(Point(er(r)))

Does Crowdsourcing Solve Big Data Problem?

Posted on May 9, 2013 by Siva

Say there is a system that allows everyone to submit their own ideas for new products and people can vote which of those ideas are best. The top ideas are decided by the entire userbase and not by a few set of individuals. Great idea and will mostly nail down the top ideas right? IMHO, Wrong!

Here is the reason why I think it doesnt work. Now, say there are a million users in this system, the more the better right? Let’s see. Say each day there are 10 ideas submitted. Then all the million people can vote against those 10 ideas and the system is bound to pick the top ideas. That’s not a problem. But instead of 10 ideas, imagine there are 100 ideas submitted. Increase that to 1,000. Increase that to 10,000. You get the picture, it’s simply not possible for anyone to review more than a handful of ideas a day. It’s not the number of votes that a person can cast within a time frame that matters, but the number of ideas that one can really read, understand, digest and comprehend the novelty of the idea and see the bigger picture of how it may have a great potential.

So, what’s the best way to solve this? I don’t think there is a simple answer. One approach is to allow people to have friends and followers so that your friends will read all your ideas and definitely vote. But there are multiple problems with this approach. First, because they are your friends, they may just vote. Second, not everyone has the time and interest to gather a big social network.

All the above I am talking about is regarding a cool company called Quirky. They are definitely pioneers in the crowdsourced product development but I think there is a lot of scope for evolving their platform.

I don’t claim to solve the problem I am mentioning above, but I think the following is likely to work much better. Say, there are 10,000 ideas a day. Instead of displaying all the 10,000 to the million users, display just 10 of the latest per user and make them vote. Based on this, pick the top set of ideas from this list and make them available again for voting the second day as a separate set. As time passes (with the limit of 30 days), fewer and fewer ideas will bubble up and only a small set of ideas are presented to each user. This will make everyone to carefully review the ideas that they are presented, real top ideas will eventually bubble up as they get more exposure by end of the 30 days.

Posted in Uncategorized | Tagged big data, crowd sourcing, quirky.com | Leave a comment

Reduce (the Order) or Map/Reduce?

Posted on March 5, 2013 by Siva

Recently there was an interesting Problem on Project Euler, Problem 415 called Titanic Sets. The problem essentially asks for how many sets are possible from a lattice grid of GxG points such that the sets have only two co-linear points. This is not very hard problem if the grid size is reasonable. But when the grid size is 10^11, then the problem all of a sudden becomes extremely challenging. In some ways this is similar to the big data problem. Everyone has been dealing with data for the last few decades. But the volume of data being gathered in the recent years is what’s making it more challenging.

To cope with this data volume, we now have several Big Data solutions. Any big data solution typically consists of a component to store lots of data and another component to process all that data. Map/Reduce is one such component for processing large volumes of data.

Now coming back to the Euler problem, as of this writing there are 41 solvers after the problem has been there for 15 days. It’s usually unusual because of all the smart people on Project Euler who solve the problems in a couple of hours to just a few days to fill up the top 50 to 100 spots. So, the fastest solvers table is a good indicator of the complexity of a problem. Of course, what’s hard to one may be very easy to another and vice versa. However, this table does serve as a good indicator on average.

When I managed to solve this problem and looked at the private forum, one of the solutions took less than 5 minutes running on a 2000 node cluster! Another PE member estimated that his current algorithm would take 6+ days.

Isn’t it excellent to be able to use a big cluster and get the result in just a few minutes instead of days? Finally we are able to make use of all that silicon that Intel is producing. There is clearly a benefit and knowledge to be gained in learning and programming algorithms for distributed computing and finally be able to actually pull of something like that in practice is highly commendable (I don’t have access to such large clusters, so I must say I envy people who do :)).

In comparison, my C++ code took 1 minute 22 seconds running single threaded on a Core i7 2.8 GHz machine. A few solutions are within the 1 minute rule. And the person who proposed the problem actually has a solution that gives the result within 13 seconds! These algorithms are sub-linear and hence very fast even for the order of 10^11.

Clearly, when one is faced with large computational problems they have choices on how to reduce the computation time. One is to reduce the order of the computation with more and more sophisticated algorithms and the other is to go the distributed computing route and use Map/Reduce or some other similar infrastructure. Of course, not every problem can be solved by reducing the order especially when dealing with large volumes of data where the bottleneck is the data sitting on the disks (as opposed to the Project Euler problem which is based on a large number and not data). In such cases, it’s better to definitely have distributed computing to take advantage of the I/O bandwidth from multiple storage devices as opposed to a single device. Perhaps a hybrid approach can provide the best of both worlds.

However, it should be noted that it’s not always easy to come up with a sophisticated algorithm that reduces the order and second it’s probably even more challenging to code such an algorithm. So, perhaps investing a lot of effort on reducing the order complexity may not give the desired return on investment. It sort of depends on how long it would take to come up with and code the algorithm but also on how many times the algorithm is going to be used (may be a couple of years running every day on a peta-byte data store?). The more it’s going to be used the more you may invest in further optimizing. Perhaps, in reality it’s not going to be develop and deploy. It’s probably a repeated cycle of develop and deploy by learning the systems behavior and tweaking the algorithm to make it even more faster. While it’s easy to throw in more nodes to the clusters and scale and not worry about the cost of continuous refinement of the algorithm, the cost of operating such a large cluster in terms of the data center space, electricity and hardware failures etc should be kept in mind.

In conclusion, dealing with large computational tasks with large volumes of data should make use of a combination of distributed infrastructure such as Map/Reduce and at the same time look at reduce the order of the algorithm as much as possible.

Posted in Uncategorized | Tagged big data, distributed algorithms, map reduce | Leave a comment

Car Performance

Posted on January 12, 2013 by Siva

This is the 4th post related to my car buying experience. The previous one was Buying a new car.

As I mentioned, my first criteria was a Hybrid. Hybrid cars give a good MPG and that’s attractive to some of us car buyers. They give better MPG by doing various different tricks. The most obvious one is to augment the conventional gas engine with a battery that delivers some power. Others are more mostly how these two systems are made to work together which I think is mostly controlled by software.

For regular cars, the city MPG is less than the high-way MPG. This is because of the stop and go pattern in the city and acceleration eats up a lot more gas than once traveling at a constant speed even at higher speeds. For hybrid cars however, the city MPG is usually listed to be more than the high-way. The reason is, when the speed is below 25 MPH or so, some cars are designed to just use the electric power for small distances. That means, there is no gasoline being used at all. Or, when the car stops at a signal or the traffic is very slow, the gas engine can completely shut-off. So, while the total energy used in cities is probably similar to regular cars and more than that on highways, hybrid cars give better MPG in cities using their battery power.

Another trick used by CT 200h is to have Eco mode where the throttle response is altered by making the driver to press the gas really hard. This may be perceived has negative by some. But CT 200h also has Normal and Sports modes where the throttle responses are perceptively better but sacrifice on efficiency.

When I decided to get CT 200h, I was already mentally prepared with the above negative points of slow response because my priority was efficiency. Once I bought the car, I experimented all the 4 modes (EV, Eco, Normal and Sports). My work to home commute is short and with lots of stops and traffic. So, I mostly drive in Eco mode during week days. Weekends, depending on the situation on the high-ways I adjust as required. I actually tried cruising around 68 to 70 MPH in Eco mode and it worked just fine (thought I am not sure if at those constant high-speeds there is much difference between the various modes).

One of the reasons the CT 200h feels slow is the horse power. It has a combined horse power of 134. I say combined because the gas engine just has 98 and together with the battery it gets 134 hp. One of the reasons I was fine with 134 hp was because my previous car also had the same hp. However, I think the combined hp probably still feels less responsive compared to the same hp for a regular car with just gas engine.

Given that there are cars with Horsepower in excess of 500, I wanted to understand what that means to the cars performance. As I understand the HP makes difference in two things

1) The acceleration time of the car
2) The maximum speed of the car

If someone is really into the fast acceleration, then they shouldn’t be buying hybrids or probably get high-end hybrids. But regarding maximum speed, that shouldn’t really be much of a concern. This is because, let’s see how the HP and velocity are related. As per Wikipedia article on drag,

$P_d = \mathbf{F}_d \cdot \mathbf{v} = \tfrac12 \rho v^3 A C_d$

That is, power is proportional to cube of the velocity. Assuming all other factors (drag coefficient and the cross-sectional area) are the same for two different cars,

HP1/HP2 = (v1/v2)^3

So, if a car with 134 HP (like the CT 200h) has a max speed of 80 MPH (it’s actually above 100, but let’s not use extreme values), then a car with 350 HP will have a max speed of 110 MPH and a car with 500 HP will have a max speed of 124 MPH. All these numbers may not mean much but just realize that HP is not linearly proportional to speed. Given that the speed limits on high-ways are only 65 to at most 75 MPH, it probably doesn’t matter. Unless, of course if you want to tow stuff with your car.

With so many cars and options, it’s very difficult decide on a car. The best way to decide is to first decide on a few key important features and of course the budget. Then the choices narrow down drastically and with a bit more of online research, it should be possible to pick up to three cars. Test driving them should narrow it down to 1.

Posted in Uncategorized | Tagged ct 200h, hyrbid cars | Leave a comment

Buying A New Car

Posted on January 10, 2013 by Siva

As I mentioned in my previous post, I recently bought a new car. I am one of those people who is not really good at haggling. Infact, in this day of the Internet age (I think we need to rewrite the 3-age system to 4-age system :)) who really needs to bargain for the best price? We typically search Amazon, Google and find the best offer and done with it. Unfortunately, that’s not the case with buying cars.

If we can buy a build to order high-end computer over the Internet and get it delivered in less than 5 business days, why can’t that be done with the cars? It’s valid question with an (not necessarily valid) answer. Typically the states laws don’t allow manufacturers to directly sell their cars to the buyers. They have to go via the dealerships. See this article Economic Effects of State Bans on Direct Manufacturer Sales to Car Buyers.

Buying a car requires doing two researches. One is to narrow down to one or two models of cars. Then, figuring out what’s the price that the cars can be purchased for in your local market. The first one is relatively easy especially if you fix a few parameters. In my case, “hybrid” is one of the fixed parameters. Of course, I also had a rough price range in my mind but I was kind of flexible. Though, I am not into vans and SUVs. So, that left me with just hybrid cars and having owned a Toyota Camry, and Prius being the most sought after hybrid and one of the best mileage vehicles, I was initially looking into it. Based on discussions with a few friends and colleagues, I decided to go with the Lexus CT 200h. It was kind of a quick decision. Though, I did a lot of research online about this car and I know it’s pros and cons and I didn’t even had to do a test drive to make the decision. The online material including the brochures on the Lexus website, forums and reviews and Youtube videos were sufficient in my case.

So, I had to focus more on the second part of the research, the price to be paid. Luckily, for this specific car, there are two main threads online, one on clublexus.com and the other on ct200hforum.com where people discussed the prices they have been quoted or paid. Given the various options available for the cars, it’s not exactly an apples to apples comparison. Further given that during November, December time both 2012 and 2013 models were selling actively, it was not easy to figure out what the right price is. But all that research will mentally prepare you to figure out the maximum you are ready to go up to.

When I started my research, a friend of mine said that Costco has an Auto program that can provide a better price. So I went to their website and I should say it’s pretty impressive with all the recent models and their options listed out and they show both the MSRP and invoice price. The invoice showed was about $2k less than MSRP and the website said that the final offer will actually be even lesser than that. I was excited. But then I wanted to find out if there were any other similar programs that can provide haggle-free best price. A few websites do have such program and in fact TrueCar.com even provides a nice graph of what people paid for the chosen car with options. Below is the graph of a sample model I was looking at (the one I bought is a bit different).

We all know there is no free lunch. So, how do some of these websites make money? I came to know via some forum threads that these guys actually get some commission back from the dealers that could be as much as $500 or more (may be depends on the amount of business they bring in). Some suggested that if one can directly go and negotiate, since the dealer doesn’t have to give out the $500 to yet another middleman, that can directly translate into savings for you.

Based on all this research, I decided to give a try of negotiating the price myself but rather than doing it in person and under pressure, I decided to do it via the online sales channel. Essentially, these days most dealers have a separate online sales channel. Some of the dealers actually are combining the fleet sales and online sales into one (will explain why in a bit).

First I submitted to one dealer. Got a reply the next day sending me a link to their “True price”. The email said

“At [dealer-name] we offer a TRUE Price. What this means to you is that we constantly shop our market and competitors, and provide you a competitive price up front – without the hassle of negotiation. Our TRUE Price saves you time and provides a more pleasant purchasing experience.”

When I clicked on that link, it just took me to their regular web page that lists all the cars and their MSRPs. Not a good impression, but may be let’s give it a benefit of doubt that the sales person linked to the wrong page. But in reality, they are just tracking me. The dealer was using eleadtrack.net. I politely replied back saying that I was just taken to the page with MSRPs which I could get from the lexus website itself. Then I got a reply with 4 different configurations and asking me which one I was interested in. I replied back with the one I was interested in. There was no reply during the weekend and then I get an email stating that there were about 20 cars in stock and if I wanted to come and checkit out first. Well, if the “True price” was supposed to offer me no-haggle price that too by constantly shopping the market and competitors, why isn’t the “true price” revealed? The sales person even called and left me a voice message. But that was the last of my interaction with them.

I then submitted my interest to another dealer. I got a response and these guys used responselogix.com to track the leads. The good thing was, in the very first email, I was given a configuration with MSRP and was given “Your Special [dealer-name] Price”. It was $1500 below the MSRP. While it wasn’t the best offer, at least this guy didn’t try to do tricks like the other guy. I exchanged multiple emails with this sales person and narrowed down on price, configuration but we had an issue with the interior color. In this whole process, he even sent me something called “Vehicle Inquiry Report” which is something I saw for the first time. It gives the MSRP and Dealer prices. He was giving me an offer $1K below invoice. However, there was a line item called “TDA” which upon research came to know it stands for “Toyota Dealer Advertisement” fee. Basically, the cost of advertising is passed on to the dealers who try to pass it on to the buyers. If a buyer is buying a car at MSRP or may be even close to Invoice, dealers may not explicitly itemize this cost but if you are bargaining well below invoice price, be prepared to see this as one of the line items which dealers might use to justify an additional fee of about $500.

I thought my negotiation and interaction with this new dealer was going smoothly but towards end of November, he said he doesn’t have the interior color I was looking for and he just has the deal that he sent me. He also said I had to close it by November to get the same price. I then thought it’s probably a pressure tactic. Since I wasn’t in a hurry to buy the car, I just waited out. Also, end of November it was raining and on the last day at 7pm he sent me an email that he was ready to go with the price I offered (which was $500 below what he was trying to get) but I had to close it the same day. I was waiting for him to reach out to me till 6pm so that I could directly go from work but since it was late into night and raining, I just gave up. He probably thought I wasn’t a serious potential buyer. I think he was trying to conn me because the Lexus website showed that cars with my color choice were available within the dealership area.

Then came December and then I thought perhaps negotiating with car dealers is not my cup of tee and probably it’s best I go with the Costco program. I then configured and submitted for quotes. Apparently, the way this was supposed to work is, every dealer would have some fleet car sales persons who typically just push large volume of cars to corporate accounts and there is not so much for them to spend time with the buyer and so the deals are typically much better. And Costco is supposed to have negotiated with some of these dealers and you take a print out of your quote and go to the dealer and meet the contact and then they would provide their special price. Some suggested getting the price on the phone rather than going, some said the dealers could do bait and switch tactics. Anyway, when I submitted it, nothing really happened. I didn’t get any email from any dealers. Also, the dealers Costco suggested after completing the form were farther than my nearest dealer. I wanted a close by dealer.

It was mid-December and I was back to square one. I submitted for a quote to the dealer with whom I was negotiating hoping I would be put in touch with another sales person. But the auto response was from the same guy with whom I interacted. So, rather than submitting via the online form, I then picked another online sales agent and sent an email directly to him. He sent me a configuration along with VIN number and gave me a quote that’s $1500 more than the price I was successfully able to negotiate but again offered the same color that I wasn’t interested in. Then I was like “damn, I lost the color and I lost the price”. But upon talking to the person and explaining him that I got a better offer for the same configuration from their dealership he agreed to honor the same price. Hurray, but still offering the same colors and not the ones I was interested in.

I was doing more research and then when I typed in the VIN number I got a bunch of websites giving me details about the same car. That’s when I came to know that there are websites rankingsandreviews.com that actually list the dealers’ inventory. Then I did a search and looked at all the various CT 200h configurations offered by that dealer. And there wasn’t a single car with the color choices I had in mind. That’s when I realized that these guys are trying to get me buy the other one because they truly don’t have what I was interested in. So, that night I decided to go ahead with the color choice they offered. In retrospect, that combination is actually pretty good. It’s Matador Red Mica and Caramel interior (instead of black interior I was looking for).

During my research, there was an article about fleet car sales and how celebrities actually buy cars without the lengthy negotiation process. Reading that gave me confidence that I was on the right track. I finally managed to get my car at a price that’s almost lowest (I think I could have gone down $500 more) and it beats the best price on the above curve by a wide margin.

Finally, on Friday afternoon I called the sales person, confirmed the price once again, setup appointment and left work a little early and visited the dealer. As soon as I entered the regular sales person greeted me, I told I was there to meet so and so sales person and I was asked to wait and in a few minutes the guy I was dealing with came and we did a test drive and then I was sitting in his office and signing the papers. Due to Lexus’s December to Remember deal I even got 0.9% APR. I have even put a portion of my down payment on my credit card (and a colleague was surprised I could do it).

All in all, a no-stress experience and my research for about a month paid of well. Obviously time is money and perhaps the additional $500 to $1000 I saved may or may not be worth the effort for others. I don’t mean to advice that this is the best route for everyone. It’s still OK to go with one of the other channels but do realize that it’s possible to negotiate below what those channels suggest (inspite a cut to them). The way I see it is, it’s a learning experience about the cars, about the various Internet websites regarding cars and also reviews, opinions and experiences of various people. That to me is priceless, or time well spent.

Posted in Uncategorized | Tagged ct 200h, hyrbid cars | 1 Comment

CT 200h

Posted on January 10, 2013 by Siva

About two months back I wrote about hybrid car savings. That was around the time I actually thought of getting a new car and was researching on getting a regular car vs a hybrid. I finally ended up getting the Lexus CT 200h, a compact hybrid luxury hatchback car.

When it comes to cars, I am a noob. This is my second car and first was Toyota Camry which I had for 14+ yrs! Probably too long to having the same car, especially here in America. But actually, when I was in India last summer, I met a guy who traded his old car after 5 yrs and got a new one, that too a diesel car. His point was that he could still get a decent price for his previous car and he could as well upgrade.

When I first got my car in 1998, the Internet was there but probably the usage of the internet to buy cars is a lot more now than at that time. At that time, I remember going the traditional route of calling a dealer, setting up an appointment, going along with a friend and getting a car with very little negotiation (I don’t remember how much research I did and what kind of negotiation I did, or may be left it to the friend, do remember using KBB).

Things have changed drastically, at least to me and in the next few posts, I want to write a couple of things about my pre and post car buying experience. Why? Just for fun. Or may be to check my notes in another 14 yrs :).

My new car is a Lexus 2013 CT 200h, Matador Red Mica with Caramel interior.

In my next post I am going to talk about my pre car purchase experience.

Posted in Uncategorized | Tagged ct 200h | 1 Comment

The truth about the truth tables

Posted on December 7, 2012 by Siva

In my previous post Learning Mathematics I mentioned how I wished I was told some practical uses of advanced mathematical entities such as polynomials and matrices. In this section I am going to talk about truth tables. Actually, I am not going to explain what they are because there is a lot of material for that. I am going to talk about what they can be used for in a real-life use case.

Say you have 2 benches and each can seat 3 students. Call these 6 students as s1, s2, s3, s4, s5 and s6 (obviously in real life they have their own unique names and identities). Let’s say all the odd numbered students are girls and even numbered students are boys. If the goal is to just find out how many ways to seat these students, we just need the knowledge of permutations and combinations (the answer is 6!).

However, what if there are constraints on the seating? Say we have the following constraints.

No two girls should sit next to each other
Student s1 doesn’t want to sit next to s4
Student s2 doesn’t want to sit in the middle

You can think of a few more constraints if you wish. Now, how many ways are there to seat the students? This is going to get more complicated. With more number of students and more constraints it’s going to get very ugly. These constraints can be expressed as Boolean functions. First, we need to be able to distinguish between student s1 sitting on the first seat with student s2 sitting. Which means, we also need to label the seats. Let’s call them c1 through c6. If we now thing of each of these 12 variables (s1 to s6 and c1 to c6) as Boolean operations, then ith student sitting on jth seat can be considered as (si and cj). For example, (s1 and c4) indicates that s1 is sitting on c4. If you want to say s2 or s5 can sit on seat c3, it can be written as (c3 and (s2 or s5)). Each one is an individual constraint by itself. Now, the solution is the assignment of True/False to each of these 12 variables such that all the individual constraints listed in the problem are satisfied.

There are actually 3 types of results one might be interested in such problems expressed as constraints. They are

1) Just counting how many ways it’s possible to do

2) What good is just being able to say there are 12,197,407 ways of seating the students for this year’s winter classes without knowing how exactly they can sit? So, we may also be interested in one or all ways of actually assigning the variables values that satisfies the constraints

3) Now, let’s say there is a cost for seating a student in a seat and the cost depends on both the student and the seat. So, C(si,cj) is the cost and what if our goal is to find not just any seating but the one that minimizes the cost? So, we are all of a sudden looking at combinatorial optimization problems.

There are different techniques to solve the problems. Dynamic Programming is one of the most common techniques for solving combinatorial problems where a problem can be decomposed into several sub-problems and the same sub problem repeats several times and it has the same outcome no matter how we reached to that sub-problem.

Dynamic Programming is usually much faster than other straight-forward techniques. However, there are times even dynamic programming is slower, especially if we are trying to enumerate all the solutions to find the optimal one. Also, if the sub-problem space is very large, we can also run into memory problems.

I recently ran into a similar problem. The problem asked for assigning numbers to a set of cells in the grid with some constraints and find the optimal assignment. Say the optimal value is X. Anyone who got found an assignment that gives at least X-3 is considered to have done a good job but the closer one can get to X, the better. My initial dynamic programming based solution gave X-3 in less than 2 minutes and X-2 in about 5 hours. So, I knew there is no way I could get to X with my existing algorithm.

I also knew that there are these things called BDDs but didn’t exactly know what they are and how to solve a problem using them (but I knew they can solve the above type of problems). So, I finally took some time to learn about BDDs and solve the problem. Based on my knowledge of the problem, I have added 4 more additional constraints that I know would help speed up for the optimal structure. With that, the answer for X was found in less than 13 seconds!

Lessons learned for me are

BDDs are damn fast
It’s also easy to express individual constraints with Boolean variables than actually coding the same for dynamic programming
Especially so if we need to refer back to the previous states of the DP
If the variables are numeric, it’s a bit challenging but not impossible to use BDDs although the number of variables will explore (looks like there are ADDs (Algebraic DDs)and FDDs (Finite Domain DDs) that makes it less challenging
It’s difficult to debug the logic since at present using BDDs with existing libraries is like writing assembly language (that is, we have to think in terms of the underlying building blocks of the library rather than at a higher level closer to the problem).

I used BuDDy BDD library to solve the problem. It’s written in C++ and very fast compared to another Python library I first tried to use. My current experience with BDDs is a) I can start thinking in BDDs to solve constraint problems b) I know how to use atleast one BDD library. Admittedly, what I still don’t know is all the theory behind BDDs and how they actually work fast. As time permits, I will spend time exploring some of that.

There is very limited information on the internet about using BDDs and solving specific problems. Most of them just hash out the theory behind BDDs which to a normal student can be easily daunting. But getting a feel for a problem, how it can be represented as BDD and using a specific library to solve the problem like I did can come a long way to clear things up or at least create the desire to explore the concepts deeper (I mean, wouldn’t you be curious if you managed to solve a problem in 15 seconds runtime which couldn’t be solved in 5 hrs and still not close to the optimal solution?).

What we need in teaching math or any advanced concept for that matter is to first create the desire to learn something with concrete examples and benefits.

Posted in Uncategorized | Tagged advanced math, algorithms, BDD, education | Leave a comment

Should CEOs be following their company twitter account?

Posted on November 26, 2012 by Siva

In this 8th post on Social Network Intelligence, I want to ask an interesting question. Should CEOs of the companies spend time following their corporate twitter accounts? Does it add any value to the company? Is it going to take a lot of effort?

One of the information that can be provided on the twitter profile is the url. In this post, we are going to dissect the followers data by the domain name of the url given in the followers’ profiles.

Let’s look at this information for salesforce and workday.

It can be noticed in both cases that their (co)CEOs are there on twitter and they are following their corporate accounts. In fact, among the list of people following from their respective companies, they seem to be the ones with maximum followers. Which is probably expected, after all they are the top leaders/visionaries that other want to follow. How else could you explain 34K followers and 0 friends and 1 tweet for a CEO of one of the top software companies? Given that both Salesforce and Workday are successful multi-billion dollar companies, I think it’s a good idea for the CEOs and other top management of any company to follow their corporate social network accounts.

Based on my analysis, there are companies whose CEOs are not following their corporate twitter accounts. In a few cases they were actually following but they haven’t provided the url on their profile or it points to their own website or a linkedin profile page. A word of advice for top management of companies with social network presence is to use your corporate website in your profiles to increase visibility to your company.

An interesting observation from analyzing twitter followers by their url domain is, it’s easy to figure out who are the people from your competition that’s following you. Who among your vendors, customers and other trading partners and organizations are following you. This is valuable information that for example can help in identifying potential leads, potential hires and even discover individual influencers within your domain that can perhaps help you market your products and services if you reach out to them and show them what you have.

Posted in Uncategorized | Tagged social network intelligence, twitter analytics | Leave a comment

Deducing global presence from social network – part 2

Posted on November 23, 2012 by Siva

In earlier post we talked about how the language of the twitter followers can be used to deduce global presence. We also discussed how it will not give a complete picture because many people may give English as their language in the profile even though they are located in different parts of the world. Obviously, another way to check the global presence is the location information in the profiles. However, note that the location information is not mandatory and also it is not a fixed list. So, it’s hard to get a good picture based on this as well. Also, location is a free form text that’s filled with city names, state names, country names making it extremely difficult to analyze without having a good data quality solution.

In addition to the location, there is also a field for timezone. Timezone, like language, is a based on a discrete list of values. While it seems to be mandatory right now, it probably was option in the past and hence there are many twitter users whose timezone info is blank.

Let’s look at the distribution of twitter followers by location for the three cloud companies which we have been so far analyzing.

In the above chart for salesforce, about 25% didn’t provide any location information. About 45% of the total (and 60% of those who provided info) are from within the US. UK is the next highest. Not knowing much about Quito I just wondered why there were so many followers from there for twitter and noticed that at least a few top users with maximum followers seem to actually belong to the US based on the location they specified. Then why their timezone is set as Quito? It is mostly because Quito is GMT-5.00 just like the East Coast USA. It’s likely that people just picked one of the several GMT-5.00 options available. Perhaps doing analysis based on Twitter Timezone field has to be taken with a pinch of salt if we want to make use of the location part of the timezone instead of just the GMT offset. May be the timezone picker should be based on a map.

Even more number of Workday users have no timezone info provided. We also see Quito coming right after London.

As discussed in part 1 of this post, Oracle Cloud seems to have more globally distributed following.

Posted in Uncategorized | Tagged social network intelligence, twitter analytics | 1 Comment

Stacking Coupons On Top Of Existing Sales

Posted on November 23, 2012 by Siva

When buying stuff online and going through the checkout process, on the billing section there is typically a Promo Code/Coupon box. I think this prompts people to start wondering if they are missing some deals. At least it does to me. I was looking for some Thanks Giving sales deals and ended up on RalphLauren.com and decided on buying a couple of t-shirts for me and my son. They did have some good deals. First they have reduced prices and on top they were giving some 25% off for the Thanks Giving. I was happy with the prices I got and was doing the checkout. During the checkout process, I found the promo code box and as I mentioned above I promptly tried to search for those promo codes. I found some coupon on retailmenot.com and I entered that and clicked the next. I was taken to the order review page and I just had to click to finish the order. That’s when I noticed that my order price has actually gone up! I almost clicked the confirmation button but luckily I didn’t. I thought my order was less before I entered the coupon code but wasn’t sure. So, I closed the browser and restarted the process. Yes, the order was indeed less.

So, it’s important to be careful whether promo codes are stacked up or not. Ideally, the software used by RalphLauren.com should have been designed to pick the promo code that gives the best discount (or ignore a promo code against a sale event if the later gives better deal) but that wasn’t the case. Next time when you are shopping online, make sure to pay attention to this.

Posted in Uncategorized | Tagged software design | Leave a comment