Econometa

A PDF version of the complete primer can be found under docs. Hopefully this will at least help a few people out there besides me who are trying to get all this stuff straight…

—– Google, auctions, and arbitrage —–

As always, the picture comes before the explanation, and clicking on it will give you a bigger version. This one shows how the Google advertising system works.

—– Google —–

Since Google has been so successful in advertising, and since AdSense is currently the main player in placing ads on smaller publisher sites, it is worth going into some detail on how Google advertising solutions work. Many other companies use elements of these solutions as well, so a lot can be learned about the market in general by studying Google.

Google has two advertising programs, one for advertisers and one for publishers:

– AdWords: This program allows advertisers to buy ad inventory based on keywords. The ads may appear on Google SERPs, within Google applications, on other search engines such as AOL, and on publisher sites subscribing to the AdSense program.
– AdSense: This program allows publishers to sell ad inventory to advertisers participating in the AdWords program. Ads may be targeted either contextually or based upon search keywords entered in a site-specific Google searchbox.

So Google is both a publisher and an ad network. One reason for Google’s success as an ad network is that every advertiser customer as a publisher automatically becomes an advertiser customer as an ad network. This neatly solves the problem all “two-sided markets” have, which is how to prime one side of the market before the other exists.

—– Keyword auctions —–

Another factor in Google’s success is its pioneering use of keyword auctions in selling ad inventory. As we noted previously, a huge advantage of search is that it automatically generates very high quality targeting data, the search keywords themselves. A keyword auction increases market efficiency by linking pricing to targeting data, and also lowers Google’s cost of sales by setting prices and fulfilling purchases automatically.

The auction itself operates in a straightforward way. For each set of keywords, advertisers choose:

– The ad that will appear
– Additional targeting criteria (geo and/or specific sites)
– The maximum CPC that will be bid
– The maximum daily dollar amount of ad inventory that will be bought

For a given ad space, Google then collects all active bids on the keywords associated with the search or web page, and places ads taking into account the maximum bid amounts. To encourage advertisers to enter high bids, Google reduces the CPC of the top bidder to be just over the next highest bid.

However, a key innovation in Google’s ad placements is that the CTR for a given ad is taken into account along with the maximum bid in determining placement position. This means that an ad with a high CTR can be placed more prominently than another ad with a higher bid, but a lower CTR. This has two consequences:

– Users see more ads that other users have previously clicked on, and which are therefore presumably more relevant and helpful.
– The overall CTR of displayed ads is higher, increasing transaction volume and thus Google’s income, which is a percentage of CPC.

—– Keyword arbitrage —–

It is neither difficult nor capital-intensive to set up an intermediary site that is paid based upon leads or referrals, and then to drive users to the site by buying keyword ads. This has resulted in arbitrageurs trying to take advantage of market inefficiencies to “buy users” for less than they “sell” them.

There are several ways to approach this arbitrage, including the following:

– By acquiring better knowledge of the auction system, one can buy users that vendors have missed for a CPC that is less than vendors are willing to pay for a referral.
– By acquiring better knowledge of user behavior, one can buy users based upon non-obvious low-CPC keywords and sell them for a higher CPC or CPA to vendors.
– By adding value to the user experience, one can buy users at a CPC and sell them to vendors at a CPA that generates a profit.

It’s probably worth noting that while they do act to decrease inefficiencies in the market, the first two techniques also can tend to pollute search results with spurious links that deteriorate the user experience. However, they also increase the number of bidders in the keyword auction, thus keeping prices high and benefiting the search engine, so it’s not clear where motivations lie on this topic.

—– Tracking and reporting —–

Along with the auction mechanism itself, Google and others also provide tools for both advertisers and publishers to track and calculate various relevant quantities, including CPC, CTR, ROI, and eCPM. In order for advertisers to take advantage of these tools, they must consider two more pages in the marketing process:

– Landing page: the page that users see after clicking on an ad.
– Conversion page: the “thank you” page that users see after completing the action desired by the advertiser, such as entering information or purchasing a product.

Since the JavaScript that a publisher places on a web page in order to insert ads comes from Google, Google can do several things:

– Identify the publisher page on which the ads will appear
– Count the impression against the selected ads
– Determine technical targeting data for the user
– Place a cookie in the user’s browser

When a user clicks on an ad, the link is actually to Google, who then redirects the browser to the advertiser’s landing page. This lets Google:

– Count the click against the ad and the publisher page the ad appears on
– Identify the ad and publisher page in the user’s cookie

Finally, to track ROI advertisers must place a Google beacon on the conversion page. This lets Google read the user’s cookie and count the action for use in ROI calculations.

It’s probably worth noting that during this process, Google can collect information on ads clicked and conversions achieved per user. This comprises additional valuable targeting data on the user, and if the user is logged into a Google user account, can potentially be aggregated across browsers and computers. Google requires advertisers to make the conversion page beacon visible as a link marked “Google site stats,” which leads to a page explaining why the beacon is present and for what purposes the data will be used.

—– Search engines and intermediaries —–

As usual, the picture comes first, with explanations to follow.

This figure shows how vendors use intermediaries as outsourced marketing in the same way that manufacturers use distributors and resellers as outsourced sales. As usual, you can click on the diagram to get a bigger version.

—– Search —–

According to our terminology, search engines are publishers: they are a web site that sells advertising space. So are sites that incorporate search in their use, such as shopping sites.

A huge advantage of search is that in performing it, a user automatically provides the highest quality user targeting data known: search keywords. Keywords tell the publisher what the user is interested in at that very moment, without causing either user inconvenience or undue privacy concerns.

In addition, search has become an important de facto interface to the Web for many users, so that ranking highly in search results and advertising on search engines has become an important factor in marketing success for many companies. The result is that search has quickly become a huge component of online ad sales, spawning an entire sub-industry in the process.

—– Search engine marketing —–

Search engine marketing (SEM) has two aspects, both of which can become complex and so are often outsourced to SEM specialist companies. The search engine results page (SERP) has two sections:

– Organic results (AKA natural results): this is the list of links that the search engine found most relevant to the search keywords according to its presumably objective search algorithm
– Sponsored links (AKA paid placements): this is the list of ads shown along with the organic results; they are usually text ads, which may either avoid distraction from or encourage confusion with organic results

Attempting to increase a web site’s rank in the organic results is called search engine optimization (SEO). SEO can range from modifying a site to be more accurately indexed by search engines to manipulating the search engine algorithm by creating false keywords, links, or other factors used to determine ranking in organic results.

The other aspect of SEM is managing paid search, that is, buying keywords on search engines and calculating the resulting ROI. Paid inclusion is a program some search engines have that allows a site to pay in order to be “guaranteed” inclusion in organic results.

—– Intermediaries —–

If you use a search engine to search for a common product, for example digital cameras, you’ll find several different types of sites in the organic and paid results:

– Manufacturer sites (e.g. Kodak)
– Distributor sites (e.g. Circuit City)
– Shopping comparison sites (e.g. PriceGrabber)
– Directories and research sites (e.g. CNET)
– Vertical specialist sites (e.g. digitalcamera-hq.com)

Distributors and value added resellers (VARs) are a normal part of the offline world of sales and fulfillment: they are vendors who buy product from the manufacturer and sell it directly to the customer. But intermediaries such as comparison, research, and specialist sites make their money in a completely different way. Instead of selling product themselves, they link the user to a vendor for purchase, and for this they are compensated.

The relationship between intermediary and vendor depends upon whether any user data is included:

– Affiliate marketing: If the intermediary simply causes the user to follow a link to the vendor site, the link is called a referral, and the intermediary is called an affiliate.
– Lead generation: If the intermediary collects and passes along data about the user such as contact information and details concerning what the user is looking to purchase, the user following a link to the vendor site is called a lead, and the intermediary is called a lead generator.

The line between affiliates and lead generators is not always distinct, but in general affiliates are smaller and more numerous, and are usually provided with a standardized way to make referrals (such as Amazon affiliate product links). Both are paid on a CPC or more often a CPA basis.

Since intermediaries are compensated by generating interest in a vendor, they can be viewed as outsourced marketing, in the same way that distributors and VARs can be viewed as outsourced sales. And just as distributors often supply VARs, affiliates can join an affiliate network, which gives them a central place to manage and track affiliate relationships with many vendors.

Next up, Part IV: Google, auctions, and arbitrage.

—– Outsourcing ad buying and selling —–

I’ll again put the picture first, with explanations to follow.

This depicts how ads flow from advertisers to publishers, including the roles of ad agencies and ad networks. Like last time, you can click on the diagram to get a bigger version.

In this post I’ll dive more deeply into the particular viewpoints of advertisers and publishers, and how they translate into motivations in the ad economy.

—– The advertiser viewpoint —–

From the advertiser viewpoint, the key question is:

Advertisers have traditionally viewed their activities as falling into two categories: branding and direct response. Branding builds general awareness and future likelihood to buy, while direct response attempts to directly lead to a sale. Although this line is blurring, branding is generally associated with graphical CPM ads, while direct response is associated with CPC or CPA ads.

Ads are sometimes part of campaigns that attempt to optimize effectiveness by controlling the order, duration, frequency, and context in which users see the various ads in the campaign. Campaigns can extend across various media including the Web, and managing them is often outsourced to an ad agency. This outsourcing includes coming up with actual advertisements (creative) and/or placing these ads and tracking their performance or return on investment (ROI).

ROI, which is usually expressed as the ratio of gain over investment, has traditionally been difficult to measure, since there was no way to know what ad(s) influenced a given purchase. In the case of advertising investments, ROI is sometimes called return on ad spend (ROAS), and is calculated as follows:

One of the advantages of the Web is that ROI is immediately apparent in the case of direct response ads that lead to a sale. The average profit per impression is

so the ROI is

This calculation is even more straightforward if the ad space is paid for on a CPC or CPA basis; if the action is a sale in the latter case, the ROI is simply

Thus CPC and CPA compensation shifts the risk of mis-estimating CTR and/or CR to the publisher, even though these measures are mostly affected by the creative and conversion efforts of the advertiser.

—– The publisher viewpoint —–

From the publisher viewpoint, the key question is:

The first answer is to simply maximize the quantity of ad inventory. Ad inventory depends upon page views and ad space, so publishers can either try to attract more viewers or increase the number of ads per page (which tends to adversely affect eCPM).

The second answer is to maximize eCPM. When buying ad space, advertisers are willing to pay more to reach the right users at the most receptive moment. Thus the key to increasing eCPM is user targeting and content targeting, as previously detailed. Content targeting is straightforward enough, but user targeting runs two risks:

– Asking for user data via a registration barrier is a hassle for users, and drives many of them to leave or enter false information
– Inferring user data by tracking or correlating to an offline database can raise privacy concerns

Clearly then, a key challenge for online marketing is to approach the end goal of individualized or one-to-one marketing while preserving the user’s privacy and sense of control.

Just as advertisers outsource to agencies, publishers can outsource ad sales to ad networks. Ad networks aggregate ad inventory across many publishers and provide a single place where advertisers (or agencies) can buy ad space in volume. This helps lower the cost of selling for publishers with ROS or “remnant” inventory and smaller publishers with less inventory; similarly, advertisers and agencies can more easily buy in volume, lowering their cost of buying.

Next up, Part III: Search engines and intermediaries.

—– Advertiser and publishers —–

I put the picture first, but the rest of this post will explain the pieces that are shown in the diagram. Note that this is a conceptual diagram, not a technical one, so for example the publisher does not transmit user segment data to the advertiser, instead segments are offered to the advertiser part of the sale of ad inventory. You can click on it to get a bigger version.

—– Ad inventory —–

A publisher has a certain amount of space on pages within the site that is set aside for ads; this is called the ad space. Then each page with ad space is served to a user some number of times during a given time period; this is called the page views.

The publisher’s ad inventory is then the ad space multiplied by the expected page views in a given time period. So a publisher’s monthly ad inventory would be the set of ads expected to be viewed by users during that month.

This seems like a pretty simple situation, but it can be complicated by various technicalities. One complication is in counting page views. For example, by page views we really mean views by a user (not for example by a robot or a search engine spider). Some concepts that are used in calculating an accurate ad inventory include:

– Hit: a request for a file on the site
– Page view: the serving of a web page, which usually includes many files, and therefore many hits
– Session: a unique destination of page views within a certain time period; usually a single IP address that has received some number of page views within 24 hours

Besides distinguishing robots from humans, counting users and ad inventory can be made more accurate by knowing if hits are actually delivered to the browser and viewed, avoiding counting spurious hits such as page refreshes, and mapping users to sessions in light of dynamic IP addresses, etc.

—– Types of ads —–

Ad inventory is characterized by various general attributes, including:

– Premium: the ad will appear on a page which is considered to be valuable beyond the number of users viewing it, e.g. the home page of a portal
– ROS (run of site): the ad will appear on any page within the site; for an ad network this is called RON (run of network)
– ROC (run of category): the ad will appear on any page within a specified category, e.g. business or music
– Targeted: the ad will appear in page views that are targeted by either content or user, e.g. an ad for golf clubs is targeted to pages that have several instances of the word “golf” or users who have expressed an interest in golf

In addition, ads come in various formats, many of which have been standardized. These include:

– Text: a short textual description and a link
– Banner: an image placed horizontally on the page
– Vertical banner: an image placed vertically on the page
– Skyscraper: a vertical banner that is taller than normal
– Interstitial: an ad that loads between two content pages
– Pop-up: an ad that creates a new browser window
– Pop-under: a pop-up that is created behind the active window
– Rich media: an ad using animation or audio/video, such as Flash

—– Ad targeting —–

Ads can be targeted in two main ways: by the content near which the ad appears (content targeting), and by the user who is viewing it (user targeting). These two techniques are part of what is sometimes called relevance marketing (as opposed to mass marketing).

Content targeting has several variants, including:

– Site targeting: targets ads to “vertical” sites that focus on a specific topic or user segment related to the ad
– Category targeting: targets ads to a category within a site (see ROC above)
– Contextual targeting: the text on a given page is analyzed and an algorithm decides which ads best match this text

Unlike content targeting, user targeting depends upon knowing something about the user’s interests. Various methods exist for obtaining user data, thereby gaining the ability to offer user targeted ad space to advertisers:

– Registration: based on data provided by the user, usually in return for premium or personalized content
– Technical targeting: based on data that can be obtained from the user’s HTTP page request, such as domain, ISP, connection speed, operating system, and browser type
– Demographic targeting: based on the user’s demographics such as zip code, age, gender, etc., usually obtained via registration or surveys
– Geo targeting: based on the user’s location, usually obtained via registration or from the IP address
– Behavioral targeting (BT): based on the user’s behavior, e.g. which kinds of pages have been viewed in the recent past
– Search keyword targeting: based on keywords the user provides in a specific search, either on a publisher site or on a search engine

In order to allow advertisers to match ads to user data, users with similar interests are often grouped into user segments or categories.

—– Compensation —–

The advertiser can pay the publisher according to several commonly used measures:

– CPM (cost per thousand): the advertiser pays a fixed amount per thousand impressions, an impression being a user seeing the ad
– CPC (cost per click): the advertiser pays a fixed amount per click, a click being a user clicking on the ad to visit the advertiser’s site
– CPA (cost per action): the advertiser pays a fixed amount per user action; the action can be an inquiry or lead (CPI, CPL), a sale (CPS), or any other kind of specified transaction (CPT)

The latter non-CPM measures are called performance-based, since the advertiser only pays based upon the ad performing, that is, generating an interaction with the user. Publishers often convert performance-based measures into an effective CPM (eCPM) by taking into account the percentage likelihood an impression will yield a click or an action; for clicks, this is called the click through rate (CTR). For example, for a CPC ad, we have:

eCPM = CPC * CTR * 1000.

For a CPA ad, the advertiser must measure the conversion rate, or the percentage likelihood a click will yield an action. If this is denoted CR, we have:

eCPM = CPA * CR * CTR * 1000.

Next up, Part II: Outsourcing ad buying and selling.

As I’ve mentioned before, I think that advertising is an important part of making the Web work well: it helps developers and writers get paid, gives users more choice in how they support sites, and reinforces the shift in power to participatory users and small creators. But it also certainly has its problems, which I’ve been trying to get my head around.

As a relative newcomer to the world of online publishing and advertising, I’ve spent a lot of time working through the details and, as I like to do, organizing my thoughts with pictures. I put together a summary which seemed to help a few people I showed it to, so I figured I’d clean it up and put it out there for everyone to check out.

I have two main purposes in doing this:

(1) I’m hoping people who have more experience than me will help out with corrections, clarifications, and comments. I’m sure I’ve missed some things and misunderstood others, and I really want to make this as accurate and useful as possible for myself and others.

(2) Once it’s been subjected to scrutiny by real experts, I’d like to put these posts into a document that can be a resource for everyone. I know that when I was looking into this stuff, I immediately ran into a lot of confusing jargon and assumptions, and no single compact introduction to help clear it up. Hopefully this can help out the next person who comes along.

OK! So I’ll start with some introductory things, and then move on the meat of it in the next post.

—– Terminology —–

To keep things consistent, I’ll use the following terminology:

– User (AKA consumer): a person surfing the web
– Browser: the user’s web browser
– Publisher: a web site that wants to sell advertising space
– Advertiser (AKA marketer): a web site that wants to buy ads
– Ad network: a company who buys advertising space from publishers and sells it to advertisers

A “publisher” refers to any web site selling ad space, and so could be an application, service, or site that might not actually “publish content” in the traditional sense. An advertiser is assumed to have a web site, since most Web ads link to a site where the user can find more information, purchase a product, etc. Finally, note that a single web site might be both a publisher and an advertiser.

—– Scope —–

There’s a bunch of terms that are used to refer to the general activity of marketing and selling on the Internet, including:

– Interactive marketing
– Online marketing
– Internet advertising

Now, there’s a lot more to the Internet than just the Web, and there’s a lot more to marketing than just advertising; but here we’ll focus on the placement of ads on web pages (which should probably be called “Web advertising,” but for some reason nobody really seems to use this phrase). Interactive marketing in general has a more general scope, some aspects of which include:

– Email marketing: ads are emailed to the user
– Adware: ads are shown by a software application downloaded by the user
– Permission marketing: the user volunteers to receive ads, by email or otherwise

Email marketing sometimes takes place without obtaining user permission and/or providing the ability to stop receiving the ads, in which case it’s called spam. Adware sometimes collects data or displays ads without user consent, knowledge and/or the ability to uninstall, in which case it’s called spyware.

OK, with that all set, the next post will cover the basics in Part I: Advertisers and publishers.

In a previous post I tried to describe an economic shift that I think is helping to support a new environment less tolerant of the monopoly power inherent in private enterprise platform determination:

The liquidity in the maturing online advertising industry, which allows new applications to monetize utility to users quickly and directly.

In other words, if you can build an application that has enough utility for a decent number of users to start using it, you can turn that utility into dollars by, say, slapping up AdSense, which pays the bills and keeps the app up and running (perhaps competing with a platform aspiring to monopoly power).

The problem with this in the real world is that AdSense, which at the moment is the easiest way to put advertising on a small site, doesn’t really pay the bills all that well. One big reason for this is that the ads are usually pretty badly targeted when the context is a dynamic app, or a constantly changing document like a blog.

This is where the Web is different from other media: as many people have pointed out, the Web is neither one-to-many (broadcast) nor one-to-one (email); it’s many-to-many. That means that everyone can provide data as well as receive it. In particular, users can “pay” for content with more than just their attention; they can pay by supplying data about their interests that lets ads be better targeted. Although smaller sites currently have a hard time monetizing this data, in theory web applications should become economically viable at a much lower user base due to both more valuable targeted ad space and the previously discussed reductions in development and operations costs.

This idea of paying with data isn’t new, we already do it in several ways:

– Supplying search terms is a payment: your interests at that moment are targeting data (e.g. Google AdWords)
– Reading content is a payment: the content itself represents your interests (Google AdSense)
– Frequenting a site is a payment: your history at that site represents your interests (e.g. Amazon)
– Registering at a site is a payment: facts like zip code, age, and gender act as proxies for your probable interests (e.g. NYT)

This trade of data for content is what Matt Blumberg calls the “New Media Deal”, as I just found out by following a link from Fred Wilson. Matt’s description of this deal is really great, but it doesn’t mention what I think is a big problem: most people don’t get what the deal is! I don’t have any handy stats, but I’d bet that if you took a survey, most people wouldn’t know that the reason they’re always being asked to register and fill out forms is to help the site pay its bills by serving up more relevant ads. They probably think it’s to spam them or do market research or something (er, well, both of which might sometimes be the case actually).

It seems to me that asking users for data might work a lot better if users really understood what it was for. Matt addresses this in part with his next deal incarnation involving more user participation, the “We Media Deal”:

The more transparent the value exchange, the more willing you are to share your data.

But the examples he gives, of us being more likely to care about sharing our data if we know we can delete it and that it will be attributed to us, isn’t really what I have in mind here. Instead of trying to get data from users in indirect ways like surveys, registration, and tracking, why not just make the deal explicit? People understand that someone has to get paid to develop apps or write articles, and if we can pay with something other than money, that’s great!

I think it’s true that a lot of people are used to thinking in the “old media” way: as Matt puts it, we pay by “tolerating” a blizzard of ads, most of which are totally irrelevant to us. If the New Media Deal is made more explicit, I think people will see that everyone wins: a less painful type of payment can support a greater diversity of sites, where participation and mutual respect are values that are reinforced by capitalism and self-interest.

In a conversation related to my previous postings on power laws, a question came up: If a ranked distribution follows a power law, what percentage of the total is in the highest ranked bin? So for the example of a histogram of users ranked by the % of taggings, what percentage M of all taggings are made by the very top user?

It turns out that this depends on whether the power law is an exact inverse (Zipf: a = 1) power law or a higher order power law.

The top user u = 1 has M percent of all taggings, so the curve is t = Mu^-a. Each bar measures the percentage of taggings by that user, so the sum of all bars has to equal 1. So for N users we have

M + M/(2^a) + M/(3^a) + … + M/(N^a) = 1

or

M = 1/(1 + 1/(2^a) + 1/(3^a) + … + 1/(N^a)).

For a Zipf law with a = 1, the denominator is the harmonic series, which diverges; so that means the % of taggings by the top user drops as the number of users N gets larger. We can calculate M by remembering that the harmonic series sums to gamma + ln(N) as N approaches infinity, where gamma is the Euler-Mascheroni constant and ln is the natural log. We can check that this is close enough after N = 100, so calculating N = 10 by hand and using this formula for the rest we have:

Gotta love NumSum. But if a > 1, the series in the denominator converges, so that as the number of users N increases, the % of taggings by the top user M quickly settles to a constant:

This is all in follow-up to the fourth point from this post:

(4) While it is true that “bigger systems benefit from both higher heads *and* longer tails,” in general this usually just makes the histogram fit the curve better; it is rather the shape of the curve that determines whether or not “most activity is from a small group of highly active users.”

A Zipf law is a case where a bigger system actually has a distinct effect: the bigger the system, the lower the percentage resident in the highest ranked bin, resulting in a lower percentage of activity from the most active users. In the case of higher power laws, this percentage quickly settles to a steady constant, so size doesn’t have much of an effect once the system is reasonably big.

As an aside, I was also asked to post the graph presented at TagCamp showing a histogram that fits a “long tail” but not a power law, so here it is:

Although this looks similar to a power law, if we disregard the top two users the histogram actually fits the curve that corresponds to a perfect bell curve PDF. This means that in contrast to a power law, where the average number of taggings per user is essentially meaningless, above this average is maximally meaningful.

In my last post, I talked about how as a technology platform, Web 2.0 isn’t that much different than Web 1.0; really, going from 1.0 to 2.0 is more of a marketing indicator that significant new value is being created on top of this platform.

But there are some other aspects to what people mean by “Web 2.0.” Paul Graham calls it “democracy, and not dissing users.” Tim O’Reilly calls it “the architecture of participation.” And pretty much everyone agrees it involves users owning and controlling their own data.

I think that one interesting way to look at this aspect is through the economics of platforms (a great example of an “econometa,” what this blog is supposed to be all about). There are two ways that a platform, or a standard technology infrastructure, can be built:

Private enterprise:
– Many proprietary standards vie for adoption
– One winner attains a lucrative monopoly
– Over time the monopoly expires and the platform becomes a public resource

Public discourse:
– The government or a standards body solicits proposals
– Private companies and individuals lobby for their solution
– A winning platform is selected and designated a public resource

Currently in the U.S. at least there is a tendency to see one of these as “good” and the other “bad” depending upon your political views; I’d argue that there are plenty of situations in which either can be the more appropriate. But one thing is certain: the first way involves a lot more money changing hands. The cost of lobbying is much smaller than the cost of many companies developing and marketing solutions, and the profits associated with being selected to build a public resource is much smaller than the profits associated with holding an ongoing monopoly over that resource.

I’d like to suggest that a part of Web 2.0 is a better understanding of how the private enterprise route towards a platform works, and a decreased tolerance for the power and profits that go to the monopoly winner. This social shift is being accompanied by several economic shifts that reinforce it:

– The ease and low cost of using the Internet to gather together interested parties and work towards a standard without any central authority
– The low cost of building a platform (or an incremental platform component) and applications that use it
– The high number of people with modern browsers and/or broadband who can easily adopt new applications
– The liquidity in the maturing online advertising industry, which allows new applications to monetize utility to users quickly and directly

Taken together, I think that all of this means that the private enterprise path (on the Internet) becomes closer to the public discourse path. In fact, it seems to lead to a “third way” that in many ways exhibits the strengths of each without the weaknesses: a winner can be selected from many different approaches based upon survival in the market, absent huge costs or large concentrations of profits and power, and accompanied by wide discourse.

So yes, I do think it is still possible for new eBays and Amazons to be created, but I think it will be more and more difficult. Instead, models like those adopted (at least initially) by Google and MySpace, which base their value upon utility to the user rather than a lock-in of user data, will become the more certain path to success.

So back to the question in my last post: “should I consider building a new application on a new, proprietary platform?” I think that the bet made by doing so could have reduced risk in this new environment. Charging for the platform itself and maintaining central control over it are becoming less viable; if a company doesn’t work with the interested (e.g. open source) communities, these communities will probably recreate the core value of the platform quickly enough to provide an attractive alternative before the company can gain critical mass. But that doesn’t mean that the basic questions in my last post don’t still have to be answered:

– Is the platform ubiquitous? (e.g. mapping and search APIs? No)
– Will there be onerous fees? (if the platform owner can get away with it, yes)
– Will the platform owner be your competition? (if there isn’t a clear way for them to make money, probably)

So I probably wouldn’t disagree with Greg on the viability of mashups as a serious business model.

One key economic shift listed above that I haven’t elaborated upon is the maturing online advertising industry. I think this is a really important part of the story, but again I’ll have to save it for a subsequent post…

Fred Wilson says that the definition of “Web 2.0″ has become so hyped up that it’s borderline worthless as a term, but that he likes the early definition “the web as a platform.”

I don’t disagree with this as a technical description of why the web is such a powerful enabler; but the thing is, it seems equally accurate for “Web 1.0.” Web 2.0 is a new “version,” and I think that versioning is always, for better or for worse, more of a marketing thing than a technical thing: sometimes it means lots of technical changes, sometimes not; but what it always means is that you’re claiming that enough new value is being created that you want to “re-launch” or offer an upgrade.

A classic instance of this is when Skype got a lot of attention and Yahoo “re-launched” Yahoo Messenger (v6.0) as “Yahoo Messenger with Voice” (v7.0). The fact is, Yahoo Messenger had offered PC-to-PC *and* PC-to-phone calls for years! It was just not very well-marketed, and so not too many people knew about it. In this case the new version was just a way to remind people that “hey, we have that stuff too!” Of course, a bunch of improvements and small new features were also included, but that’s not really what the new version was about.

In the case of “Web 2.0,” I think we’re also talking about what is primarily a marketing term: I like Paul Graham’s interpretation, that it really just means “using the web the way it was meant to be used.” There are a few new features like RSS and the ping infrastructure, but the real reason for the new version is to say “it’s worth taking another look, really cool things are happening.”

Regarding platforms, looking at the comments to Fred’s post, it seems to me that people might be mixing up two separate ways a company might build value:

(1) by building a new application on an existing platform
(2) by building a new platform

The word “platform” in software usually means a standard infrastructure you can build applications on. The web is such a platform, and one that to an increasing degree supplants the PC OS. Applications like Google search and del.icio.us tagging are of the first kind: they are built on this platform, “the web as platform.” But Google and del.icio.us APIs, and for that matter eBay and RoR, are of the second kind: they are *new* platforms. Moreover, they are *proprietary* platforms.

A question that a lot of people seem to be asking is, “should I consider building a new application on a new, proprietary platform?” The standard answer is that you can, but this consists of making a pretty big bet:

– that the platform will become ubiquitous
– that the fees charged by the company owning the ubiquitous platform will not be onerous
– that the company owning the ubiquitous platform will not decide to compete with you, and do so by altering the platform to give itself an advantage

But I think that this answer might be changing. This brings me to what I think is a second aspect to “Web 2.0,” a new value that is not so much a technical improvement, it is more of a social and economic shift. I’ll have to save that for the next post…