Analytics startup EDO teases consumer behavior from TV ads with machine learning

EDO, a startup in the business of analyzing media, is using machine learning techniques to study how TV advertising can impact consumer behavior. The company's founders believe they can show the causal relationship of what a consumer watches on TV and what they then do in search, Wikipedia lookups, and other forms of brand-related activity.

(Image: LorenzoPatoia, Getty Images/iStockphoto)

Though advertising dollars continue to shift broadly to interactive media such as online display ads and "programmatic buying," there's still tens of billions per year in advertising in plain-old "linear" television, and plenty of efforts to improve the use of the medium.

The problem is how to connect marketing, the advertising "spot" that airs on TV, to behavior in a way that can really prove the value of that ad.

Machine learning of various forms is one tool that is making possible an increasing ability to connect TV viewing and consumer behavior. That's the focus a three-year-old startup named EDO, which has dual headquarters in New York and Los Angeles.

EDO, which has some heavy-hitters from the media world, has made a business watching television feeds in a giant DVR farm and figuring out how ads correlate -- and perhaps cause -- consumer buying behavior online. EDO's DVRs have captured 47 million "airings" of television ad spots over the past three and a half years. That database of ads is compared against publicly available data of things people do online such as searching by keyword or looking up things in Wikipedia, a data set of "trillions" of consumer actions, the company claims.

Also: Advertising data going forward: Transparency and scale

The data science matches between the airing of an ad and the spike in behavior of various kinds that relates to that product or brand, creating a kind of "A/B testing," seeing which spots among several from a brand marketer had the biggest boost to behavior.

EDO Thursday morning announced a new round of funding to the tune of $12 million from Breyer Capital, run by former Accel Partners venture capitalist Jim Breyer. (EDO had a previous, seed round of funding, which the company is not disclosing.)

The company was founded by actor Edward Norton and Daniel Nadler, formerly CEO of data analytics company Kensho (later sold to S&P Global) and Kevin Krim, formerly the general manager of CNBC Digital. Customers include Walt Disney's ESPN, and the Bravo network, which want to optimize how they promote TV shows with their on-air ads. EDO claims to have customers in a number of other industries such as apparel and automotive, though the company would not share names of those clients.


Krim, who serves as CEO of EDO, and Joshua Lee, the chief technical officer, who was formerly with Airbnb, took some time to speak with ZDNet to explain the company's uses of artificial intelligence to make sense of the masses of data they're observing.

Clients are looking to figure out which ads yields an effect, down to the individual time slot, to maximize the return on investment as they shift money around. is as they shift dollars here and there. Sometimes, that means figuring out what works even before planning the ad campaign, by tapping historical patterns that EDO has observed.

"We had an auto client who was launching their first-ever subcompact SUV," explains Krim. "That client was not able to look at their own history [of advertising] for that product, they couldn't see based on their own experience which [cable TV] networks are most effective to generate engagement."

"With our help, they were able to look back at a very specific set of campaigns by competitors, just for the mid-luxury segment, and look at six or eight historical campaigns by competitors" to see what campaigns seemed to prompt corresponding behavior by consumers.

"What's so cool is that we can contextualize data for the entire industry, by segments such as luxury versus non-luxury, and SUV versus non-SUV," says Krim. "Clients are able to get a nuanced view by looking at their data but it's only by looking at entire data." The output of such differentials in ad impact can be money saved, Krim explains. "We had an auto maker launching a fully redesigned version of a flagship model of theirs, and they were over a long weekend this summer running six different 'creatives.' We could show them that between the top- and bottom-performing spots there was a 50 percent difference in impact -- the top ads were 50 percent more likely to engage a person than the lower performers."

Also: AI startup gets $30 million from Andreessen Horowitz to free up workplace drudgery

As a result, they had the possibility of doubling their budget allocation to the top performers, and maybe cut the allocation to the bottom ones. That choice would generate the equivalent of an 240 additional prime-time spots. If the average cost is $50,000 per spot, you're talking an extra $12 million in effective spend."


The EDO team.

EDO, which began life as "Big Data Oracle," combines hours of TV "viewing" with masses of data publicly available about the "funnel": The online paths that TV viewers turn to for things such as searches. The company is vague about how it goes about gathering that data save to say that it involves no special arrangements with Google or other internet companies. The consumer data is aggregated data, and anonymized, so no one single person is being scrutinized in how they behave. It's a statistical sampling of behavior.

Also: 10 digital transformation tips to help your business TechRepublic

The point, CTO Lee says, is that it is not the data sources that make the difference: Watching linear TV and scraping data can be done by others. It is the data science, the ways that data is analyzed that matters. The data science team at EDO employs a number of machine learning technique, from simple linear regression to "more advanced techniques" involving Google's TensorFlow and Random Forests, Lee says. "We have philosophy not to over-complicate things if we don't have to," Lee says.

But the key question, of course, is whether the company is actually discovering merely correlation -- people seem to search for cars at the same time that a car ad airs -- or some kind of causality that's meaningful.

Lee insists it's the latter.

"We do feel strongly that there is a causal relationship" in the data, Lee says. "We don't have an ability to do individual attribution, it's true, because the data is aggregated, but the way to think about it is that we see a baseline established for a given granular product category."

"We see a given stimulus, the only stimulus in a certain time period is a TV ad airing -- and we see clear a departure from the baseline at that time, and remember we've seen this enough across 47 million ad airings."

How digital leaders can make the most of data and machine learning Bloomberg CTO Shawn Edwards talks about the role of IT leader and his firm's exploitation of big data.

Krim explains the company is taking a page from direct-response advertising.

"Direct-response people would run an infomercial on TV, and they would have a very detailed record of the time it aired," Krim explains. "They would watch their various, what they would call channels, such as email or Web site visits, and they would measure those changes in the baseline."

Also: 10 ways AI will impact the enterprise in 2018 TechRepublic

"We wanted to essentially bring that acuity of direct-response to brand marketers. The issue with brand marketers is they have a distributed funnel [the online venues where people search and such] where they don't own the data at all.

"We can access the gateway behaviors that lead people down funnels to purchase outcomes; we are picking up the early consideration of shopping behaviors that become new car buys or changes in consumer packaged goods preferences -- something like, I'll stop the next time at Taco Bell and ask for a Doritos-flavored shell for my taco."

Previous and related coverage:

Early AI adopters report big returns

New study shows artificial intelligence technology is paying off, but organizations face challenges.

Oracle introduces new enterprise digital assistant

Going beyond typical chatbots built for a single purpose, the Oracle Digital Assistant can be trained to support domain skills from multiple applications

AI delivering returns for enterprise early adopters, but not industries created equal

Deloitte's annual AI survey reveals a bit of realism, cybersecurity worries and a 17 percent median return on investment.

Machine learning now the top skill sought by developers

SlashData's latest survey of 20,000 developers identifies machine learning and data science are the skills to know for 2019.

What is deep learning? Everything you need to know

The lowdown on deep learning: from how it relates to the wider field of machine learning through to how to get started with it.

What is digital transformation? Everything you need to know

Digital transformation: what it is, why it matters, and what the big trends are.

AI and Internet of Things will drive digital transformation through 2020

Research study reveals IoT, AI and synchronous ledger tech (blockchain) priorities.

Related stories: