Archive for February, 2007
This is an installment of the ongoing “Metrics Series.”
Like many other folks, I think Alexa data is just plain wrong.
When I compare Boxxet’s numbers (using two totally different methods) to Alexa’s numbers, there is absolutely NO correlation. As many others have reported, Alexa is heavily skewed with tech users and it shows. One quick example, Boxxet’s most recent biggest day, according to Alexa, was our launch date (Jan 16, 2007). Yes, we were pleased with that day, and we got lots of attention from the tech bloggers. However, Jan 16 generated only 1/3 the visits of our actual best day (Feb 12). And Alexa reports Feb 12 to be 1/7 of Jan 16. Alexa is off by 21x.
Despite my frustration of our incorrect Alexa numbers, and the near-daily exasperation I have with the many people who rely on it to make snap judgements (including me in the past), I chose not to blog about it. Also, there may nothing better out there and I do like Amazon.
But this weekend, I saw that Alexa had an upgrade to their service and in their blog announcement, they said:
I expect that this new traffic data will be particularly useful to potential advertisers. It will help to size up an audience before making an ad buy.
Until now, I wasn’t sure why Alexa was doing what it was doing. I figure it was some sort of giving back to the community. But with this statement, Alexa sees itself has a useful business tool. PLEASE STOP, YOU ARE NOT!
Bad data is more than an inconvenience, it can be downright hurtful to a business. Great web sites may be shut out of advertising, business deals may never happen because a company did not pass a fictional Alexa test, potential employees may choose not to join a company b/c of its Alexa numbers, adversaries may aim their resources at the wrong competitors.
Amazon/Alexa, you are encouraging people to make business decisions using your very suspect data. I believe you have the resources to make Alexa site traffic numbers great. Please please do so.
Filed under: Metrics Series
February 19th, 2007
You Mon Tsang
This is an installment of the ongoing “Metrics Series.”
Boxxet spent a long time in a closed beta, longer than we expected. Some reasons were normal: stamping out bugs, making sure the application scales, and determining weak areas that need improvement or protection. More importantly, after our first wave of users of the closed beta, we felt that we needed to add more automation to the system.
There are, of course, pluses to an open beta (get quick feedback from the real world on what works or does not) and a closed beta (you make sure your first impression to the masses is a good one). We felt we did the right thing by staying in a closed beta. It allowed us to make large changes without worry about massive disruption, shore up our infrastructure and experiment with new algorithms at will.
However, the downsides are significant and I’d like to share some metrics for others who need to make such a decision. Perhaps the numbers will help folks make a more educated determination about how to proceed with a beta.
When we announced Boxxet was going into beta, we got slashdotted. We estimate that approximately 40% of the people who came to our front page signed up for the beta. Frankly, this is more than I expected. However, upon letting people know that they have been given access to the beta (within a few weeks), 60% returned to take a look. In some cases, we waited a few months before we gave people access and the drop off was even greater than 60%.
Of the people who came to our front door, 40% signed up and 60% returned when we emailed them. That means 76% of the people who stop by to a closed beta did not immediately return to see Boxxet when we let them in.
Of course, if we did not effectively serve these 76% (slow pages, crashed servers), we may have done something worse: lose them forever.
But this is a choice and I wanted to make sure that others understood the numbers on both sides of the equation. As we all know, the cost of user acquisition is quite high and not giving 76% a first impression hurts.
Filed under: Boxxet, Metrics Series
February 12th, 2007
You Mon Tsang
This is an installment of the ongoing “Metrics Series.”
A few weeks after we opened up Boxxet to search engines (about 7 weeks ago), I was worried. Traffic from Google started to come in at a good clip, but traffic from Yahoo, MSN and Ask was miniscule to non-existent. Given the latest search engine stats, I wanted non-Google traffic to give us roughly half of what Google was giving us. Here’s what Hitwise estimates as search engine share in Dec 2006.
We believe we went through the appropriate steps to let search engines know about us and told them through available APIs and submissions that we were ready to be crawled. But Google completely dominated the organic search traffic to Boxxet. We talked to friends at the search engine companies, scoured for tips and decided that maybe time will solve the problem.
But time has not brought change and at Boxxet, Google continues to bring in over 90% of all organic search engine traffic.
But recently, I have also noticed that other high-profile sites have similar organic search share. For instance, the top 11 referrals at Techcrunch in Dec 2006 look like:
1. google[organic] 391,034
2. (direct)[(none)] 326,796
3. digg.com[referral] 192,774
4. google.com[referral] 78,960
5. news.bbc.co.uk[referral] 46,621
6. netvibes.com[referral] 32,444
7. techmeme.com[referral] 25,561
8. stumbleupon.com[referral] 22,294
9. reddit.com[referral] 22,035
10. my.yahoo.com[referral] 19,643
11. techcrunch.com[referral] 18,869
On this list, there is no other organic search site besides Google, so I will make the following assumption:
12. yahoo [organic] 18,000
13. Others search engines: 18,000
So Techcrunch’s organic search share looks very similar to Boxxet’s.
Over at Search Engine Land, the folks there shared their top 25 referrals from Dec 2006 and Google is the only organic search site! So here again, I estimate Yahoo and the others as follows:
Google 6,172
Yahoo* 280
Others* 280
Search Engine Land also looks the same as Boxxet (as well as Techcrunch).
The organic search share at Boxxet, Techcrunch and Search Engine Land are virtually identical and heavily skewed towards Google. So the next step in solving this supposed anomaly is looking for commonalities.
- Boxxet and Search Engine Land are pretty new sites so perhaps the non-Google crawlers haven’t figured us out yet. On the other hand, Techcrunch has been around for almost two years.
- Techcrunch and Search Engine Land are tech sites, but Boxxet is focused mainly on entertainment and sports.
- Techcrunch and Search Engine Land use blogging software (looks like Wordpress and Movable Type). Boxxet has its own presentation engine.
- Only thing that we share is Google Analytics, but let’s pray that is impartial.
Bottom line, I am starting to draw a heart-stopping conclusion: Google’s search share is > 90%. I am open to any other thoughts or theories, but I am preparing for this new world. Rich Skrenta has a very thoughtful take on the era of Google dominance and that was assuming 70% dominance!
When a well-known and well-branded site like Techcrunch gets more people coming in through a Google search than by typing in “www.techcrunch.com,” I get a chill up my spine. Will future successes on the Web be determined by a bot originating from an office park in Mountain View?
Filed under: Boxxet, Metrics Series
February 6th, 2007
You Mon Tsang
As many of you know, Boxxet uses a highly-rated photo from a Boxxet (like, say, the iPod Boxxet) and uses it for a banner. Sometimes, the banner looks terrible. Sometimes, it looks great. I was on Boxxet late last night when this one came up. It made me smile.
Filed under: Boxxet
February 5th, 2007
You Mon Tsang
