From Istanbul To Sand Hill Road

You Were Partially Outbid!

Vulcan's are the wisest of Star Trek characters, and they have a equally wise farewell remark, which is, "live long and prosper."  Well, if I keep using the product from Prosper, it is certain that I won't live long.

But before I get into the sad experience I had with this company, I must say that I was a big fan, and have a lot of respect for their investors, some of which I know.  In a conference I named Prosper one of the most interesting sites on the web today.  I think microlending has the ability to change the world and democratize finance.  I would have invested in this company.  It is because of this admiration, I put some money in my Prosper account to lend and try it out.

I had a great plan.  30% of my total was allocated to higher risk borrowers, 40% to mid, and 30% to low risk borrowers.  I had a spreadsheet lined up ready to go.  Of course I was thinking, "if I can lend at 13% and borrow at 8%...maybe there is the opportunity to arbitrage".  So I wanted to test it out and figure out the origination costs and one time fees, that would eat into my hypothetical arbitrage window.  All great in theory.

The reality on the other hand is that, since there is an auction process at Prosper, you bid for the right to lend money and the one who bids the lowest interest rate wins.  Auctions worked for Ebay, it surely works here too right?  Well on Ebay you are bidding for a product, most of the time it is something you know about and you have only the price to think about.  You can set limits easily.  Intuitively, you know what a pair of shoes or tennis racket should be worth.  When you are bidding for a financial instrument, you need to calculate or at least think, whether 10% is high or low for somebody with an A credit rating.  That's not so simple. 

And because this is an auction, you may be outbid.  In this case, it is almost comical but you can be "partially outbid".  Get that?  That means if you bid $100, $56.5 was accepted at 12%, but the other $43.5 needs to be bid lower than that.  How is that for simplicity? If this happens you get the wonderful "You Have Been Partially Outbid!" message.  Oh my God!  I can live with being outbid, but partially?  I  cannot be seen in public being partially outbid. 

Jokes aside, the point is that this is not as simple as it looks, and that was the disappointment on my behalf.  There is some learning you have to do that they don't help much with.  Shortly my inbox got filled with the thrilling "You are Partially Outbid!" messages causing me to bid carelessly just to "lend the f... money and get this over with".  Since we are talking about small amounts, and I was trying to learn the system, getting 8% vs. 9% wasn't so much of a concern.  Maybe they should just show you how much you will get paid per month, and let somebody bid for that instead of interest rates.  More intuitive, one would think.

The user interface is well designed but the user experience is not.  That's the point of this article.  It needs to be fixed.  It is way too complicated, and I suspect it is because somebody made an Ebay analogy that didn't work.  I will still use it, I want to be a fan, but if I get one just one more email saying "You Were Partially Outbid!"  I am going to completely, not partially, outbid Prosper and try Zopa.

New Ads In New Places

I wrote this piece originally for Matt Marshall's Venturebeat.  You can read it directly there or just continue below.

-----------------------------------

New Ads In New Places

There was a great discussion on video ads earlier on Venturebeat. The discussion surfaces two questions that are facing all online services. If the ad industry is to grow from $16B to $30B in 5 years, what will the new kinds of ads be and where are the new places to put them?

Surely, it won’t be more of the same; our eyes are already beginning to ignore the ‘ad gutter’ no matter how relevant the search engines are trying to make them. So what will they be? VCs are seeing three new emerging categories of online ads, one of which could become the lion’s share of all of online advertising.

1. Online ads that look like content
Pre-roll and post-roll ads inserted in a video stream are in this category but long term success is still questionable. You don’t click on pre-roll ads because you are waiting to watch the video, and you don’t click on post-roll ads because you rarely watch a video all the way to the end. The bigger end game is product placement. When you view an image on Flickr, there is no reason why the background cannot be altered to show an ad relevant to your demographics. That’s making an ad look like content. There are technologies that will place ads right inside user generated images and video. It can be done in a number of ways. Your image or video editing tool can help you place ads at the time of content creation.

Alternatively, ads can be inserted after content is created. A piece of software scans an image of video to find appropriate places to insert ads. As a result, a man wearing Adidas shoes, could all of a sudden be replaced by Nike shoes in an image. This method is more challenging because one can’t rely on the creator to make the right judgment on where to put the ad. The algorithm has to be smart enough to figure it out, but there are teams in Silicon Valley tackling this task as well.

2. Online ads that are content
Some ads can be content that people look for. It’s done every day. Those ads are called coupons. When somebody does a local search for pizza in Menlo Park, Applewood can put an ad in the ad section, or put a coupon right next to their address. Either way, it’s an ad to bring in a customer and may cost the store the same. However, there are empirical results showing that coupons are clicked on 10x more frequently than text ads, because they are considered content. It is no surprise that Google did a deal with Valpak to enable local coupons.

Another example of ads that are content is the products of companies like Winster. They are very simple casual games where people buy credits to play to win movie tickets, and promotional products. The game is little more than a gimmick to show you an ad. It’s a customer acquisition tool like a coupon, but people see it as content.

3. Online ads that are offline ads
Now this depends on your point of view. If you are a heavy 2^nd Life user, spending 40 hours a week, that’s more than 2 full days a week (adjusting for sleep) in the virtual world, by now you probably think a billboard ad in the game is an offline ad. The trend I am talking about is in-game advertising. This is another area where we will see a lot of innovation, because the amount of advertising that could be shown in a virtual world is far bigger than the potential for text ads. After all, for each person in the US, advertisers spend $2000/yr across all media, TV, magazines, direct mail etc. If 2/7^th of this total were spent ‘in game’ the total would amount to more than all of the online advertising business today.

Conclusion: The first two categories will yield interesting companies, who discover new and effective ways to show ads. There is clearly room for innovation there and that is enough to get VCs excited and companies funded, but it is in the third category that the most money will be made. It’s simply a much bigger market. I would highly recommend entrepreneurs to find ways for their services and technologies to capture a piece of this emerging market before Google and Yahoo do. It must surely be on their radars.

Do It Yourself Vertical Search Engine

There are a number of companies, big and small, trying to build a better search engine.  Whether it be vertical search engines, or more general search engines with better algorithms than Yahoo & Google, they all face a big hurdle to success.  Simply stated, the hurdle is the near zero marginal cost of doing a search.   And the more companies I see the less comfortable I get that any will cross that hurdle.

The promise of vertical search is a good one.  If the search engine focused on one sector, such as shopping, blogs, electronics or health, they don't have to crawl the entire web, and focus on much fewer sites and hone their algorithms to deliver more relevant results faster.  As a result, the company will attract more relevant ads, therefore both cost per click and click through rates will be higher than others.  Eventually, the company will be runaway success, reach out and conquer other verticals, and then eventually take over from Google.

Almost all of these companies, when fundraising, take a keyword, like "digital cameras"or "san francisco" and compare their highly relevant search results with the generic results you get from Google.  It looks great.  However, what's sorely missed is that when comparing search engines, one should not compare results for a given keyword, but compare the overall cost of finding the information that's desired.  If you type the wrong keyword, you may have to refine your search, lengthen the phrase, or reach out to page 2-3-4 of the search results to find what you need.  Eventually you find what you think is what you want.  It may take 1-2 minutes longer than in vertical search engine, but I claim that that cost is not a big problem for any consumer.  Let's test that hypothesis.  How many times have you searched something in Google and got frustrated concluding "I just can't find what I need!".  I would guess that that is never.

When a user tries a number of keywords or lengthens his/her phrase, what they are effectively doing is adding relevance manually.  That is what the vertical search engines try to do for you, but when the cost for the consumer doing it themselves is nearly zero what do they bring to the table?  If "soccer" is too generic, try "weekend soccer" or "weekend soccer san francisco", eventually you bring the relevance you are looking for and you do it happily without complaining.  So what's the real value proposition for the added relevance from a vertical search engine and how does that compare to the cost of acquiring the users and getting them to your web page routinely?  That's the hard question I have not yet seen the answer for.

As long as the cost of doing another search on Google is zero, I see a difficult time for vertical search engines to keep the relevance edge over time.  I would love to be proven wrong.

Google and The Mathematics of History

Seems like whenever I think of something interesting to write, one way or the other, it ends up being related to Google.  I am developing a fascination with what they can do with all the data they are collecting.  I am becoming to Google, what Dr. Hugo Strange was to Batman, a psychologist, so consumed by analyzing Batman that he ends up trying to be Batman.  But I digress…

Before we get to Google, I have to tell you about Anatoly T. Fomenko, a Russian Mathematician, who makes the bold claim that the timing of historical events, Chronology, as we know it is wrong (Sir Isaac Newton makes this claim in a paper too).  Fomenko asserts that history is off by about 1000 years.  Cultures we think to have existed say 200BC is actually 800 AD.  Ancient Egypt, ancient Rome all existed but a lot later than we have come to believe.  More specifically, the claim is that the clergy in the 16th and 17th centuries duplicated medieval European history into biblical history.  Documents describing the life of kings of ancient Rome are actually documents describing German kinds in 1300.  The most controversial claim that falls out is that Jesus Christ was born somewhere near year 1000.  Before you guys go “no friggin way, you’ve watched the Da Vinci Code too many times”, let me be the first to say that I am not interested about the accuracy of the claims.  You can buy the book here, and form an opinion on it yourself (but I warn you, this is heavy heavy reading).  What I am very much interested is what gave Fomenko the initial idea, and the methodology by which Fomenko reaches his claims.

The initial idea comes from analyzing lunar eclipses.  There are working laws of physics that allow astronomers to predict exactly when and where on the world an eclipse will happen.  We get news of this all the time.  They can go backwards in time as well, and match all the recorded eclipses to what the theory predicts.  But Fomenko notices (in the works of astronomer Robert Newton) a serious mismatch.  Eclipses that were recorded to have happened between 700-1300 AD show lunar behavior that vastly differs from theory and could only be explained by a mysterious non-gravitational force applied on the earth-moon system.  However, this mysterious force completely disappears, and everything matches theory, if the dates of the eclipses were wrong and each one actually about 1000 years later than claimed.  This surprises Fomenko and he gets obsessed with analyzing chronology. 

But he’s a lot smarter than many of us, and devises a statistical method that can determine whether two pieces of text are written in the same time period.  Now this is very important, and touches the crux of what I am talking about.  Language changes over time.  The words we use change.  New words are added and some words disappear.  The kinds of sentences we construct change over time, and all those changes can be quantified.  In a way, the syntax of a document, not its semantics, can be used to determine when it is written.  This sounds reasonable and acceptable mathematically.  As a result, Fomenko takes a documents said to be written in the times of ancient Rome, and compares it to a documents many years later and concludes that they are statistically from the same time period.  There are chapters and chapters comparing kings of ancient Rome to kings of Germany, saying that these two kings were actually the same person.  This sounds very reasonable and acceptable mathematically, but accepting is socially is another matter.  To make it socially acceptable, the mathematics needs to be extremely accurate.  Can it be?  Is there a company out there who has the computing power, the smarts, to take vast amounts of text data and analyze it statistically?

Enter Google.  They have more data and computing power than Anatoly Fomenko could have ever imagined having.  So can this statistical method be improved to get the time interval more and more accurate?  Can they date text to within 50 years, 10 years, 2 years?  This kind of statistics could shed a lot of light into what historical document is fake, what document is real.  It would solve a lot of the world’s problems about how history really happened.  Google can, in effect, take history from the hands of historians, and place it in the hands of mathematicians.  And that my friends, would be an earth shattering event.  Only Google could do this.  It would start a whole different way of debating history, one which is more scientific and less ideological.   This is a big big step towards organizing the worlds information.

There was a time in history where it was unheard of to discuss human origins in biology class.  Who knows, maybe in the near future, human history will be taught in statistics class?