Probability and statistics blog

I blurred out my personal information, but otherwise the check shown above is exactly as I received it. Yahoo! really did send me a check for ZERO DOLLARS 00 CENTS. Obviously, mailing out checks for nothing is wasteful and pointless (shall I try cashing it? Will the bank “credit” me for nothing?). It could be that this check isn’t even “legal”, if I’m to believe the Best Answer by Yahoo!’s crack team of expert volunteers. But instead of making this a “Ha! Ha! Big Company Does Something Stupid” post, I thought I’d dig a little deeper into the story behind the check.

Until recently, Yahoo! had an advertising program similar to Google’s Adsense. Webmasters could show ads from Yahoo!’s network and share in the profits when people clicked on the links. Although basically similar, Yahoo!’s version was much more restrictive: fewer websites were accepted into their network and webmasters could only display ads to visitors from the United States. I can understand why Yahoo! would want to limit payment to clicks coming from a single (huge and wealthy) country, but the way they implemented this requirement struck me as odd. It was the webmaster’s responsibility to make sure no one from outside the US saw their ads. This meant that I had to install Geo-IP software on my server, which maps the unique numerical identifiers of visitors to their country of origin. Installing this software is a pain, and you have to update it regularly with the latest free or paid versions of the database, or else the information becomes inaccurate. Yahoo!, with all its infrastructure and CPU power and economies of scale, was making thousands of individual webmasters perform a task best handled by a giant IT company, like, say, Yahoo!

The other strange, and occasionally frustrating, thing about Yahoo!’s program was the high volatility in payment amounts. On average, the earnings from their ad network were good, but I would see days when a website made $50 and others when it made $2.11, sometimes from the same number of visitors or even the same number of clicks. Variability is built in to programs like this, but I found Yahoo!’s results to be way more inconsistent that Google’s or even some of the much smaller companies with their own advertising networks. You might think that only the average matters, but high volatility leads to poorer decision making. Imagine the following two scenarios: in one, you see the revenue from a advertising program slowly and steadily decline over the course of months. Time to find a different program. In the other scenario, payouts vary widely from month to month. Some are fantastic, some very poor. How can you tell when to leave the program, and when to expand your use of it? There are ways to measure non-obvious trends with time series analysis. But no matter which tool you use, the more “noise” in your data stream, the worse your predictions will be.

High volatility in Yahoo!’s payouts was so prolonged that it couldn’t have been just a matter of growing pains. Either no-one at Yahoo! was looking into payment variability at the level of individual users (it’s possible that the total revenue generated by their program, summed across all users, was much more stable), or they didn’t see why this was a problem.

Finally, we have the issue of the worthless check. When Yahoo! closed their program, they had to send out final checks, even to those users with balances below the regular minimum threshold (most programs require $50 or so in earnings before they send you a check). One possibility is that Yahoo!’s algorithm never looked to see if the amount owed was positive before cutting a check. That would be really dumb. Another, slightly-more-favorable explanation is that they really owed me 0.5 cents, so their algorithm saw this as a positive balance. It looked like I should be getting a check, but then Yahoo!’s check-printing software rounded down to the nearest penny and I end up with $0.00. Either way, I see this as a final indicator that Yahoo! just isn’t thinking very hard about their data. By contrast, Google and other companies seem to understand that their business depends on managing data with a huge amount of human intelligence and understanding. Data is dumb, Yahoo! needs to be smarter.

Tags: algorithms, yahoo

This entry was posted on Wednesday, June 23rd, 2010 at 9:32 am and is filed under feature, stats. You can follow any comments to this entry through the RSS 2.0 feed. Both comments and pings are currently closed.