Frazier Miller
Frazier Miller serves as the general manager for Yahoo! Local, one of the most visited local network of sites online. Bringing a strong background in product management, Frazier is responsible for setting overall strategy and prioritization for Yahoo! Local properties. He currently oversees all of the products that are core to Yahoo!’s ongoing local initiatives, including Yahoo! Local, Upcoming and Yellow Pages.
Prior to his role with Yahoo! Local, Frazier ran product management for Yahoo! Messenger, a leading communications product. Here he was responsible for growing the Yahoo! Messenger business from 60 million users a month to over 100 million users. During his tenure leading the Yahoo! Messenger team, Frazier also oversaw its expansion into the VOIP market, growing both users and revenue from the ground up.
Prior to Yahoo!, Frazier served in a variety of product management roles with high-growth technology firms including director of product management for BEA Systems and Nimble Technology.
Frazier holds a bachelor’s degree in history and English from Dartmouth College and a master of business administration from Harvard Business School. When not working, Frazier enjoys the outdoors, skiing, wind-surfing, and spending time with his two children.
Shailesh Bhat
Shailesh Bhat, Product Manager for Yahoo! Local
Shailesh Bhat serves as a senior product manager for Yahoo! Local, one of the most visited local search sites online. In this role, Shailesh is responsible for driving key monetization efforts and helping build and maintain features that increase the overall Yahoo! Local consumer experience.
Prior to working with Yahoo! Local, Shailesh worked at Yahoo! Bangalore as a product manager for the mobile products and market innovation group. In this position, he conceptualized and helped drive new products to market for India’s quickly growing mobile user base. Shailesh brings nearly a decade of experience in product management, sales, and software development to Yahoo!, having also held earlier positions with MitoKen Solutions and Quark.
Shailesh holds a Masters degree in Business Administration from the Indian Institute of Management and a Bachelor of Engineering from the BMS College of Engineering, Bangalore.
Interview Transcript
Eric Enge: Can you talk about the types of data sources used in local search by Yahoo! to find the business information? I am specifically focused on business listings.
Frazier Miller: Right. There are a number of sources that we use to do this. It can be a confusing world and there are a lot of different data providers out there. But, in general, we rely pretty heavily on several areas. One is the licensed feeds that we get through data providers like InfoUSA, Acxiom, and Localeze. We rely heavily on these to build our backbone.
But, we realize the data changes very quickly and there are better ways to update that data quickly, so we rely heavily on merchants and give them the ability to update the information themselves directly. We are also using our users more and more to help us update and correct information on the site as well. And so, we have all three of those different areas as ways of building out the database and making sure we have our information as accurate and up-to-date as possible.
Eric Enge: Right. Sources like InfoUSA, and Acxiom, and Localeze, provide authenticated feeds, right? In other words, the data provider has gone through some level of trouble to validate that the location information for a business is correct.
Frazier Miller: Yes. That is a lot of the service they provide and the reason why people like us pay them for their feeds. They have people who are calling on the businesses ensuring the freshness and accuracy of their information. This is important because statistics show that over 50 percent close in their first five years of business (U.S. Small Business Administration).
You also have far more businesses than that whose phone numbers or addresses are changing, so we find that it is not sufficient just to rely on these authenticated sources and that we need to take other means to do that. We are relying on multiple authenticated sources, and some have strengths in one area and others have strengths in others.
So, we need to provide a layer of heuristics to select which of the data providers has strengths in given areas. We will then compare that data provider’s information against these other sources like user input and user feeds, which can be good. But there are also some pitfalls because you can’t rely solely on that. We will match and compare data sources of these different types to ensure we get the latest and most accurate information.
Eric Enge: Right. With user-contributed data, for example, you have the risk that it is intentionally incorrect.
Frazier Miller: Right, there are fraudulent cases. You can imagine a case when someone comes into a competitor site and says, oh, here is their phone number, and they are providing the phone number to their own site, although we don’t see this very often.
We do have human and manual moderation that goes on for changes, so consumer submissions all go through a moderation process where we look for patterns and we actually do validation of data to make sure it is accurate. There are a number of steps we use to try to keep fraud and spam in control.
Eric Enge: Right. A business owner is able to submit data directly to Yahoo as well.
Frazier Miller: Absolutely, there are a couple of different ways. They can add their information for free, as we give business owners the ability to do so, and they can correct and update information. But because we don’t have credit cards and we don’t have authentication against that business owner, it can stay open for other people to add or change as needed.
We don’t lock the information down to that particular user ID or that business. When they become a paid business partner, they basically lock up their profile so that they control the majority of what is said in that profile. When they pay, we obviously have more ability to authenticate and make sure that it is the right business.
Eric Enge: Right. How do you handle discrepancies between providers?
Shailesh Bhat: What you are referring to is essentially how we handle the merge of all these entities. From any of those providers, you could potentially get the same listing, and the value of getting the listing from multiple sources is essentially the enrichment that we would get. One source would give one data set for the same listing as compared to others.
So you are right, we need to have heuristic models in order to arrive at the best-aggregated listing and the most comprehensive information for that particular merchant’s listing.
Eric Enge: Right, so if the business pays, then they control the listing, within the constraints of editorial judgment, of course. If they say the address was 41 Temple Street, and InfoUSA says it is 39 Temple Street, you use 41 because you have an authenticated paid relationship, is that right?
Shailesh Bhat: That is right. But for the specific example that you used of address, we would also use other parameters like geo-coding. So for example, if that specific address did not geo-code to an actual location on a street for example, then we would handle that listing. So it may even get rejected even if it is from the paid feed.
Eric Enge: So, you are calling out the example where they say it is 41, but 41 doesn’t exist as an address?
Shailesh Bhat: Exactly.
Eric Enge: And so, that brings up the whole larger point here, the huge problem is that businesses forget to update their data, or they might not even remember that it is necessary, and changes are happening all the time as you pointed out. You gave the example of 50 percent of businesses closing their doors within five years.
There are also businesses with two hundred locations, and they just added ten new ones, and four of them changed addresses and three others changed their phone number. It is an arduous administrative task for the business, right?
Frazier Miller: Yes, one of the things that we encourage our businesses to do is to update the backbone providers when they have a change in information. If they change their phone number, it is not just a matter of calling up the telephone company and saying my phone number is changed.
There needs to be awareness about all the other providers who are picking up their information and getting it out there. So it is no longer as simple as just the yellow pages, right? We find that paid businesses make more of an attempt to do this. It is not often that we have people who are just paying year-after-year who are not updating or managing their listings in an active way. So, there is certainly a reminder there. We do have tools for merchants when they have a change, but that falls under a paid relationship with the merchant.
We don’t have anything for free where someone can just come and update a mass amount of data. Getting back to the problem of abuse, you could put out a lot of data in a pretty short period of time, so we generally make sure that there is a paid relationship with a merchant and then have the ability to update via feed.
Eric Enge: So bulk submission directly to Yahoo is not available in the free service?
Frazier Miller: That’s right.
Eric Enge: Do you do any validation yourself in any of these scenarios?
Shailesh Bhat: Yes. So for example on the self-serve side, which is essentially listings.local.yahoo.com, that is where a merchant would go to submit a listing. All listings that get fed in through that mode go through a manual moderation. A similar edit is possible from the details page of the listing itself.
There are links on the details page of a listing that would allow someone to clean that listing and submit changes. Even that would go through manual moderation.
Eric Enge: Right. So is there any scenario in which you send an automated fax, or email, or give a phone call to validate contact data?
Shailesh Bhat: In an automated fashion to all our listings, no, we do not do that. There may be specific instances where we would reach out, especially if there are multiple listings that have been provided from one particular ID or something like that.
Eric Enge: More of a spot-check type scenario?
Shailesh Bhat: No. When it is manually submitted, each change that has been made is viewed by someone, but it does not mean that we call each of them.
Frazier Miller: However, there are sometimes exceptions where we will reach out and call businesses, but that is not a standard practice. One of the standard practices we do have though, which is very important to our users, is the URL checks, especially where you get SEMs and others who are intermediaries for these businesses and are trying to get direct clicks into their representation of the business. It is fairly easy to do a more mass scale check via a URL address validation to make sure we are showing the business itself instead of an intermediary.
Eric Enge: Right. So, if you are a business and, for example, you have more than a hundred locations, what is the smartest way for them to manage this relatively confusing situation?
Frazier Miller: I think it comes down to making sure that the core data providers, like InfoUSA, Axiom, and Localeze, are up-to-date.
They are super eager to work with you. They have a business model and they can go and charge folks like us to handle these feeds for free for the business. And then, we get the feeds directly. For a lot of these businesses, it is not just about Yahoo, it is about yellowpages.com, and Superpages, and Google, and a whole set of directories.
So again, I think that is a smart way for businesses to go about keeping their data up to date, whether it is an individual business or a chain. And then, with regard to us entering into a paid relationship, we are certainly a leader in the marketplace and continue to command a lot of market share.
We have a couple of paid products. We offer enhanced listings which are basically $10 a month per listing, where they can enhance the data fields and write a description of the business. There are a couple of additional URL links that they can use and they have the ability to control. We also give them other tools for them to manage their listings.
Eric Enge: Now, if somebody has hundreds or thousands of listings, do they get better per listing price?
Frazier Miller: Bulk discounts?
Eric Enge: Yes.
Frazier Miller: We are always open to negotiation. It generally has to be a lot of listings. When you get into the thousands, like a Starbucks they generally may also be making search marketing buys or display buys. So generally, they tend to be an account that will manage more proactively, so we will absolutely look at how to get the deal across a number of different advertising products.
Eric Enge: What is the best way for someone to engage with the paid listing product if that is what they want to do?
Frazier Miller: We have a self-serve portal that they can go and provision themselves up and running. It is a very easy five-step process. For the larger guys, it would be to engage our inside sales team, which also is a relatively straightforward process. The best starting point for reaching them is at advertising.yahoo.com.
Eric Enge: Do micro-formats play a role or is there a possibility of that occurring in the future as a way of people sliding authenticated data direct from their website?
Shailesh Bhat: Right now, the biggest use of microformats and RDF is all in the context of the SearchMonkey application.
Eric Enge: So you are not using it as a way of extracting location information for Yahoo Local at this time?
Shailesh Bhat: Right now, on Local we are not. But we will keep it open as an approach.
Eric Enge: The next question is about cleaning up the old records, and I think we have probably discussed the importance of that, but the corresponding question is when you have conflicting records, does that essentially dilute the strength or the ranking of your listing?
Shailesh Bhat: Let us go back to the heuristic model that we were talking about earlier. Essentially, the aim is to have as few duplicates as possible. So if there are four or five listings of 39 x street and 41 x street and so on, the intent is to use other parameters to figure out whether they are the same listing and at the end of the day, publish one single listing.
It does make it pretty complex because at 39 x street you may have two businesses running, right? Or it could be different addresses altogether. So we run the algorithm to try and merge listings. And in such cases, depending on the confidence we have on the source, we would potentially not accept some of the listings from a given source.
Eric Enge: Let us say you do accept it, and it passes the threshold of enough confidence to accept the record, you probably have some way of calculating your confidence level in the data. It would seem to me that it would be very logical for something with a lower confidence level to just be presented with a lower ranking in the results.
Shailesh Bhat: Right, so this one factor in itself may or may not be a tilting factor on a specific record. What I am saying is, it is quite possible that just because we have conflicting data from two sources for a record does not mean that it will not show up as the first result for a given query. There could be other parameters, like the details on that record, the type of query that is there, the keywords, or what exactly the query term is and so on. But, this can be one factor.
Eric Enge: Right, fair enough. So let us talk about the other ranking factors if we could. Just so that you know, I have drawn these largely off of a post by David Mihm in which he interviewed a bunch of local SEO gurus and this is more or less what they came up with as important local ranking factors.
Shailesh Bhat: The distance parameter in itself is definitely one factor, but I think it is a slightly overrated factor in many cases. Categorization, I think, is especially important because of the way queries get generated. Your question is: how is this verified?
When people claim a listing, they typically give websites. Also, the description that they provide is used to make sure that the categorization is correct when there is a submitted listing on the self-service side.
In cases where we do not have a self-serve listing, where the merchant has not provided any data, but we have data from other sources, we essentially look out for the degree of agreement between various sources. That is one heuristic element that helps us in figuring out the right category.
Product keywords in the record are very important. Reviews, ratings; essentially you could say depth of content when a merchant submits a listing is a factor that is useful.
Eric Enge: Right. So along those lines, the depth of content emerges in two dimensions. One is the quality and depth of the data you get. The second way of looking at it is the number of times it is referenced across the web on yellowpages.com and local.com and sites like that, but also on other websites. If a business is extremely frequently referenced on other web sites, is that likely to count for something?
Frazier Miller: Yes, it’s a subtle difference, but one that I think is constructive. Web search looks a lot more at the web index and prevalence of things, whereas in the local context, which really is listing specific, we look at that a lot less. There are some things we do, we go out the web to validate, we go to our web index to validate, but it doesn’t play a huge role.
Some of that is back to the reasons that we talked about before. InfoUSA has undue weight in the web index for a listing because they publish through so many people. And so, it could be that we’ve gotten a couple of comments from users that all corroborate with each other.
So, some of it is important because you’ve got these big content giants who unduly influence the web. But it’s an interesting area and certainly, there is more to explore with how we use the web index, but it doesn’t play a huge role today.
Eric Enge: Right, I understand. Let’s start switching gears and talk a little bit about mobile integration.
Frazier Miller: Local searching, local queries, maps, and directions are some of the top use cases, obviously, on mobile devices.
We also have a product called oneSearch. It assumes that you are searching across your email for a given user’s name, or maybe it’s a web search, or maybe it’s a local search. And so, they have an algorithm at their level to basically differentiate different types of queries that go into oneSearch, and whether it’s more likely to be a local query versus a web search query versus an email inbox query.
And then, the queries that come out to us use the same index and the same database that we provide the web search to present their results. In most cases, there will bias factors like distance, if they know the distance, or if they know the location of the user, so they will bias factors like that in their algorithm, but generally, it is the same dataset and similar ranking algorithms.
Eric Enge: What about advertising opportunities in mobile?
Frazier Miller: Yeah, advertising is really emerging in the mobile space and we as a company found a lot of success there, but it has largely come from larger national advertisers who are pretty sophisticated about advertising. It also has a bias towards more display and brand advertising to date as opposed to search marking or keyword terms.
I think that’s just a product of sophistication of the market and the users, and there is certainly a lot of search inventory happening on mobile devices. But it’s extending fairly quickly as more and more people are browsing the web and using the mobile phone for doing searches.
So, there is nothing unique or specific, there are no products that we provide to businesses like we do on the desktop to say, oh here you can appear in this set of pixels for local queries on the web. The enhanced listings have certainly become part of the core-index so that factors in just as it would factor into the web listings. You know desktop listings, but it operates much in the same way on mobile.
Eric Enge: What about geo-targeting options?
Frazier Miller: We see a lot of geo-targeting happening with advertisers. In fact, at the keynote that I spoke at a few weeks ago at SMX Local, we released some data that we’ve seen a 200% increase in geo-targeting advertisers in the last twelve months at Yahoo, and so advertisers are definitely getting the importance of geo-targeting their message.
The context for that I think is trying to minimize marketing spend in a tough economic environment, and if they can bid only on keyword search terms or on display ads in a given DMA or given geography, then they prefer doing that because it saves them on costs. We are very bullish on geo-targeted advertising and continuing to add lots of tools and capabilities to allow ever more granular geo-targeting. As a company, we are very excited about this.
Eric Enge: You are planning to increase the granularity in which you offer geo-targeting?
Frazier Miller: Yes, and I think we will continue to head that way. I think there are very few advertisers who want point-level advertising right now, like just a point on a map. People have gone from the state level and DMA level, which we have today, and are increasingly looking at things like city level and zip level. So, I think that’s definitely an area that we are looking a lot at.
Eric Enge: Right. So certainly it gets challenging if you get more and more granular. If they say they want to advertise in a particular zip code, well it’s pretty easy to have somebody that you geo-located to outside their zip code giving the quality of IP lookup data today.
Frazier Miller: Yes, we have a lot more technology besides just IP lookup that you can imagine from users of Yahoo. There are a lot of inputs of location, whether they are reading local news in that area. We’ve got technology that allows you to store locations that are important to you, and so they give us a lot of information about the places they are interested in.
One of the challenges that’s interesting and that speaks right to the heart of our overall company’s strategy is: let’s assume that we can target effectively down to zip code level. If you are searching on a keyword term like women’s shoes and 94028, the amount of inventory gets pretty small, even for a company like Yahoo where we have tons of inventory.
You’ve probably heard about our exchange concepts, that at the Yahoo Advertising Network, we are aggregating the supply of inventory across a number of different properties outside of Yahoo.
That’s a key piece that I think will come into play much more as we get into ever more granular levels of geo-targeting. Even the biggest players today, Google, ourselves, Microsoft, none of us have enough, or even in the unforeseeable future will have enough, inventory to effectively service a lot of these geo-targeted queries. So, that I where I get excited about the ad network to deliver a lot more inventory against that.
Eric Enge: Yes, indeed. And of course, as people move more and more dollars in online spending, that inventory will go up naturally from that as well.
Frazier Miller: Yes.
Eric Enge: Thank you Frazier and Shailesh!
Frazier Miller: Thank you, Eric
Shailesh Bhat: Yes thanks, Eric. We appreciate it.
