Get the most important digital marketing news each day.
Google’s Surojit Chatterjee: Here’s Why You Should Trust AdWords Estimated Store Visits
Head of mobile search advertising offers more details on what goes into measuring estimated store visits in AdWords.
With mobile ingrained in our daily behavior, technologies that can tie digital advertising to in-store visits and sales are finally becoming a reality. When it comes to store visits reporting, though, advertisers have to have faith that the data is accurate and actually holds meaning.
Earlier this month, when Xad launched its in-store visits technology, Blueprints, Google reached out to Marketing Land to offer more detail on how their own in-store visits metric works in AdWords. On Monday, for the first time, Google’s director of product management for mobile search advertising and AdSense, Surojit Chatterjee, shared more explicit details about how Google determines if someone has actually visited a store rather than simply walked nearby.
Google began reporting on estimated in-store visits in December, 2014 to show advertisers how ad clicks help drive foot traffic to their stores. At the time of the launch, Google simply said that this estimated data was based on user proximity to the advertiser’s location on Google Maps, captured when location history is activated on their phones. Turns out it’s not quite that simple.
Understanding what goes into the metric could help assuage advertisers hesitant to rely on “estimated” data.
Speaking at HeroConf in Portland, Oregon, Chatterjee explained the signals that go into estimated store visits include:
- Google Earth and Google Maps Street View data
- Mapping of the coordinates and borders of hundreds of millions of stores globally
- Wi-Fi-strength signal in stores. With permission of the stores themselves, Google teams go in and measure the store’s Wi-Fi signal strength within that location
- GPS location signals
- Google query data
- Visit behavior
- Panel of over 1 million opted-in users provide their on-ground location history validate data accuracy and inform the modeling
Asked why advertisers should trust data that is presented as estimates, Chatterjee, pointed to the breadth of signals and volume of data that get taken into account in the sampling. “We are extremely accurate when we show the data,” he added, “This isn’t a user survey.” It’s also why many advertisers don’t get this data at all, yet — Google doesn’t have enough of it to provide statistically valid results.
When I caught up with Chatterjee after his talk, he likened in-store visits measurement to what the company has been doing with products like Google Now, a service that can decipher spoken words despite a user’s age or accent. He explained, Google is good at gathering huge amounts of data and using machine learning to automate the process of understanding signals (grasping a word whether it was spoken by a 4 year-old and a 60 year-old in Google Now) and predicting outcomes (estimating whether an ad click generated a store visit). “We’re very good at this kind of thing.” says Chatterjee, and a big factor in being good is the wealth of data Google indexes. We don’t want to just know the borders of an advertiser’s store, he says, we want to map all the stores surrounding that store to get precision.
The user panel data is used to inform the algorithm and verify that store visits actually happened. Chatterjee says, If we think a user visited a store, and the user panel data shows that a visit happened, then we know we’re good. If we think a user visited a store, but the user panel data shows there wasn’t a visit, we feed that back into the model and vice versa. Google’s model also does not automatically count a user entering a store as a visit, but considers behavior patterns and queries to understand intent and whether the visit was most likely informed by an ad click.
Google has also been working with agencies and larger clients on tracking in-store sales back to ad clicks in AdWords. He cited data from Express, working with digital agency RKG, that showed the retailer saw a 98 percent increase in ROAS on brand keywords and a 174 percent increase on non-brand keywords when looking at store transaction data.
Google partners with the major third-party store transaction analytics firms to pull data for advertisers. Google support teams then work with advertisers to feed that back into AdWords, matching the sales data to ad clicks to show paid search attribution down to the keyword level. Chatterjee says many retailers are now using store transactions and store-visits data in conjunction to inform campaign and bidding strategies.
Chatterjee hinted that the long-term plan is to integrate and automate store transactions measurement into AdWords, but that’s a ways off. Store visits data, meanwhile, is available to a larger set of retailers. With all the signals and ongoing machine learning that go into it, Chatterjee stresses, advertisers that have estimated store visits results in their accounts should have faith in the data.