Google Inc. looks for ideas that are
``really crazy'' when sizing up potential purchases, the Internet
company's top dealmaker said.

``We look at everything very carefully,'' Salman Ullah,
Google's director of corporate development, said yesterday in a
speech at a meeting of the Los Angeles Venture Association. ``The
really crazy ones do really well.''

Google, owner of the most-popular Internet search engine,
has about 15 people working on acquisitions that meet with dozens
of companies a week, Ullah said. Mountain View, California-based
Google responds to every e-mail pitching a company, while phone
calls have a 10 percent response rate, he said.

The search engine, which had more than $11 billion in cash
at the end of the fourth quarter, last year bought video-sharing
site YouTube Inc. and DMarc Broadcasting Inc. to move into the
market for radio advertising. Google also bought smaller startups
including online software company JotSpot Inc.

``The crazy ones mean they ignore the usual restraints of
investment levels required or design parameters or `Gee I need
more servers than anyone ever thought was possible','' Ullah
said. ``When you free yourselves from these constraints, you
create crazy, cool things.''

Google wants companies that can build revenue streams from
their users, instead of buying firms with a lot of users that
don't bring in much in sales, Ullah said.

``We don't do traffic for traffic's sake,'' he said. ``It
has to be highly monetizable.''

In the past Google has also used a technique called Monte
Carlo analysis to size up a deal, where computer algorithms are
used to answer questions.

Shares of Google rose $2.86 to $467.39 at 4 p.m. in Nasdaq
Stock Market trading. They have climbed 1.5 percent this year.

Via Bloomberg

Hmmm... sounds like a way of attracting more pitches

Though they sure have the resources to do just about anything technically.

Yep... $11 billion in cash, that's what we call spare money no?

sh!t

They are asking for clear monetization models that have nothing to do with traffic? 

that's a bit hypocritical isn't it? Their entire business model is based on giving stuff away, building traffic, and selling ads

taffic is all they do

yep but traffic allows them to sell ads 

Yeah but that's what i am saying - why put a call out for sites that do NOT just generate traffic

I think this is somewhat mis-stated

the traffic has to be highly monetisable, that's what they mean I think. 

the google guy is actually saying dont worry about resources your apps might use

the reporter is writing that to mean have crazy ideas (not resource usage)

so they want traffic, but traffic that can be monetised 

ok - well badly written

April 12th, 2007

Google’s acquisition strategy should think small (and mobile)

saying that they use the 'monte carlo method' doesn't say much about how they obtain an approximate future value of a business, to somebody who knows what monte carlo methods are this is like saying 'they use numbers to size up a deal'. Monte Carlo methods are used a lot by futures traders (though if you ask one, they probably don't know that the underlying method is referred to as a monte carlo method) to predict the future value of stocks, bonds and warrants etc. The way it works is that you take what it is you are trying to figure out (in this case, the value of a company) and then break it down into all the different components that have an effect on that valuation - so if you are forecasting a stock price you 'input' seasonal info, competition, users, trends, costs, and you can break these down even further (eg. minimum wage has an effect on cost, health of the economy has an effect on the minimum wage - you can refine it down to the smallest components that have an effect, infact the finer you break it down the more accurate your model is). each of these variables has a potential range of outcomes (eg. the cost of silicon can vary from $8 to $14). for every unknown component, you use random numbers. you then run the simulation multiple times to see what the most likely outcome will be. it's used when you need to determine an outcome in a model where not everything is known or can be quantified. You get probabilities that something will occur - eg. 'what is the probability of the stock price being above $5, or above $4.50' - analysing multiple probable outcomes leads you to the 'most likely outcome'. this works really well and is suprisingly effective - just ask a meteorologist. 

 a recent story where you hear about this is with the value of the australian dollar. it shot up another cent late last week because soem analyst somewhere said that the probabilty of an interest rate rise next month is now 60% (most likely outcome). all the other analysts trading USD/AUD obviously then punched this into their systems and were probably told their 'most likely outcome' is the USD/AUD at 83c (or greater)

 But in short, I think Google just said that to sound 'smart' while most companies are using such models, just not referring to them in that way. it would be more interesting if htey talked about what variables they look at and what they think affects startup growth and potential valuation. also, valuation is relative to the valuator - what is worth $1B to Google may only be worth $5M to MSFT. it is only a valuation when it is realised - so the valuations they are calculating are about what it is worth to them - so it is interesting to hear that they place value on traffic that can be monitized 'highly' - and note that he said 'highly' meaning that they want have efficient traffic (eg. less traffic but more hit-per-pound). they made search engine traffic very highly monetizable..

Re Mone carlo simulation .... it still all comes back to how good the model is -- if it doesn't match reality (ie. puts too much weight on one or more factors, or disregards the real factors altogether because of lack of  vision) it will fail...no matter how 'sophisticated it may appear on the surface.

In many cases it comes down to hocus pokus.... with magic masquerading as engineering ... and  an undue faith put in the numbers because it sounds better than cooking Cariboo bones in a camp fire and deving the future by which way the cracks form....

they also use something similar to the monte carlo method for reinsurance for catastrophe modelling,

Ah.... there it makes a lot more sense........ but some interesting writings by Mandlebrot and others would suggest that the curves used are still out of wack with the reality and underestimate the risk of 'the really big catastrophe" .... what's needed is a power-law distribution and its not usually put in place for most models... hence the suprise when disasters bigger than forecast happen.

reinsuarnace also works on the theory of large numbers - that if you cover enough distributed areas one major catastrophe will not wipe out the total fund - ie New York / London / Sydney are not going to be wiped out by that onemajor catastrophe. Although there have been some very bad luck cases where such things have happed

and reinsuraners generally only accept about 10% of the total risk of one 'entity' and then share the risk between hundreds of other reinsurers (the long tail )

re: Although there have been some very bad luck cases where such things have happed

 thats the point... its been suggested that the "some very bad luck cases" are not that.... but an underestimate of how likely things far out on the long tail truly are... (that is the distribution is not a standard curve but a power law)..   

re insurers: but the pyramid has been getting narrow at the top has it not? ... at the very top the number is pretty small... something really big has the ability to get the behemoths to merge and the risk of collapse of the whole system climbs.... 

"Prediction is very hard, especially about the future." - Yogi Berra.