IBM’s Watson Advertising, formerly The Weather Company, is partnering with personalized data marketer Jivox to create and deliver contextual ads in real-time. Ads can be tailored based on dynamic local weather information–meaning, not only might customers in a cold snap get more ads for hot coffee, but if it’s snowing outside, the coffee ad itself might be altered on the fly to include snowflakes.
For Mark Risis, Watson Advertising’s head of global partnerships, simple targeting like this, even accounting for sudden changes and local variations in weather all over the continent, is just the first step. The broader challenge is understanding the subtler effects of weather on customers’ emotions and purchase intent.
“When we analyze weather we don’t just look at temperature, we look at up to 30 different attributes, some of which are apparent to the consumer, and some of which are in the most literal sense invisible to the consumer, like barometric pressure,” said Risis. “And we look at which of these contribute to sales increases, and traffic increases: the kind of KPIs [key performance indicators] that marketers really care about.”
That data is often not obvious, and varies widely based on geography. For example, according to Risis, coffee sales in the U.S. Pacific Northwest are driven more by relative wind conditions than any other variable, while in the Northeast, changes in temperature are both more dynamic and factor more strongly into a purchase decision. And while weather is still the company’s bread and butter, Watson continues to add more contextual data as it learns what shapes shopper’s decision-making.
For Jivox, data like Watson Advertising’s is valuable for two reasons, according to CEO Diaz Nesamoney. First, it’s less deterministic and more contextual than most other forms of data mining. “It’s literally a forecast,” said Nesamoney. It doesn’t assume that past behavior predicts future behavior: a customer who buys snow boots is actually extremely unlikely to buy a second pair of snow boots, but might very likely buy an umbrella and raincoat when the weather changes and that snow turns to rain.
It also levels the playing field to some extent between companies that have rich data about their customers and those that have less information. Contextual data like weather can deliver a highly-targeted ad without necessarily requiring the level of personal information a direct retailer might have about a customer’s past purchases, what they’ve viewed on the retailer’s site, and so forth.
“Machine data and learning algorithms can fill in the gaps in the data we have,” said Nesamoney. “It can also handle cases that humans can’t really do, where the combinations and correlations are too sophisticated to be obvious … But this data that’s contextual in nature is every bit as compelling and in many cases offers better predictive power … It’s going to be very exciting especially for CPG brands, who too often lament that ‘we just don’t have enough data.'”
For Watson Advertising, its years of analyzing weather data give the company a robust data and technology platform it can use to process huge amounts of data in real-time. As the company gathers more data on consumer choices, it can ideally produce similar kinds of forecasts to predict buying behavior. Weather is a key data set and it provides a technological backbone, but it also provides an analogy for understanding consumer behavior.
“We often talk about how the core process of taking a lot of information, crunching it and generating what is the most likely forecast is at its core very applicable and extensible to predicting what consumers will buy, or where they will consume their media or where they might be going and what they might be thinking about,” said Risis.
The key is whether Watson, Jivox, and their data platforms will be able to learn as much from their ad deployments, and whether those ads convert into sales, as the Weather Company has been able to learn from predicting and then measuring weather events. The ad platforms can measure clicks and interest, but the really rich data is in measuring sales. Otherwise, it’s like predicting rain, and watching rainclouds, but never being able to measure just how much rain really fell.