Solving the digital advertising causality problem
- 03 September, 2019 07:00
One of the great benefits of digital advertising has been its inherent measurability.
But while digital advertisers have a far easier time of understanding how many eyeballs are witnessing their messages, there is another conundrum that digital technology has yet to prove so adept at solving – causality.
A viewer might have seen an ad, and might have taken a subsequent action, but it doesn’t necessarily follow that one action definitely led to the other.
It’s a puzzle Harikesh Nair has devoted much of his professional life to solving, both as the Jonathan B. Lovelace Professor of Marketing at Stanford University’s Graduate School of Business, and more recently during his secondment as the chief business strategy scientist at China's second largest ecommerce company, JD.com.
Speaking ahead of his presentation at the Melbourne Business Analytics Conference, Nair described his role at JD.com where he used marketing science and applied econometrics to drive growth for the brand and its partner brands in China. It was work that saw him developing ecommerce marketing, pricing and advertising solutions for improved brand-building and monetisation.
The centrepiece, however, was his team’s work developing an external facing ad-experimentation platform, which facilitated scalable, self-serve experimentation for advertisers to evaluate and enhance their campaigns.
“There is huge frustration in the entire world of advertising and marketing that they don’t know what they are getting in return,” Nair says. “Digital advertising provides a vastly superior measurement and tracking and therefor that has been its source of success. But while measurability has improved quite dramatically compared to traditional media, there is still a huge problem in the confluence between correlation and causality.”
Nair’s first line of investigation was the randomised control trials and experiments used in drug testing. While these can be performed at scale in the digital world, he says they also present a significant opportunity cost, in terms of the lost revenue from those users who might be held in the control group.
“Experimentation can be actually costly, and to get exposure to users I have to bid and pay money,” Nair says.
Also, he says this form of experimentation could really only provide answers to behaviours based on whether the user had or had not seen the brand’s ad and could not take into account the behaviour had that same user seen a rival’s ad instead. The solution developed at JD.com was experimentation as-a-service, where JD.com used modern artificial intelligence-driven methods to minimise the cost of experimentation while learning the value of advertising.
“It is very hard to assess causality unless it is an experiment, and the experimentation has to be done by the platform,” Nair says. “If I put money into one platform as opposed to some other platform, what do I get? Can it show me that I am achieving my goals? So along with my team I built these external facing self-serve experimentation products, which are now available on JD.”
One example was JD Comparison Lift, which uses online experimentation to help advertisers discover a creative-target audience combination that provides the highest expected payoff. This is done via an algorithm that adaptively allocates traffic during the test so as to minimise the cost to the advertiser.
Building digital advertising momentum
Interestingly, while the results of these experiments have enabled some of China’s top advertisers to improve the effectiveness of their marketing effectiveness, Nair says JD has not seen any shaping of advertising spend. Instead, these capabilities have only served to increase the dollars flowing to digital platforms.
Nair says this visibility of effectiveness also increased advertisers’ acceptance of digital channels for brand building – something which until now has remained the domain of television.
“Delivery is a solved problem, and targeting is a well understood problem, so the next question is what am I getting out of this and how is it delivering on the various goals that I have, like building brands,” he says. “I think it is the next phase in the evolution of the digital advertising business.”
More on business and marketing analytics:
- Why we need to look inside the black box of AI
- Why Spain’s BBVA is opening up customer data analytics for all
- How predictive analytics is tackling customer attrition at American Express
- Why marketing analytics is not about ROI calculation, but innovation
- Telstra's four Ps of marketing analytics
Follow CMO on Twitter: @CMOAustralia, take part in the CMO conversation on LinkedIn: CMO ANZ, follow our regular updates via CMO Australia's Linkedin company page, or join us on Facebook: https://www.facebook.com/CMOAustralia.