You are viewing dmorr

Poker, politics, philosophy
why is Google crushing Yahoo? 
26th-Oct-2007 02:12 pm
poker
[Note: much of this post is talking about the pre-Panama world for Yahoo!. The whole point of Panama is that it will close this gap, but that it will take time. If my analysis is right, then Yahoo! is doing exactly what it needs to do to close the gap. I should have made that more clear in the original post.]

jgische asked the other day why Google is doing so much better than Yahoo!. The answer is a little subtle if you aren't in the search business, so here's my take on it. The main difference is that Google makes more money per search -- about 20-25 cents per query -- than Yahoo, which makes 10-11 cents per query. [These numbers came from here but are currently wrong, and may have been wrong at the time.]

The first thing that people think is that Google's search results are better, so that translates into more money. That might help to explain the market share difference -- Google has a bit over half the US search market; Yahoo has about 20% (Neilsen stats). But Google makes a ton more money per search than Yahoo! does, so the gap in market share doesn't explain the revenue difference. Also, no external measures of quality that I've seen find that Google's results are better than Yahoo's (the last two I read found Yahoo to be better, though both had some methodological problems). From the standpoint of the organic search results, Google and Yahoo are tied.

So what's going on? The difference is not in the search results; the difference is in the quality of the ads. Google's ads are way better than Yahoo's. To see why, we need a little bit of history.

Ads in search are based on keyword matches. If you want to place an ad, you write the ad, then you buy the keyword, and the ad will show up somewhere on the page when you run a search. (Incidentally, this was patented by Overture, which Yahoo! later bought. Google licensed the patent from Yahoo for 2.7 million shares of Google stock.)

In the old days, how well your ad placed depended on how much you paid. Willing to pay more per click? You get placed higher. Bidding was what mattered for ad placement. Both search engines did that for years.

However, some bright spark at Google realized that to maximize revenue, what matters isn't the price per click, but rather the price per click times the clickthrough rate. If you have an ad that gets twice the clickthrough at the top spot than some other ad, you can charge that buyer half as much and make the same amount of money. So that advertiser needs to spend much less to get the top spot. The other way to think about that is that Google offered advertisers a discount for quality. If you have a relevant site and a good ad, Google said, you can pay much less per click to get good placement on our page.

This turned out to have a huge impact on search revenue over time. What happened is that advertisers started to compete on ad quality, and over time, the ads became more and more relevant to the searches. In fact, after a while, users who were previously disinclined to click on ads learned that the ads could be useful after all, and they started to click more. So Google started to ramp up its monetization, pulling away from Yahoo, and in the process, making the search experience better for its users. Even if Google's results weren't any better, the overall experience was better because the ads were so much higher quality that they were like having extra useful results, whereas the old-style Yahoo ads were still more of an annoyance than an asset.

So where was Yahoo! in all of this? Well, they realized pretty quickly that they needed similar technology, and so they had Overture (remember them, the ad patent?) create a new system. This was a long project, planned for a year, because of the approach they took.

There are two ways to find out what the CTR*price number will come out to. One is to occasionally vary the order of the results, and see if the CTR changes. If the new ad gets a high CTR*price, then you can promote it. However, this isn't ideal: it takes a while for you to figure out which is the best ad, and advertisers change their ads; plus, since you're sampling, you can have all kinds of sampling errors or biases. You can also notice when the CTR profile of the ads is unusual -- if the second ad is getting many more clicks than you expected relative to the first ad, then you might promote it. I think this is how Google initially started their system, though it's not what they do now (or not only what they do now).

The other approach is to have a system that can estimate the CTR of an ad before it gets shown, based on the ad text, the query, and attributes of the page. If your estimate is good, then you just calculate the estimated CTR*price, and you have an ordering. This gives advertisers quick feed back on how their ads are predicted to do, and then you can adjust based on actual CTRs once the ads are out. (Here's a paper from WWW 2007 on this problem.)

Yahoo decided to skip the earlier order experimentation and CTR observation method and went straight to the predictive model. This was a big effort, an attempt to skip the early steps and jump straight to the end in an effort to leapfrog Google. More critically, they decided to do it along with a whole revamping of their entire advertising platform, revamping the whole way that advertisers bought any kind of ads, not just search ads, from Yahoo. As happens with big projects, it hit some problems and came out late, but with surprisingly high quality and functionality.

So has Yahoo caught up? Not yet. The reason is that marketplaces take time to work, and Google has had an ad quality marketplace for years, whereas Yahoo has only had one for a few months. Google's ads are therefore higher quality, and Google's users are used to high quality ads so are more likely to click on them. Yahoo will continue to catch up as time passes, but the gap is huge: several years in Silicon Valley is an eternity, and that is the head start that Google has in search monetization.

This monetization also translates directly into market share. Here's how. Google makes 20 cents or so per query, whereas Yahoo makes half that. So when AOL solicits bids to provide search technology, Google can bid twice what [more than] Yahoo can and still make a profit.

So there you have it. Google is making much more money than Yahoo not because of search result quality, but because of ad quality. Usually if you take care of your core business, serving good results, the rest will take care of itself. Google figured out how to turn advertising, formerly parasitic on the core business, into a way to actually help customers. Ads became part of the core business, and revenue skyrocketed.

Note: I work for Yahoo, but none of this is corporate opinion. I've used only publicly accessible information for this posting. Not that anyone will care, but you never know.
Comments 
27th-Oct-2007 07:19 am (UTC)
If you're testing ad quality, why don't you hire doofs for $8/hr to click on ads? One page of N add, and the one they pick gets promoted. A team of 10 guys like this could click through ten pages a minute, for five hours a day and get you rock solid quality filtering in a month.

I'm sure the Overture system will be great, but till it's done something like this should work pretty good.
27th-Oct-2007 07:56 am (UTC)
I don't work on ads quality, but I'm sure the stock answer within Google would be "doesn't scale". There are billions of ads for millions of keywords (and combinations of keywords) and dozens of languages. I'd also speculate that it's pretty much impossible for any small team to have the domain knowledge necessary to judge ad quality - for example, if I search for "mesothelioma", all the ads look to be about the same quality to me because I'm not out to make a quick buck suing folks.
27th-Oct-2007 02:54 pm (UTC)
It doesn't need to scale. You're only worried about covering the ads which represent the highest volume of searches. Low lying fruit for the stopgap measure.
28th-Oct-2007 08:03 pm (UTC)
Searches and ads have a long tail. I think you may be greatly underestimating how many searches you'd have to hand-evaluate to get any substantial fraction of revenue.

And of course, hand evaluation goes stale over time, as the meanings of queries and the ads that exist change.
27th-Oct-2007 03:38 pm (UTC)
The new system actually rolled out earlier this year, called Panama.

It takes a while for the ad marketplace to react, and of course Google has a big head start, but Panama is already helping.
27th-Oct-2007 04:05 pm (UTC)
Yay!

Did the roll out party feature Van Halen songs?

;)
29th-Oct-2007 06:35 pm (UTC)
Ah. Gold-farming for ads :)
27th-Oct-2007 02:12 pm (UTC)
I remember that one of the things that turned me away from Yahoo as a destination wasn't just the quality of ads, but the intrusiveness of ads. IIRC (which, to be honest, I'm not entirely certain, but it does connect in my mind) I went to Yahoo to do a search, and a second after the page opened, my speakers blared this swooping sound, and a picture of some spaceship flew to the front of my screen and almost scared me half to death. I swore right then that a) I wouldn't use Yahoo anymore, and b) I would never see that show. Sadly, that show turned out to be Firefly, so I did change my mind on that account. Perhaps it is time to change my search habits as well. *8)
27th-Oct-2007 03:39 pm (UTC)
Yeah, those ads are annoying. Note that I was talking only about search ads, though, which are text based and just sit there on the page. If you go to search.yahoo.com you won't ever see those really annoying ones.
27th-Oct-2007 03:56 pm (UTC)
Ah, so Yahoo has a search only page. That is nice to know.
27th-Oct-2007 08:51 pm (UTC) - Easier to type
Anonymous
"google.com" than "search.yahoo.com"

Maybe its even easier to type google than yahoo. I don't know.

How about a search engine called "asdf.com"?

-Ming
27th-Oct-2007 04:41 pm (UTC) - My take on Yahoo
I've had my.yahoo.com as my home page for many years, yet even before Google, I never really used Yahoo as a search engine, choosing Lycos, Alta-Vista, etc, instead. I really have no idea why. However, when I do use Yahoo to search I generally find the results less satisfying than what I get from Google.

I am disappointed that Yahoo has dropped Bill Pay, I've been using it since before my bank offered it and I'm lazy enough that I didn't want to change all the autopays I had set up. I'm sure Yahoo didn't have enough customers for it to pay any more since almost all banks now have it for free.

Your FFL Stattracker? Awesome.
27th-Oct-2007 06:30 pm (UTC) - Re: My take on Yahoo
It's not surprising you find Yahoo's results to be less satisfying. People generally do. Here's what's going on (I almost included this in the original article, but decided it was a bit off-topic.)

If you take some result set for a query and show it to a set of users with Google branding, and take the identical results and show them to a group of users with Yahoo branding, the set with the Google branding will consistently be rated as more relevant and less cluttered. Even though the results and formatting are the same.

Google has a totally kickass search brand. Just marking search results with Google makes people like them more. Annoyingly, this is true even of people who work on search at Yahoo.

Since I work on search, no comment re Bill Pay or anything else -- everything I know I find out the same way you do.
28th-Oct-2007 02:16 am (UTC) - Re: My take on Yahoo
That definitely rings true. As a test as I was posting, I searched on both sites for "Rimadyl", which I've started giving my dog. The top few were identical, but for some reason Google felt more comfortable. Not sure if it was the larger font, the lack of numbering, or the text being displayed, but I liked the Google experience more. Yahoo seemed colder than Google.
27th-Oct-2007 04:43 pm (UTC)
Hmmm... Yes, could you please put that into a chart and use short phrases so that I can pass this 'round to friends that advertise on Google but know as little as I do about what we are actually doing?

K, thanks. And don't forget the TPS report.

Oh, and you have to come and play Zombie Fluxx. That is a Messed Up Fluxx game. You'll enjoy beating me at it.
27th-Oct-2007 06:41 pm (UTC)
Hm, well, how to advertise well is kind of a separate topic than what makes a good system from the search engine perspective.

By which I mean, I have no idea how to create a good ad. Experiment lots?

The stuff I do know is pretty obvious: be clear about what you're offering, make sure that the query terms you bought show up in your title (ideally) or at least the abstract you write, and make sure the end user can tell what kind of site it is (is it an online store, or the official site for a physical store, or the official site for Dorothy of Oz, or whatever).

I also don't know much about Google, they probably have this too, but Yahoo has a lovely set of tools that tell you all kinds of things about how your ad campaign is doing, whether you're making money, which terms are working out and which are sucking, which creatives you wrote are working well, etc.
27th-Oct-2007 06:45 pm (UTC)
Know what? A large breakfast and a big nap need is prohibiting me from dealing with work stuff right now.

I love vacation. Except in the newspaper the hotel offers there's a whole article about this advertising crap stuff.

27th-Oct-2007 07:59 pm (UTC) - APIs matter too
I worked a bit with both the Adwords and old Overture APIs(pre-Panama) and the Adwords API was much better. Both simpler and more flexible, better documented and better example code. They also have very good ad tracking tools.

It was clear that the Adwords API could be useful to anyone who wanted to do any automation, but the Overture API was only designed for big ad buyers. Yahoo was reluctant to give us API access unless we were going to buy alot of ads. I suspect this is another way that Google was beating Yahoo. By scaling tools for small ad buyers all the way up to ebay-sized ad buyers they give small ad buyers confidence to scale up ad buying quickly..

To be fair I don't know how it works in the Panama world.
27th-Oct-2007 10:41 pm (UTC) - Re: APIs matter too
The Panama API has been very well received by all accounts, and lets you do a bunch of cool stuff like geotargeting that you couldn't really do before.

The previous API sucked. That was known and one of the goals was to fix that. I think they have.
28th-Oct-2007 08:01 pm (UTC)
I have lots of opinions on this, since I worked on Google search quality and now work on ads (and specifically ad auctions). Let me think which of them are for public consumption, since Google, while incredibly open internally, is very closed about information getting to the rest of the world.

Hmm. Nope. I tried typing some things, and decided I couldn't go on record with any of them. Ah well.
29th-Oct-2007 11:36 am (UTC)
But did you search for public versions of what you typed using Yahoo! or Google? ;)
29th-Oct-2007 05:06 pm (UTC)
the problem is, the fact that I think them is still meaningful. A lot of things are said about google, some are true, many are not. My opinion on which are which is meaningful inside information.
29th-Oct-2007 12:52 am (UTC)
Thanks for a fascinating perspective on the yahoo vs. google adwar. Ironically, I am like many others who posted here, that they use yahoo for a lot of things except search. I use yahoo every day for financial info., news, and most recently, fantasy football and mail. (although the mail is only related to fantasy football.)

I think google enticed people for search because their page has always been plain blank white and un-intimidating, while yahoo's page is so super-busy and scary. Google seemed so much easier and no-nonsense if you just wanted to know where to find weird stuff, the symptoms of a disease, or the definition of a word someone just called you. I was an early gmail user, and I can't imagine using anything else for email, even though the big brother factor of it is pretty weird.
29th-Oct-2007 01:07 am (UTC)
(via patrissimo's journal.)

Nice post, but IMO the best part is this quote from the article you linked:
The news comes as Wall Street, dealing with its annual August dog days, fills a news vacuum with rumblings that Google's IPO is facing weak demand and will have to be made more attractive to supposedly value-obsessed investors.
..Bwahahahaha!
29th-Oct-2007 05:47 am (UTC)
It sounds strange now, but remember that Google's IPO wasn't really a success. They cut back on the number of shares sold and dropped the price from where they were expecting.

It was a huge success for the people who did buy, of course. And it's done ok since then.
29th-Oct-2007 04:20 pm (UTC) - Oh I think it was a massive success (for the interested parties)
I think that google had to drop from their auction calculated $100/share down to $85/share as a bribe to the investment banks. I don't have any idea what backroom deals went down, but the banks made out like bandits.
29th-Oct-2007 05:13 pm (UTC)
I wouldn't describe going up almost 7-fold in 3 years "ok" :).

But agreed that the IPO had problems, and the price got cut due to low demand, the playboy interview during quiet period thing, etc. Too bad, since the Dutch auction is awesome, as is keeping more nmoney out of the hands of the banker middlemen.
29th-Oct-2007 05:43 pm (UTC)
I wonder how many of the problems with the dutch auction existed because the banks didn't like being cut out. It's certainly a better deal for the company and the investors, even if it didn't work out that way for Google this time.
30th-Oct-2007 06:57 am (UTC)
OK, wow. $.20-$.25 per query? Is that per query, period? Per monetized query? Is it only revenues per query? Making $.20 per query for every single search doesn't pass the envelope-back test for me.

I can't imagine that more than 10% of searches get monetized, and even that seems absurdly high, but I'll go with it as an upper bound. I can't imagine that a single click averages more than $.50, so that would mean an average of four to five paid clicks per monetized query.

Explain?

BTW, I found this utterly fascinating. Thanks for posting it.
31st-Oct-2007 06:41 pm (UTC)
I got those numbers from here, but they are probably wrong. For reasons that will become clear, I updated the post to reflect that.
This page was loaded Oct 22nd 2014, 12:11 pm GMT.