The Gong Show

Mark Zuckerberg said that Facebook is adding 5MM global users per week currently.  That number is incredibly large, and I want to add some context to this number.

We have a number of social media investments at Union Square Ventures, and when we first invested in each of them, it was pretty early in the size of them.  The gross numbers were low but the growth rate was very promising.  Around the time of the investment, we talk about goals for the first year. We believe in Big Hairy Audacious Goals (BHAGs) around here, and a typical just-barely-attainable BHAG for an early social media company is to reach 1MM global registered users within 1 year.

So, the best social media startup can strive early on to add 1MM users in a year. By contrast Facebook is adding 52 * 5MM = 260MM users in a year. So, Facebook, despite already being gigantic, can add users 260x faster than some of the fastest growing startups.

A few weeks ago, I wrote that I was interested in finding (or seeing someone build) a good Map/Street-Art mashup that would curate interesting street art via a web service. As I was looking through the finalists for the SXSW web awards today, I found a project very similar to what I described, called Street Art Locator, which is nominated as a finalist in the Art category. The data in Street Art Locator is still quite sparse, but it’s a great start. It even has a cool “TwitterVision-esque” fullscreen mode.

I don’t know exactly how to describe this, but I really like this new pattern in web services of adding something to a URL in order to use a web service.  Perhaps this is best described through two examples:

• If you put “http://bacolicio.us/” (no quotes) in front of a URL, it will put a piece of bacon on the website. Example: http://bacolicio.us/http://thegongshow.tumblr.com/
• If you put “pwn” in a YouTube domain you can download the video you are watching… for example, if I’m watching “http://www.youtube.com/watch?v=oHg5SJYRHA0” then I can insert “pwn” between “www.” and “youtube.com” to generate the following new URL: http://www.pwnyoutube.com/watch?v=oHg5SJYRHA0 which allows me to download the video.

I’d love to see more web services do this kind of thing. I want to use this method to do stuff like, save to del.icio.us, get comScore/Compete metrics for a site, or email a link. It’s fun in a geeky way.

I just read a great piece in the NYT on the way in which scientific research is funded (and a request for change). In short, researchers need to support their work through grants, which encourage less risk-taking and safer hypotheses with more predictable results. The author would like to see a less “free market” approach to funding research in order to encourage researchers to be more ambitious and take greater risks with their work, be more wild and provocative in their theories.

This reminds me of a comment Albert made that really stuck with me.  I fear I may be mangling his idea, so consider this a loose paraphrase, but it would be of greater benefit to everyone if we could get more research into highly unlikely, but devastating, events. If you spread some money around to funding research for a variety of independent, catastrophic Black Swan events, the likelihood that one of those events will happen is actually quite high, and the philanthropist will have done a HUGE service to the good of everyone on earth by getting in front of a life-threatening problem before anyone saw it coming.

For example, I’m going to reserve the nitty gritty of the math for a follow up post on binomial distributions, but lets take the following example: Lets say you have $100,000 to contribute to scientific research. That’s not a large grant size the grand scheme of all money put toward research, but it’s a large enough amount that it can make a real difference to an academic. Let’s also assume that all black swan events happen with an independent probability of 1%. If you give $100,000 to one person working a single crazy black swan event, then there’s a 99% chance that your grant will be wasted because the black swan will never occur.  But, if you allocate your $100,000 to 68 independent black swan events evenly (which is $1,470 per event) then there’s a 50% chance at least one of those events will happen. Using this portfolio play, did not contribute that much money to the devastating event(s) that occurred in this scenario, but at least there now exists some research in advance of this catastrophic event(s), which is a great public service to everyone.

I’m a fan of the freemium business model, but it’s rare that I encounter a company that has done a proper analysis of price sensitivity. If you’re going to charge a customer for your service, you need empirical evidence to prove you’ve got the right pricing model.

If you could model your customers’ responses to changes in price via a reliable function (linear or otherwise) then you could use some simple math (stuff you likely saw in high school) to maximize revenue, but that’s very unpractical for real world web services.  Instead, I love it when I see a company that has done either A/B testing or a cohort analysis in order to determine the optimal price through iterative testing.

A/B testing is the most statistically clean way to optimize pricing. You have your control, or A,  as your current price and then your B, B’, B”, B”’ etc are the various other price points you want to test.  You test all these prices at the same time live on your site, and after the end of some period of reasonable churn, you can conclude which price point is optimal based on the amount of relative revenue earned from the group.

The drawbacks of A/B testing are in the details of the implementation.  Most A/B testing is implemented by cookies. If a user logs into a computer and is assigned a B” cookie, then they’ll see B” cookie pricing on the site.  If you give a special deal to B users but not B” users, then one of the B users might blog the special deal and the site will look discontinuous (or buggy) to B” users because not everyone will see the same price at the same time.  Side Note: Google employs this method of testing ALL THE TIME (hundreds of A/B test going on at once testing multiple variables), and that’s why sometimes you’ll see a blog post about a new Google feature that you can’t see yourself.

To avoid the inconsistant experiences of A/B testing, you can a cohort analysis instead.  In this situation, the control is your current weekly revenue, and then the variables are changes in pricing that are done to the site in sequential weeks.  So, if you started doing a cohort analysis at the beginning of February, the users that signed up to pay for your service in the first week of the month would be in the first variable group, and then the users that signed up at a different price in the second week would be the second variable group, and so on… After a reasonable amount of time has passed to allow for churn, you can see which cohort maximized revenue.

The plus side of a cohort analysis is there is no inconsistant experience because everyone is seeing the same pricing at the same time.  The downside of a cohort analysis is it introduces an additional variable into your experiment: time. You can’t control for time in a cohort analysis, so if there’s a significant event in the real world (or a big event on your site) that would lead to different behaviors amongst your cohorts, then you have noise in your study.  For example, you could never do a cohort analysis around Christmas season because people’s spending decisions fluxuate too much from week-to-week during that time.  Similarly, if you were doing a cohort analysis of price sensitivity during Oct 2008, you would have to throw out your experiment because the huge drop in the stock market during those 4 weeks would introduce too much noise into people’s purchasing decisions.

These are my two favorite methods of doing pricing studies, but there are plenty more. I hope that if you’re using a freemium business model you start experimenting with your pricing, because it’s really low-hanging fruit in terms of maximizing revenue.  If done properly, you could significantly increase revenue without changing a single feature in your service.

John Borthwick wrote a killer post today regarding Google and the cycle of creative destruction. I wrote a comment on that post, and I’m reblogging my comment here. There’s some details that don’t make sense out of context, so it’s probably best to go read John’s post first: it’s well worth your time.

Great thoughts John; nice mental food to chew on.

I think Twitter search today = WebCrawler of old… WebCrawler is the first search engine I ever remember using, and according to Wikipedia it’s the first full-text search engine on the web. WebCrawler was very dumb. If you searched for meatloaf, it gave you the page with the most references to the word meatloaf… there was no recentness priority, there was no pagerank, there was no weighting on page titles or H1 tags. WebCrawler ended up getting VERY spammy as porn sites figured out they just needed to add giant blocks of popular keyword to their pages in order to game the system. It showed people that search could in theory be a great way to find what you want, but it was crippled as the popularity of the web increased.

I think Twitter search today is equally dumb. When I search “plane Hudson” a few weeks back, I received only an ceaseless stream of RTs and links to CNN articles. No news that told me anything more than what I already knew: a plane was down in the Hudson. That’s not to say that there weren’t very important Tweets happening at that time. On the contrary, the TwitPic from the guy on the Ferry that approached the downed plane (you know which one I’m talking about because it was RT’ed 92498327492 times) was an epic piece of citizen journalism happening in real time. But, Twitter search didn’t source that well… all it did was show me what the most recent RT was, not the most important original tweet.

So, I agree that the “notificator” is important, but we are at the VERY BEGINNING (top of the first inning) of its development, particularly in search. And, I think Google sees this coming a mile away (as does Facebook), and both will innovate on this front. I wouldn’t be surprised if Google released a real-time search engine comparable to Summize in the coming year, but with a different, more-useful algorithm than recentness. Perhaps Facebook too, though that would come with a lot of privacy baggage due to their permissioning system so it’s unlikely.

All that said, I have a lot of faith in Abdur. I think he’s one of the smartest people I know, and he’ll find a way to make Twitter search more useful. I hope it is his priority right now.

PS: one small detail. Google didn’t own YouTube in Nov 2005, so I assume you meant to say Nov 2006.

Originally posted as a comment by andrewparker on JohnB blog using Disqus.

This bring me to my latest crackpot web service wishlist entry. I wish there was a good, highly visual Google Maps mashup for graffiti.  The features:

• Think like TwitterVision for the Look and Feel of the service, except instead of tweets it would be recently spotted graffiti.
• Anyone could contribute to the service via emailing or MMSing in a pic from a cellphone, or simply uploading a pic and dropping a pin on a map.
• A “recent” view and a “popular” view, so that you could see both new stuff and amazing stuff, as desired.
• From this, graffiti tours could emerge.  It would be like an outdoor museum, curated by the crowd. Art meets geocaching.

“The Obama administration is expected to impose a cap of $500,000 for top executives at companies that receive large amounts of bailout money,” according to the NYT.

That’s a nice gesture by The White House to try to force responsible use of TARP funds. Stories of high executive pay, $4BN in 2008 Wall Street bonuses, and extravagant AIG bailout parties are frustrating examples of irresponsibility. But it doesn’t address the largest concern in my mind: Dividends.

The EU proposed a moratorium on all stockholder dividends for any company receiving bailout money, until the governments interest on their preferred stock has been repaid.  It’s a punitive measure to ensure that companies are incentivized to repay the government quickly. I think it’s a great idea, and I wish the US would enforce the same.

Citibank and Bank of America intend to pay out a $0.01 dividend/share in Q1, which is down substancially from Q4, but still represents ~$60MM each.  That’s $1.2BN in TARP bailout that’s ending up in shareholders hands, which is far greater in sum than executive comp issues. Furthermore, PNC and Bank of New York Mellon have received billions in TARP funds and neither have cut their dividend. That fact that any unprofitable company or any company with more liabilities than assets would continue paying a dividend reeks of greed to me.

Critics claim that canceling dividends will drive private investors even further away from stocks that receive bailout money, which would necessitate further government bailout.  However, I disagree: if an investor is only sticking with an investment because he/she expects to receive dividends that flow directly from the federal government, and that investor would otherwise sell the stock, then that stock should not exist.  Instead, the government should focus the money on helping the company fail gracefully; forget about the shareholders.