July 1, 2010

57) Google Image Labeler

危 wēi  danger

Google owes much of its success to its phenomenal search algorithm, invented by Larry Page and Sergey Brin while they were attending Stanford University as Ph.D. candidates.

Broadly, Google search works in three distinct parts:

  • Googlebot, a web crawler finds and fetches web pages.
  • An indexer sorts every word on every page and stores the resulting index of words in a huge database.
  • The query processor compares your search query to the index and uses the algorithm to recommend documents that it considers most relevant.

Google harnesses a distributed network of thousands of computers to parallel process this information.   This approach has proven incredibly effective, with perhaps one major exception: image search.  Image search is less reliable because the indexer mines pages for words and therefore only labels images based on their context (and most images on the web are untagged).  Many companies have tried to build software to interpret images but it’s tough to do  – that’s why identifying unclear letters remains one of the last ways of evidencing that we are in fact human, not machine.

Proof you’re [sort of] smarter than machine:

Although there is clear value in being able to search for images accurately, even Google couldn’t afford to have people complete the labour-intensive task of tagging images one by one.

机 jī opportunity

Google asked a different question  - how can we have consumers do this for free?  Answer: make it a game.  Google licensed ESP gaming technology, originally conceived by Luis von Ahn of Carnegie Mellon University and launched Google Image Labeler in 2006 as a beta.

In the game users are paired with another and they compete in tagging images.  The game is great fun: some users reportedly play over 40 hours a week.  The game has enabled the company to ensure that its keywords are matched to correct images, building an accurate database for Google Image Search.

Gaming has great potential for good, other recent examples include Matchin (helping build a database of the web’s most attractive pictures) and Solarstormwatch (helping astronomers spot explosions on the Sun to give astronauts an early warning if dangerous solar radiation is headed their way)

How About…

  • Developing a game to harness consumer power economically?
  • Applying gaming ideas & principles to your existing offer?

–

Here’s a screenshot from the game:

May 6, 2010

50) McKinsey

危 wēi  danger

In its 2009 report, the UK CIPD’s Employment Survey claims that the average cost of filling a job vacancy is between £4333 and £7750.  This is the average across all sectors and doesn’t even include legal or training costs.  For Management Consultancies this number must be far larger – the firms visit universities in the recruitment drive and often give signing on bonuses.  With this in mind surely McKinsey’s average employee stay of about 3 years is a fundamental floor to the business model?

机 jī opportunity

Far from it, McKinsey cleverly keeps its leavers close to the Firm, recognizing their potential value.  It delivers this through its alumni services – both formal events and informal networking. This dynamic network is widely understood to be a lasting benefit of a career with McKinsey, thereby improving its appeal as a recruiter. The backbone of the McKinsey network is the firm’s alumni directory which lists the details of 3,500 ex-McKinseyites and is more up-to-date than the alumni rosters at Princeton or Harvard.  The strategy of setting up its alumni to be potential future clients must be working – McKinsey has produced more CEOs than any other company and is referred to by Fortune magazine as “the best CEO launch pad”. And you can’t blame the Firm for not publicizing the fact that Enron’s Jeff Skilling was among those high-flying CEO alumni.

How About…

  • Keeping departing employees close: supporting them where possible and viewing them as potential ambassadors for your company?

—

April 14, 2010

47) Facebook

危 wēi  danger

Many technology startups, particularly social networks, are heavily dependent on network effects (the phenomenon whereby a service becomes more valuable as more people use it, thereby encouraging ever-increasing numbers of adopters).  This makes it very challenging to attract the first users to the system because they see relatively few benefits until others join. When Facebook was launched in February 2004 by Harvard undergrad students as an alternative to the traditional student directory it faced exactly this challenge, how would it get to critical mass?

机 jī opportunity

Facebook overcame this challenge because it tapped into an existing offline community – Harvard University students.  And that group had a real need – it was difficult for them to meet fellow students outside their social groups.  The most pressing need to meet students came from those searching for dates – the group that became the system’s earliest adopters.  Facebook initially only allowed students with a Harvard email address to use the site, and then opened it up to other Ivy League schools.  This enabled the startup to control its growth and in the process made it more aspirational to the wider population.  Since launch it has grown to more than 400 million active users, with over 50% logging in on any given day.  Profits are believed to exceed $1bn per year.

I noticed that for the week ending March 13th 2010, Hitwise said that for the first time more Americans typed Facebook into their browser than Google.  It made me wonder if Google’s algorithms for search are being disrupted by different types of recommendation engines – if Facebook can harness its community to offer ‘discovery’ it might become the search engine of choice.

How About…

  • Piggy-backing off an existing offline (or online) community?
  • Initially targeting those that are likely to be influencers for other groups in due course?

—

« Older Entries
Newer Entries »