Flipkart - a good story in making

I had a bad online book purchasing experience with  firstandsecond.com, an online bookstore in India. I ordered a book, which was not delivered. I had another bad online purchase experience in buying a flower bouquet after which I applied brakes to online purchases.

After several months of keeping myself away from online purchases, recently, I purchased few books on Flipkart (http://www.flipkart.com), a Bangalore based startup. The first order was shipped the next day. A day later, I had the books in my hand. I was impressed.

I purchased another set of books for my daughter's birthday. There were only 4 days between the purchase date and the purchase date. The book detail page indicated 7+ days of delivery. I contacted the customer service to see if Flipkart could expedite the shipping of these books. Flipkart did not promise the result. However, they were considerate enough to try and deliver the books in two installments. The first set of books arrived in 3 days time. The second (and the last) installment the next day. This is a very pleasant experience, which makes me a loyal customer of this budding site. I hope they maintain their standards. Thank you Flipkart.

Are people in Bengaluru so insensitive to Ambulance Siren?

Today, near the Marathalli outer ring road signal, I felt really bad to see an Ambulance stuck in traffic. All drivers around the ambulance were looking at the vehicle that was emitting emergency siren. None of them tried to move away to give way to this Ambulance. The traffic signal is just 50 meters away and there were few policemen at the signal. None of these policemen made attempts to clear the traffic to get the Ambulance moving.

Even after the green signal, and the traffic started moving, several vehicles tried to move ahead of the Ambulance from the sides.

Did we loose the concept of 'emergency'? Why can't we understand that we should stop on a side and give way to emergency vehicles?

Public Service Ads by Google

I was going through this website http://www.tecrec.in. Usually, I ignore the advertisements on any page. However, this advertisement caught my attention. The advertisement was about a public service organization http://www.grameenfoundation.org, and the bottom-right of the ad reads "public service ads by Google". I was happy to see organizations trying to promote something with an intent, which is other than money making.

SEM by startups - can they win the war!

Several startups use SEM for marketing. Is this really a good idea? The answer is yes and no.

It is a 'yes' because people get instant exposure to these sites based on the bid words. Traffic to relatively unknown websites (like startups) are very welcome because website visitors are always the revenue generators.

Can this 'yes' transform into a 'no'? The answer is 'yes'.

Let us go back and look at SEM on Google. A startup xyz.com bids on keyword(s) and gets itself to the top rank for those keyword(s). Similarly, a very well known company pqr.com, which also has a good SEM click history bids on the same keyword set. The bid value of xyz.com will be higher than pqr.com. When the user searches using the keyword, Google will show these first xyz.com and then pqr.com in the advertising box. Let us consider the following scenarios:

1. xyz.com has a better ad content. The user is impressed and clicks on the advertisement. xyz.com is charged for this click. Many sites have very high bounce rates. The cause for the bounce off could be because of the site content, the site organization and flow, and also because of the trust level of the user on that site (xyz.com). Majority of times this click results in a loss to xyz.com. However, the SEM optimization engine thinks this as a success for xyz.com and increases its credibility/trust factor. A good vote in favor of xyz.com. If b% of users convert then the effectiveness of the amount spend is b%


2. The user does not click on xyz.com even though it is at a higher rank but, this user clicks on pqr.com because pqr.com is a known website. There are publications that indicate the effectiveness of having ads positioning at positions 2 or 3. At these positions, statistics indicate higher click through rate. This click for pqr.com (and not for xyz.com) reduces the trust factor for xyz.com.

For a relatively unknown xyz.com the probability of 2 is much higher than 1. Thus, the SEM optimization engine is likely to penalize xyz.com because it did not receive the clicks even though it was positioned higher. Now, if xyz.com wants to maintain its ranking, it needs to bid higher. Further clicks not-in-favor of xyz.com will cause xyz.com to bid higher for the same set of keywords. The same is not true for pqr.com, which has already established its trust factor. Also, the b% effectiveness due to bounce rate also adds to the SEM expenses of xyz.com.  Thus, in SEM space, startups rarely win a war against well established companies.

What are the alternatives?

One alternative is to reduce the cost of SEM while keeping the effectiveness of keywords higher. This is something SEM experts are known to achieve. These experts are known to find keywords that are effective and cost low (mainly due to lack of competition). However, their search volume is also very low. A startup has to identify and dynamically modify its keyword set on which it bids to increase visitors through SEM program. This is a search through a long tail of rare search keywords, and user typography errors. Can this search go on for ever?

There is this other alternative some companies had adopted to get to their target user base. Instead of making users come to their website, these companies started to go to the users by identifying areas where the users are more likely to visit. One such approach is to create applications on social networking sites like Facebook, Orkut, and MySpace. Partially, this was a step in right direction. Users, who rarely visit sites like xyz.com, would visit/add the application on facebook/Orkut/... with some curiosity, and after some clever marketing strategies by the application developers. After spending considerable amount of time on this application, these users get used to the content of the xyz application, which establishes a trust factor. Now, getting these users on social networking sites to xyz.com is another big hurdle. Note: I did not see an application that had converted the entire user base to its independent website. Thus, this step is 'partially' in the right direction. It only established the brand of the application. It would have been a completely successful approach if it had made people come to xyz.com.

Disclaimer: xyz.com and pqr.com were used just for explanation. These pointers are not to the actual websites (I did not check  if real websites exist and what they do). Also, my thrust on the trust factor being one of the variables in computing the bid value/ranking for a site by Google is based on some discussions I've read over the internet about Google's way of optimizing its SEM systems.

Netflix Prize Contest - recommendations

Netflix prize contest is about designing a recommendation system that can do 10% better than Netflix's own reference system. The grand price was set to $1 million. It had the buzz factor when it was launched. Very soon, people had realized that it is very difficult to achieve the 10% improvement target that was set by Netflix.

People tried several approaches. One of the very popular initial approaches was "Simon's Approach", which was almost similar to the LMS algorithm.

Researchers are trying several methods to grain incremental gain in the score. They are spending significant energy, time, and effort in achieving the target. Many of these researchers are combining their individual approaches to obtain further improvements.

Yes, after trying several approaches and tweaking the algorithm parameters, these researchers might reach the ultimate goal of 10% improvement. However, are all these approaches practical to implement? Even if these could be implemented very efficiently, do these really give good movie recommendations for the real user? Finally, can these approaches be used for recommendations in other domains?

This research has brought many teams to collaborate with each other. It has given graduate students a research topic to work on and get their thesis/publications. More importantly, this research has given a common ground for researchers to compare individual algorithms.

Finally, I'm not sure how much Netflix benefited from the practical use of the submitted algorithms and approaches. It has definitely benefited in terms of getting the right to use several algorithms in their recommendation system, and getting hundreds of researchers to work on a problem by paying a small amount (per annum per researcher).

Blogging begins !!!

Initially, I was reluctant to put any of my thoughts online. Now, don't know why, I'm thinking of putting some thoughts out. Hopefully, this log accumulates enough information for use by the rest of the world.