A Blog by Jonathan Low

 

Sep 27, 2017

As IBM Ramps Up AI-Powered Advertising, Can Watson Crack the Digital Marketing Code?

IBM is consolidating a lot of data and analytical capability under the Watson brand.

The question is whether the cost-benefit for potential customers is competitive with internal efforts and newer innovations that may accomplish many of the same tasks in less time and and at a more affordable price. JL

Marty Swant reports in Ad Week:

Watson promises to kick start cognitive advertising, seeking to transform marketing from image and voice recognition to big data analysis and custom content. “Brands are looking to AI (to) distinguish them and give them a competitive edge.” Integrating The Weather Company with other data IBM is hoping to transform advertising. “Weather impacts your mood and your emotions (which) are a huge input into your decision-making modality,”(But) “Watson has to prepare data before they analyze it and people are finding that the price tag is very high.”
As much as we might want to try, nobody can change the weather. However, weather does have the ability to change us—our moods, our health, our daily routines—and those that know its effects might in turn be able to affect us.
In recent years, The Weather Company, which produces forecasts for 2.2 billion locations every 15 minutes, has been using its troves of data in ways that go far beyond what’s happening outside. Since its acquisition by IBM in January 2016, the company has also begun swirling deeper and deeper into the world of advertising with the help of Watson, IBM’s artificial intelligence service that’s working on everything from diagnosing diseases to crafting movie trailers. Now, IBM is finally bringing several major components of The Weather Company’s data capabilities under the Watson umbrella with the launch of Watson Advertising.
The new division—encompassing data, media and technology services—will offer a suite of AI products for everything from data analysis and media planning to content creation and audience targeting. By integrating The Weather Company’s signature WeatherFx and JourneyFx features along with all of the other data at IBM’s disposal, the company is hoping to transform what is in many ways still a legacy business into a cutting-edge advertising powerhouse.
“Weather impacts your mood and your emotions, and your moods and your emotions are a huge input into your decision-making modality,” says Cameron Clayton, the former CEO of The Weather Company who is now general manager of IBM Watson’s Content and IoT Platform.
Watson Advertising promises to kick start the era of cognitive advertising, a field that has both legacy tech companies and startups seeking to transform every aspect of marketing from image and voice recognition to big data analysis and custom content.
While there are countless ways to use Watson—through dozens of APIs or studio-like projects that can cost millions of dollars—its new advertising division is structured into four units. The flagship service, focused on audience targeting, will utilize Watson’s neural networks to analyze data and score users based on how likely they are to take an action (like purchasing a product, viewing a video or visiting a website). Another piece of the business will use AI for real-time optimization. A third, Watson Ads, will build on a service that launched last year with a number of high-profile brands, employing AI not just for data analysis or targeting but also for content creation. As part of a Toyota campaign, for example, Watson became a copywriter, crafting messaging for the carmaker’s Mirai model based on tech and science fans’ interests.
“The Watson Ad opportunity is an exciting first-to-market idea that advances our learning opportunities in the AI space,” says Eunice Kim, a media planner for Toyota Motor North America. “Not only are we able to create a one-to-one conversational engagement about Prius Prime with the user, but we’re able to garner insights about the consumer thought process that could potentially inform our communication strategies elsewhere.”
There have been plenty of other advertising opportunities for Watson. Earlier this year, it transformed into a doctor, promoting Theraflu while answering questions about various flu symptoms. For Campbell’s, Watson put on its chef’s hat, personalizing recipes within display ads using data about consumers’ locations and what ingredients they had on hand. For a major partnership with H&R Block, Watson turned into a tax expert, deploying an AI smart assistant to help clients find tax deductions.
“Brands are looking to AI as a feature that they might add and what that can do to distinguish them, modernize them and to give them a new look and a competitive edge,” notes Marty Wetherall, director of innovation at Fallon, which created H&R Block’s campaign.
As more marketers become interested in the potential of AI, the rebrand to Watson Advertising allows IBM to separate the advertising capabilities of The Weather Company from its other less-known operations—industries including aviation, insurance, energy, finance—explains Watson CMO Jordan Bitterman. Earlier this year, IBM created the Cognitive Media Council, a group of senior-level executives from agencies and brands that meet a few times a year to shape how marketers think about the future of AI.
“I was a mobile believer early,” Bitterman says, “and I can tell you that cognitive technology and AI will be 10 times bigger than mobile.”
The forecast for AI seems to be just heating up: According to Statista, revenue from AI services worldwide is expected to grow from $2.4 billion in 2017 to $4.1 billion in 2018 to $59.8 billion in 2025. Meanwhile, the market for Big Data is projected to grow from $33.5 billion in 2017 to $88.5 billion in 2025.
“We could certainly call AI the ‘new black,’” remarks Forrester analyst Joe Stanhope. “And marketers are getting pummeled on a daily basis with AI-type things.”
One of the first agencies to sign on for Watson Advertising is UM, which has been testing several of IBM’s new offerings for clients, including an unnamed auto brand. Since earlier this year, the agency has been piloting cognitive capabilities for localized ad campaigns at scale, meshing Watson data with client stats to analyze metrics across a large number of car dealerships in a way that optimizes ad spend while also checking local inventory to see whether or not it should personalize an ad to someone in that market.
According to Kasha Cacy, U.S. CEO of UM, the company’s 100 data scientists aren’t always enough to manage the level of complexity that Watson offers on its own. She says that bringing Watson Advertising onboard allows the agency to “take the shackles off” and be more creative with how they use data, citing one client that is already using a combination of weather data, Google searches and pollen counts to trigger when media should be bought in various markets.
“There are things that we would love to be able to do that we just can’t right now because we don’t have the human or technology processing to be able to do it,” Cacy says. “That was kind of the impetus for us talking to the Watson folks, because I think as I looked across the business I saw all these places where the thinking was there, the strategy was there, the ideas were there, but the implementation was just too onerous and I hate the idea that we’re limited by implementation.”
Indeed, it seems like AI is on most marketers’ to-do lists, whether that means researching ideas, meeting with AI companies or spending actual money. However, Stanhope cautions, the space is so new that there is also a lot of room for over-hyping it.

Building brand Watson

For years, IBM and longtime agency Ogilvy & Mather have sought to explain the capabilities of AI while also humanizing it for a world not yet ready to integrate man with machine (or vice versa). First came Watson’s now famous 2011 appearance on Jeopardy, where it managed to win both the competition and the hearts of nerds around the world. A few years later, it delved into music, with a memorable 2015 TV spot featuring Bob Dylan that showed Watson bragging to the songwriter about being able to read 800 million pages per second—devouring Dylan’s entire career catalog and identifying themes in his work.
Last year, it made the first AI-created movie trailer for the sci-fi flick Morgan. This past February, Watson analyzed the work of Spanish architect Antoni Gaudi to inspire an art installation at Mobile World Congress in Barcelona.
According to Ogilvy U.S. CEO Lou Aversano, the goal has been to give Watson a “humble, friendly, ‘I’m here to help’ personality” that takes the fear out of using AI. He says Watson has become a frame of reference in the category for ways to use data across a variety of industries.
“Technology has always been and will always be there to help businesses do more to increase productivity,” he says. “You can go back to the mainframe, you can go back to the card punch, when accounting systems were computerized. And this is just the next manifestation of it. This happens to be the first time where that technology has an ability to learn. It’s not just a data input machine.”
While some marketers that have worked with Watson are pleased with the results, it’s clear that even cognitive celebrities have critics. Some experts in the AI community think the Watson brand is already overhyped, resting too much on its early Jeopardy victory without offering enough tangible proof for what it can win for clients today. Financial analysts have also been a little wary. In a report released in July, Jefferies financial analyst James Kisner said that IBM is being “outgunned” in the race to hire AI talent, with Amazon posting 10 times more AI job listings on sites like Monster.com while Apple had twice as many.
Kisner—who talked to a number of buyers, users, competitors and evaluators for the report—also found that Watson tends to be more expensive than many other AI platforms in the market. That could be a risk as other companies create their own massive data sets and offer services at a much lower price point. (In October 2016, IBM cut its API query prices by 70 percent, according to Jefferies.)
“Watson is what we call a picky eater,” Kisner says. “So they have to prepare all this data before they analyze it and people are finding that the price tag to do that is very high.”
Some industry observers say that early hype in a marketplace full of risk and confusion could damage the likelihood of companies investing in the right places. Forrester’s Stanhope doesn’t fault IBM, in particular, explaining that the company seems to have relative parity with competitors in terms of competency. On the other hand, he says IBM does seem to be all-in with Watson.
And Watson’s brand certainly comes up a lot more in IBM’s earnings as its capabilities expand. According to the Jefferies report, references to Watson in IBM press releases and prepared remarks have skyrocketed from just a few each quarter in 2013 to more than 200 in third-quarter 2016 before falling to around 100 earlier this year. Meanwhile, the company’s overall quarterly results have been less than flattering with 20 straight quarters of losses, while revenue from Watson’s unit has remained fairly flat. (Asked about earnings, Watson chief revenue officer Carrie Seifer declined to comment, mentioning only that the company is still in the “early stages” of bringing Watson to market.)

Training data

While Watson might promise a lot, it’s not a turnkey solution, to use a favorite phrase from the world of ad tech. That’s because getting the right answers requires having the right information, but it also requires asking the right questions—and that requires having the right data in the first place. Jacob Colker, an entrepreneur in residence at The Allen Institute—an AI think tank founded by Microsoft co-founder Paul Allen—believes that AI-based success is rooted in having the proper data to train algorithms with.
“Let’s say you and I are going to drive to grandma’s house,” he explains. “There are 100 ways we can get there, but over time through simulations we understand the easiest way to get there. Then you can add in reinforcement learning, where you reward the computer for making the best amount of turns.”
According to a report by Demandbase and Wakefield Research, 80 percent of marketing execs expect AI to “revolutionize” marketing by 2020. However, there are plenty of uncertainties—60 percent worry about integrating AI into their existing tech stacks, 54 percent are concerned about training employees, and 46 percent are nervous about interpreting results.
Francesco Marconi, strategy manager and AI co-lead at the Associated Press and researcher at the MIT Media Lab, says that while brands and publishers are excited about how AI can help with scale, scope and speed, the technologies are also changing the importance of accuracy in terms of targeting and content creation. Marconi suggests agencies and media companies cannot simply settle for one or two people that know a thing or two about algorithms. Rather, they need to train their teams to be proficient with AI in the same way that modern-day workers are familiar with computers and smartphones. Because algorithms are written by humans, there is a risk of human error if data is bad or programming is faulty.
“No system is going to be 100 percent accurate,” Marconi says. “So which would you rather tend towards, false positives or false negatives? In the context of advertising, would you rather a bad ad be labeled as ‘good’ or would you rather have a good ad labeled as ‘bad?’”

Watson rivals

While IBM might have been one of the earliest enterprise software companies out of the gate—it tends to get much of the public credit for its role in the world of AI—Watson is certainly not the only game in town. Around this time last year, Salesforce introduced its own AI tool, Einstein. (In fact, earlier this year, Salesforce and IBM announced their own marketing-cloud partnership between Einstein and Watson.) Less than two months later, Adobe released Sensei. Of course, there are also the social and search platforms like Facebook—which recently announced plans to open an AI research office in Montreal—Google, Microsoft and Amazon. And it’s not only the big names, either: Startups like Persado, Albert and LiftIgniter have also entered the AI space.
Another company, Amenity Analytics, says that it is able to provide Watson-like services at between one-fifth and one-tenth of Watson’s cost. The 20-person company, which has its origins in the hedge fund world, says it’s able to process media stories to help brands understand how they’re being perceived in any given news cycle—for example, analyzing every story written about Pepsi across thousands of sources.
“Think of it as ‘moneyball’ for media companies,” says Amenity Analytics CEO Nathaniel Storch. “We’re not going to tell you when Pepsi is going to buy more ads, but we’re going to tell you when Pepsi is more likely to buy more ads.”
That begs the question of whether IBM, one of the earlier pioneers in artificial intelligence, is being passed up by younger companies. Addressing industry criticism that IBM might be overhyped, Clayton points to the value in his company’s comparatively lengthy track record in AI, suggesting that some others in the field still might be trying to catch up.
“Time to learn matters. AI systems or cognitive systems learn kind of like children. You have to teach them the basics, the foundational elements, and that takes time,” he says. “And Watson has been at this for a while: Watson’s almost a teenager, whereas some of these systems are very new and are still learning.”
As for why Watson is reportedly still much more expensive than some competitors, Seifer says by offering tools meant to make clients more efficient, Watson can ensure that a big investment pays off. “We’ve done a lot of ROI analysis to make sure that this is something that is ROI positive for our agencies and clients,” she notes.
So where’s this all heading for IBM Watson and artificial intelligence? Ask any agency executive and they’ll say they’re thinking about the technology, either researching it on their own, talking about it with clients or experimenting with various budget sizes. Nobody doubts it’s the future, but how that future plays out is something that even Watson has yet to predict.

The 4 pillars of Watson advertising

Audience targeting
Of the four main components that make up Watson Advertising, the flagship is the AI-powered audience-targeting technology that will be used on campaigns across IBM’s client base. Targeting will not only be available on The Weather Company’s owned-and-operated properties, but on TV, print, radio and other platforms. There’s also a new partnership with Cognitiv, whose technology will be used to help computers learn to think like a marketer.
Bidding optimization
One of the new features of Watson Advertising will be enhanced-bidding optimization based on a brand’s KPIs. IBM says the feature will use deep learning and neural networks—a technology meant to mimic the human brain by creating connections between various data inputs—to help reduce time spent on manually optimizing digital campaigns in a way that lowers cost per action.
“I think we can materially accelerate the speed with which decisions get made from agencies and marketers,” says Cameron Clayton, general manager of IBM Watson’s Content and IoT Platform. “And that’s a huge advantage because a huge amount of time is spent trying to crunch data, and I think a media plan’s role today is about letting them run with the data and not the data running them.”
Watson ads
Last year, IBM began using Watson to create actual ads powered by AI. Since its launch in June 2016, Watson Ads have been used to leverage natural language to let users interact with ads from brands like Toyota and Campbell’s so they can get information about car inventory or recipes.
“Watson’s been learning for a while,” Clayton says. “But a big difference is Watson is for the enterprise primarily. It’s not been designed to target consumers the same way that Alexa or Siri have been.”
AI-powered planning
IBM is also partnering with Minneapolis-based Equals 3, which will use its Lucy platform to uncover extra insights and research. “What’s even better is that deep learning algorithms learn and get more sophisticated over time,” says Jeremy Fain, CEO and co-founder of Cognitiv. “This automated technology has given clients up to 20 times ROI and lowered advertising cost per sale by up to 40 percent when compared with other solutions. It has also enabled more effective advanced KPI targeting including incrementality testing, drive-to-store optimization and high-value customer targeting.”

Who’s Who in AI

Salesforce
Since Salesforce debuted Einstein, its own AI platform, last September, the company has launched 18 AI-powered features across sales, service, marketing and commerce verticals that generate 475 million predictions per day. A recently announced partnership with IBM also enables weather insights that inform customer interactions. “It’s not an extra screw that somebody needs to buy or something somebody has to learn in a deep way or you need a data science degree or a Ph.D. to figure out,” says Chris Jacob, director of product marketing for Salesforce Marketing Cloud.
Adobe
After investing in AI for a decade, Adobe debuted Sensei last year, which now gives marketers services such as personalized recommendations, predictive email subjects, budget optimization and auto targeting. “We’re centralizing and bringing in a lot of the machine learning and AI services to a centralized place which then advantages all of our operations,” says Amit Ahuja, vp of emerging businesses for Adobe Experience Cloud. “The reason it’s so important is if you look at any of these different areas, there are very foundational components across every one.”
Xaxis
For the past two years, WPP-owned Xaxis has been developing a co-pilot for AI media buying, using machine learning to predict viewability rates and decrease CPMs for North American clients. At times, the agency has been running around 50 percent of its ad spend through the AI technology, according to Sara Robertson, vp of product engineering at Xaxis. Now, they’re experimenting with custom algorithms. “It’s pretty simple once you have the pipeline existing. The big investment in the beginning is how do you process the data, execute it and push it reliably,” Robertson says. “Once you have that down, writing a few lines of code to target different success metrics is the easy part.”

0 comments:

Post a Comment