Behavioral Insider: Zillow is known for its "Zestimates" of all our home values, but what other data is being packed in now?
Schwartz: Transactional data, what people paid for the house, what their tax base was when it was last sold. We started showing for sale listings and are up to 3 million. Now we have a broadened mission to serve homeowners through the whole life cycle of ownership. That is where behavioral targeting comes in. Every homeowner has the opportunity to dialogue with Zillow, whether you are prepping a home for sale, shopping homes or just monitoring your asset. In markets likes Los Angeles and San Francisco, well over 85% of all the homes have been searched on Zillow. It creates incredibly compelling targeting opportunities.
BI: What is the state of the ad business now at Zillow in terms of clientele and product mix? And how are you weathering the housing decline?
Schwartz: You buy so much when you buy and renovate a home that home-related advertisers have done very well for us. The retailers, financial services and mortgages are very important. The endemics like builders and brokers spend with us. We have a self service ad product, and 17,000 advertisers have geo-targeted down to the Zip code individual advertising: agents, mortgage brokers, gardeners and architects. I get some eye-arching on this one, but turmoil in a market drives change. Agents and brokers are becoming savvier about how they allocate their ad spend, and the Internet is an amazing beneficiary. We have seen substantial growth in our core endemic ad categories.
BI: How can this data become behaviorally targeted ad product?
Schwartz: The most important is "home attribute targeting" and the cycle of home ownerships. We are able to model from usage patterns on the Web site what status we think the home is in. Is it likely to come on the market and it is being staged; what we call pre-sale. Is it being actively sold now? Has it recently sold? Is it about to be renovated, or is it back into stable ownership? Being able to put digital advertising in front of the owner or shopper for that home is imperative to a lot of ad categories. When you renovate, the retailers are all over you. Before selling, you do a lot of painting and staging and of course, real estate agents want to talk to you. We call it the cycle of home ownership. We can predict when a home is going to be put on the market in advance of it going on the market.
BI: What are some of the indicators of those behaviors?
Schwartz: If an owner comes to her house once a month in general to check the value of their homes and suddenly they are getting very interactive, like uploading photos of a home, correcting the home facts, or rewriting some text about the home and checking comps on neighbors homes, that is very indicative of prepping for sale. We are able to validate those models upward of 78% accurate.
BI: Are you targeting behaviors around specific properties or around cookied individuals?
Schwartz: We do both. We profile activity on the specific home level, and then we try to connect that activity with the specific visitor and correlate an address or area of interest. If we see a home shopper who may not own a home but is actively searching in a neighborhood for sale listings, that is someone likely interested in a neighborhood. Our advertisers are pretty wild to reach that individual. If someone claimed a home, we know the address and then we can profile activity on that address and send ads at the cookie level.
BI: How are some brands already leveraging the data?
Schwartz: Many telecoms advertise down to the home level where they have high-speed networks. We deliver graphical ads down to the individual address because all of our searches are at the address level. A number deliver a fiber message to a home that is actually wired for fiber. For a John Deere lawnmower campaign, we targeted by acreage. Over an acre, we showed ads for riding mowers, and smaller acreage had ads for push mowers. In the Zillow Mortgage Marketplace, we let consumers post their loan applications without name and contact information. Brokers reply with non-binding quotes. If a consumer is shopping for a 30-year fixed mortgage in N.J., that is a high-quality ad impression. When you go to a home details page, we will [retarget] you with 30-year mortgage ads.
BI: With 5 million users, you have to run into scale issues when parsing to Zip code level on BT campaigns.
Schwartz: Too often we get into dialogues where folks want to geo-target and run BT on a very specific segment, then you just don't have enough scale to move the needle. A national bank coming to retarget folks who want a 30-year jumbo, we have more than sufficient scale to serve them really well. When you want to do a 30-year jumbo in a Zip code that isn't a dialogue, we will enter into it. We focus on national or large regional BT opportunities not overlay BT with geo. If someone is targeting specific homes, we can do that as long as they are targeting homes across the Northeast region. The home ownership cycle scales nicely, since we assign a home ownership score to every visitor who comes to the Web site: own a house, shopping for a house, house for sale, is renovating. We stick to those broader buckets.
BI: Any metrics on performance so far?
Schwartz: It entirely depends on the advertising creative and how specific we get in the targeting. On mortgage targeting we see a 2x to 3x improvement in front-end click-through rates and consistently strong back-end conversion rates. Home-related targeting like renovations is incredibly predictive and very powerful. We have home-improvement retailers buying every impression on that. We have the ability to tune our mover bucket to days from transaction. We found that is very important to conversion rates. You are looking at different product at different times in your move. You select your cable or telephone provider within four weeks of a move. We are spending a lot of time figuring out which products line up at what periods, pre or post of a move. The ability to do that at scale fundamentally affects ad performance.