works tirelessly to move what should be personal decisions about content choice out of the hands of the carriers and into the hands of the – duh! – individual. So I don’t want us to ever fall into the trap of blindly imitating the behaviors of those that have come before us.

When ever we are accused by outsiders of crossing some line – be it with regard to allowing the possibility of hosting offensive content or copyright abuses or whatever – it is invariably a result of our position as a disruptive force in their industry. With so much control in the hands of the corporations entrenched in these industries, it takes a strong will – and sometimes an extreme position – to have any affect at all.

Tweaking how we identify and deal with potentially offensive content is part of our overall goal of providing a friendly and comfortable environment for our users while still allowing for a maximum amount of personal choice and self-expression. Rest assured that we’ll continue to revisit this topic until the end of time.

As always, massive amounts of flame mail are invited!

Myk

P.S. Have a spare half hour? I have some more musings on my blog that may be of interest to provide further background on my personal thinking on the subject: http://mykwillis.wordpress.com/2007/07/28/holy/