Internet Outsider

The (Big) Problem For Hakia, Powerset, Mahalo, and Other Google-Killers

The "highlighter" feature that search-engine Hakia announced yesterday wasn't worth a press release, but it did get me to try the company's "semantic search" service, which is actually pretty cool.  As instructed, I asked Hakia three English-language questions:

Why did the stock market crash?
Where do I get good bagels in Brooklyn?
Who invented the Internet?

As promised, I got intelligent results for all (even the last one, which was a trick question).  For example, Hakia understood that, when I asked "why," I would be interested in results with the words "reason for"--and produced some relevant ones.  If I'm ever in the mood to ask an English language question--and I remember that Hakia exists while reaching for the keys--I might use the engine again.

But therein lies the problem--indeed, the problem for Hakia, Mahalo, Powerset, and the dozens of other companies that are pursuing next-generation search.  Contrary to the premise upon which most of these companies are based, I don't agree that current search sucks.  On the contrary, I almost always find satisfactory results immediately, conveniently, and with minimal frustration.  I also don't find myself wanting to ask the Internet English language questions all that often: It's usually easier to just type keywords.  The results (and display) could always be improved, of course, and maybe I'm always missing out on fantastic sites that have just the info I'm looking for, but ignorance is bliss.

On the questions I asked, Hakia certainly delivered nice results. But I'm used to using Google and Yahoo, and Google and Yahoo usually get the job done, and I almost never wonder whether I'm getting "the best possible results."  So unless Hakia, et al, focus on tight, defensible verticals--or sell their technology to Google/Yahoo/Microsoft--I don't think their future is promising. 

Don't believe me?  Check out Hakia's modest traffic over the past year. Or just ask the guys at IAC's Ask, who, despite being widely viewed as having the "best search on the web", despite massive advertising, and despite the brilliant Barry Diller, haven't budged off of 2% market share.

Calacanis's Mahalo Off to Solid Start

Jason Calacanis's new human-powered search engine, Mahalo, appears to be off to a solid start.  According to Jason, the site is currently serving about 50,000 pageviews a day (1mm a month).

Mahalo also recently launched a new "Greenhouse" program that addresses the primary business concern about the company--namely that the cost of hand-building SERPs will be so high that Mahalo will never survive.  The Greenhouse program allows the public to apply for part-time paid work building Mahalo search-results pages (SERPs).  According to Jason, Greenhouse received 800 applications in two weeks.  The company has hired 150 people as Part-Time-Guides (PTGs) so far and hopes to have 500 by year end. 

Greenhouse PTGs create SERPs based on the company's "Most Wanted" list, and Mahalo's full-time editors then accept or reject them. Mahalo is giving the first 100 PTGs to hit 100 accepted SERPs an iPhone. 

The Greenhouse program is currently producing 10-20 SERPs per day, at a cost to the company of approximately $10-$15 each. The company expects its PTGs to be producing 100 SERPs a day by the end of the year.

For most search queries, a human-edited SERP will be superior to one generated solely by an algorithm.  Here, Mahalo's ability to prioritize SERP-building is the key to success.  If the company had to compete with Google on speed or breadth, it would fail.  Assuming search requests follow standard distribution laws, however (e.g., 20% of terms account for 80% of searches), Mahalo ought to be able to invest its resources in building only the most popular and profitable SERPs, with links to Google for the rest. 

Assuming Revenue Per Search (RPS) of, say, $0.05 to $0.10, Mahalo would have to generate 100-300 queries per SERP to generate a positive gross margin.  It's not clear how long a shelf-life each individual SERP has (presumably minutes for some and years for others), but given the traffic the site is already generating, this doesn't seem a high hurdle for the company to meet.