By Paul Sobel, the CEO and founder of Dataline
When advertising entered the big data era, the promise was unparalleled accuracy and precision when it comes to messaging audiences. The concept was that data can’t lie, and hard numbers tell an unassailable story.
That’s just marketing hyperbole, of course, and there are plenty of errors in this data-driven age. In fact, for all the talk of data advertising has remained dependent on humans making decisions about gathering and using those insights. . Humans must build, market, select and deploy the audience segments used in programmatic advertising. Even more robust custom models are still assembled by human data scientists.
The thing about humans is that no matter how hard we may try, we carry biases. These biases then make their way into the audience building and selection process, leading to missed opportunities and potentially wasted ad spend.
Fortunately for data buyers, AI may be able to help with that, finally helping data-driven advertising move past its human biases.
Accumulated knowledge & preconceived notions
The concepts of big data and programmatic ad buying represented major shifts in how brands approached advertising, yet many elements stayed the same. Brands still turned to agencies as their trusted advisors. Campaigns still relied on market research and past performance. Data scientists, newly elevated to a critical role, had a sense of what kinds of variables should go into a model.
All of these preconceived notions stem from experience and expectations, and they can certainly contribute to ROI. But they are also inherently limited because they are based around a human-held bias of what they believe should work.
By eliminating these biases, brands can unlock opportunities that are hiding in plain sight, including finding new customers that may be outside of their preconceived notion of their customers.
Biased composition
Let’s look at this in action. Programmatic campaigns rely upon audience targeting parameters The method of execution necessitates a brand or its agency select syndicated audiences, either directly within their buying platform of choice, or imported from a DMP.
This audience selection is often based upon a number of factors that aren’t necessarily aligned with hard data, including:
- Did this audience work last time we picked it?
- Does this segment descriptor match our desired audience?
- Can this audience afford our product or service?
Brands that want to move beyond syndicated audiences have begun using custom models. These bespoke models frequently outperform off-the-shelf segments because they are more tailored to an advertiser’s campaign needs. But data scientists still introduce biases to their models.
This isn’t always a bad thing, because often it happens when a data scientist is exercising judgment on how to best build the model. Constructing a model requires data scientists to sift through thousands of variables as they try to connect the dots based on a sense of who they think is the best target audience. The data may surface a correlation between a brand’s first-party data and an external database. If the data scientist feels the correlation is inaccurate, irrelevant, or goes against their ideas, they may remove that variable from the model.
This “human touch” happens because data scientists are making decisions built on experience. It’s important to note that this action is actually going against what the data says.It’s done with the best of intentions and based on expertise, but the decision is still biased. Sometimes this is good, and sometimes it means missing out on valuable audiences and new customers because a correlation didn’t make sense to a human.
Where AI changes things
AI capabilities have reached a point where machine learning can perform many of the tasks of a data scientist, including advanced audience model construction. AI is built and trained by data scientists, based on years of experience, institutional knowledge and real-world campaign data. Yet AI does not bring with it human opinions of what’s “right” or “wrong” for a custom audience model.
So while data scientists inform the AI, it doesn’t have gut instincts or feelings the way that humans do. AI-built models are designed specifically to take advantage of correlations between a brand’s audience and behaviors exhibited within a database, no matter how counterintuitive those correlations may seem on the surface.
In doing so, AI helps brands uncover new audiences that may have been missed or ignored because they didn’t fit into a preconceived notion of the target customer.
This may be the purest form of data-driven advertising that many marketers have ever seen. There is always a time and place for market research and the big spectacle of a broad-audience brand buy, like a Super Bowl ad. But when it comes to making ad budgets work efficiently and effectively, it’s never been easier to create audience models that are built solely around data and free of human bias.
