The following is the transcript of an interview of Shashi Seth, Product Lead, Search, of Google. Shashi is responsible for putting together the new Custom Search Engine strategy just announced by Google. Read our analysis of the interview with Shashi Seth, where we discuss the implications of what we learned during our discussion below.
Interviewing Shashi is Eric Enge and John Biundo of Perficient Digital. Eric and John have had early involvement in the Google Co-Op program since its launch last May and visited the GooglePlex in September 2006 to preview the Google Custom Search Engine program.
Eric: Can you provide a general overview of the Google Co-Op strategy? I.e. are there more new products coming, or will the focus be on enhancing the existing ones?
Shashi: When we launched back in May, our goal was to introduce a new direction for search, to get human expertise to enhance our machine algorithms. Our first launch was an eye-opener in many ways. The one thing that we learned was that Topics work extremely well. Our users who interact with Topics are happy. They come back to the Topics and come back to them many times. In many ways, health and destination guides, the two Topics that we launched, proved the concept for us and paved the way for us to move forward.
The one thing that we did learn there, the value proposition for our partners was not as crystal clear as we would have hoped it would be, and that’s why we embarked on building another product called Custom Search Engines, which actually is intended to enhance the value proposition that our customers see, which is to create a custom search engine for yourself. And that’s the new direction for Google Co-op
Eric: Do you see yourself enhancing existing products, or are there new ones in the Co-op hopper?
Shashi: I think the focus for the next coming quarter will be to enhance and shore up the existing products. And there is much work to be done.
Eric: What are the main search problems, from a quality of search experience point of view, which you’re trying to address with Custom Search Engines?
Shashi: The quality problems for Topics were not that many. We did not encounter any problems with them. There is no doubting that there are problems left to be solved. This was a long journey that we have embarked upon, and we have not reached the destination for sure.
But I think it was more about the value proposition and making sure that our partners felt like they were getting something out of engaging in, and contributing their knowledge and expertise to Google, and we’re getting something in return, and that was the custom search engine. This is the product that people wanted and had asked for in the past. Now that we had the infrastructure built out with Topics, it was an easy next step for us.
Eric: Are there issues in Core Search that the family of Co-op products is really targeted at solving?
Shashi: That is a really good question. The problem that we’re trying to solve is essentially this – when a user comes to Google, they often enter a query that is a couple of words at the most, for example, “BMW X5”, and we don’t know what the point, what the intent or what the context is, and that’s a tough problem to solve, when you get those few keywords as part of the query.
When we start trying to retrieve answers for “BMW X5”, you are getting a slew of answers that may or may not pertain to the user’s intent. For example, the user may really have intended to say “what I really want is some accessories for it”. Or they may have really meant that “I want to buy a BMW X5”, or read some reviews about it. Or even more specifically “I only want to read negative reviews”.
Eric: To make myself feel better about the car I bought. (laughs)
Shashi: (laughs) Yes, or about the car I am about to buy. We don’t know that intent. So this is a journey to try and solve that problem. To solve that problem we need to solve another thing, that in our search engine we have a fair amount of knowledge of what a page is about, but we are not able to associate its very specific intent and context. For example, it is easy for us to say that this page is about treatment, but is it not easy for us to say is this page really meant to be for a patient, or is this meant to be for a health professional.
And only when an expert in that field, or a person who is knowledgeable in that field, look at the page can they really make that statement. So somebody like CDC or Harvard medical school can tell us “yeah, this is really meant for medical professionals and I don’t think a patient would get much out of it”.
Eric: That’s a big thing where human input is critical, context and intent.
Shashi: Exactly, and so when we get that kind of input in the form of meta information about URLs or URL patterns, we try to understand the intent of the user and it becomes easier to get the user closer to the answers that they were looking for. There is another problem that is very relevant many times, that is often people know what they are looking for, but they don’t know how to express it.
A good example that I give to people is that imagine that a young mother takes her 6-month-old to the doctor because the child has been crying, and the mother doesn’t know what the problem is, and when they go the doctor they figure out that the child has Rheumatoid Juvenile Arthritis. And the mother is so distraught that she doesn’t really remember all the words, she just remembers Arthritis, because she has been exposed to that word before.
Late that night she comes back home, and when the child goes to bed she goes on the net and types in the word “arthritis”, and she then tries “arthritis treatment”, but that is not necessarily what she really wanted. What she wanted to say was “I am looking for juvenile arthritis treatment” or “rheumatoid arthritis treatment”. What happens is that she goes through a number of pages before the word “rheumatoid” pops up, and then she changes the query to add that word to it, and lo and behold, the answer comes up.
If we do a good enough job of looking for the user’s intent, we could do a better job of asking the user “Are you looking for Arthritis treatments for children?” And that might have solved the problem right there. If we look at health, it’s not as clean as trying to match labels to the results, we also actually switch context For example, when you are in health, and you click on “for health professionals”, we switch the entire context on you to a set of labels that are actually meant for health professionals.
It is this approach that I think will, over time, make the user experience better, and help people find what they are looking for. At the end of the day, that’s the goal of Google search.
Eric: Will Google help users find the best CSEs? If so, how?
Shashi: Not initially. We currently do not have a directory, or any discovery mechanism for custom search engines today. It’s something that we want to do, but in our list of priorities, it is probably much lower.
Eric: We are actually building such a directory here, and that will be live on launch day.
Shashi: That’s awesome. We have always relied on partners and third parties to help us. A great example of that is that there are significantly more comprehensive directories of our iGoogle modules of home page modules than we have, and that’s the type of relationship we like to keep with our partners that can bring value to our products and our services. That is absolutely fantastic.
Eric: Do you see Google taking a role in the future, of trying to make the best CSEs known by some means or another?
Shashi: I think at some point our users are going to demand some of these things, for example, “can you help me find an appropriate custom search engine”. But like I said before, there is a list of things we would like to do, and it will be a while before we get to that point, and we are going to hope that people like you provide the service for us.
Eric: What kind of support for tools and APIs is Google planning (or announcing at launch)? Will there be anything beyond what we previewed at the Googleplex a few weeks back?
Shashi: I think the two things that are on the radar in the near term are statistics on usage, etc. and some debugging visualization tools that we can offer to custom search engine builders as well.
Eric: I think one of the suggestions we made is being able to show default, or normal, Google results and your CSE results side by side.
Shashi: Absolutely
Eric: It would give you a real fast comparison and you can make sure you are actually making progress. But in any event, it sounds like there is nothing on day one, but it’s high on the priority list.
Shashi: As far as debugging tools?
Eric: Yes
Shashi: Absolutely, there will be nothing coming out on day one. Our goal was to get something easy to use out there, and then come back with features that we know our users will want, and we have heard about from people like you.
Eric: Fair enough. Let’s switch directions a little bit, and talk about existing custom search engines. What are the top 2 or 3 differentiating features of Google Customized Search Engines v.s. the competition (Yahoo Search Builder, Rollyo, and Eurekster/Swicki)?
Shashi: Sure.
- First and foremost, the power and capability in our tool. We pretty much are trying to give you as much as any user would need to create a custom search engine of their own. Not only restrict and filter options, exclusions, a list of sites you can have in there, which a couple of competitors have
- The ability to customize the look and feel to offer refinements much like what we offer in Topics
- Collaboration features so you can invite your community, and have them help you
- Ultimately the ability to monetize, I think many of our competitors don’t have this feature at all. How do you create a custom search engine, put it on your blog site, and still be able to make money off it?
So I think those are key differentiating factors, but I think the one I always like to play on is quality. You are getting a slice of a really quality search engine to begin with, which is Google, so the quality of your product should be very high as well.
Eric: So you are making the statement for us that you think Google has a pretty good search engine? (laughs)
Shashi: Yes!
Eric: The Eurekster/Swicki “buzz cloud” seems like a neat idea. Will Google CSEs have a feature like that?
Shashi: I have not heard of it. I have never used that
Eric: John, can you clarify this feature for us?
John: Sure this is something that comes in the user interface for the search box. It’s a little box of terms that appears in the entry box for the search terms, and what it does is display the most frequently searched terms, and it indicates their frequency by the size of the font.
Shashi: Ah, just like a tag cloud.
John: Right. Their pitch is that it, the term they use is that it’s the online water cooler, and it helps people discover what’s hot in a particular topic area.
Shashi: This is similar to what we used to call zeitgeist. So they use this to call out the most popular queries for that day, and so forth. That’s pretty easy to do, and we could create those tools pretty quickly if that’s what people want. If search engine builders want this, then it should not be really hard. I think it’s a side benefit.
As part of things we plan to offer to our custom search engine builders in terms of statistics and information on queries, this might be an interesting feature, and we might give them control over it, whether or not they want to display it. I can’t imagine that every partner of mine has the space available to display this kind of information next to a search box. Working with Intuit, I learned very quickly how premium space is. I think it’s fun, but definitely not something that everybody would want to use.
Eric: It seems to us that the market has misunderstood the essential benefit of Subscribed Links. People talk about Smart Answers, and Oneboxes and things like that. We have always viewed Subscribed Links as a way for a Publisher to provide their own Smart Answers, and you can define that as not having to actually go to the website to get the answer to your question because the answer is there in the search engine results.
But, are Subscribed Links something you are going to continue to push, and how do you see that expanding over time?
Shashi: Subscribed Links are exactly what you described for us. It’s essentially a tool that allows our partners to do two things. One is to provide contextual and relevant answers in the search results page, and build this brand association with their users. For example, users may come to their site say, for example, say WallStreetJournal.com, and ultimately want the Wall Street Journal’s answers or opinion on something right at the top of the page. When a user subscribes to the Wall Street Journal (their feed), that is the signal that they are giving us, saying I trust WallStreetJournal.com, and where relevant, surface their answer to the top of the page.
Eric: Right
Shashi: And that’s really the benefit that the user gets out of it. If I am interested in cricket scores, and I trust cricinfo.com, when I come to Google and type in something like cricket scores, I should be able to see all the cricket scores that are going on right now and have them come from Cricinfo.
The other element that I think that people have also missed here is that there are lots of pieces of information that Google cannot index, real-time information for example, like site statistics. Those are things that are changing every second, or every minute, and therefore don’t belong in our long term index. This is a great way to actually get to that information, both for our users and our partners, so that users find benefit and value in using them, and build that brand association with our partners so that they go over to the partner site when appropriate.
A great example of that is if I type in the name of a movie today, and I subscribe to Fandango, I should be able to see the movie results, the information about the movie, and when I click over, I should be able to go to a page on Fandango that actually puts me on that movie page, knowing full well that I am in 94043, and shows me only the relevant results so I can buy the tickets, if I want to, right then and there.
I think that’s the power of it, so we are absolutely going ahead with it. There are going to be lots of exciting changes that are coming to this product. We are working on internationalizing these products and making it easier for our partners to work with.
Eric: OK, can I slip in one more question?
Shashi: OK, go ahead
Eric: It seems to us that the long term success of these programs, not just for Google, but the people who implement them, is based on the strong editorial approach that brings value to users so that the user will be compelled to return to either your Subscribed Links or your CSE. What’s your sense about that?
Shashi: Can you explain that a little bit?
Eric: Sure. You can take the CSE and remove the whole web, and have only your own site. You could do that, but you would not need a CSE to do that, first of all, but it’s very parochial if you will. You are not adding a lot of value. Similarly, you could make Subscribed Links and provide nothing but ads, in Google search results, or you could provide Subscribed Links that provide the answer, like Smart Answers.
Eric: Similarly, if your CSE is a high-quality implementation that truly offers an improved experience where you remove spam sites and promote sites that you know to fit the theme of your custom search engines, including ones that are not yours. That’s what I mean by quality editorial input.
Shashi: We expect that there are going to be partners that build very, very good Subscribed Links and CSE because they put in the time and effort to make them really good. And there are going to be people that put in a minimal amount of effort, and put something out there that may not be as appealing to the users.
Shashi: At the end of the day I think our users are smart enough to make that decision as to what they find useful, and what they don’t, and that alone will compel our partners to put in that extra effort. Even today, if you go into our directory for Subscribed Links, you will see that there are lots of Subscribed Links that are really, really good, and you will see that people use them very often. Some of them have a lot of subscribers already and are used a lot, and some of them don’t have a lot of subscribers but are still used very, very often.
So I think it’s going to be a self-selection mechanism here. Which is lots of good ones will get created. People will use them. Our partners will also see those great examples and will strive to also make Subscribed Links and CSEs that are also of that quality. Ultimately the users will tell them by usage or non-usage of these products.
Eric: Thanks for taking the time to chat with us.
Shashi: Thank you.
We have published an analysis of this interview titled One Basic Problem with Algorithmic Search in our blog.
