“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Over the past few years, advertising has become far more data-driven. AI is playing a large role in the transformation, helping advertisers measure campaign effectiveness and transform data into actionable insights.
But AI is far from infallible. The technology reflects human biases. That’s why, to make the most of AI, advertisers must diversify data analytics teams to bring varied perspectives and talents to data collection and decision-making. Then, advertisers can combine artificial and human intelligence to maximize the value of each.
Here’s a road map for how to do just that.
Understand and mitigate AI’s limitations
To understand AI’s limitations, consider this year’s Australian Open. The tournament used AI to process match data in real time and make predictions about probable winners. In the men’s final, Rafael Nadal was down two sets to love against Daniil Medvedev. He was projected to have a 4% chance of winning. But Nadal defied AI’s data-driven predictive capabilities and won the next three sets to clinch the match.
The same applies to AI in advertising. If we ask a consumer survey question in a different way, we get different answers. If we scan a data set for certain demographic information but make omissions, we risk coming up with skewed analytics and suboptimal decisions. To that end, advertisers need to be aware of AI’s blind spots when using the technology for marketing analytics.
One crucial step the data scientist should take to ensure the highest possible accuracy and quality of AI is to actively recognize any selection biases in the data collected and use randomization or statistical correction to remove it.
This is especially true for survey data. Certain survey formats and media attract particular types of participants. Various incentive programs offered by panel-based research organizations may also affect respondent composition. Data scientists need to think about the sample composition rather than just what the data says.
Diversify data analytics teams to minimize AI’s blind spots
People work differently when they see a business challenge or hypothesis, and their backgrounds and past experiences inform their approaches. Only by diversifying teams, and the problem-solving approaches they prefer, can we become maximally competent in accounting for all possible solutions.
Let’s say a market research team is designing a survey to understand how customers of different genders respond to advertising. A team without anyone who identifies outside the gender binary can fail to account for gender fluidity and nonbinary folks. If the product caters to people of a certain gender or to young and urban populations who often identify as nonbinary, the failure to diversify teams can skew research and undermine the precision of analytics.
Still, diversity is lacking. Nearly six in 10 marketers are white, according to a survey by the ANA of 61 of its 1,400 member companies. Surveys rarely include options for respondents to identify as nonbinary. And anyone who’s worked in marketing knows that diversity drops among key decision-makers. For example, people identifying as female account for 75.1% of admins and 70.8% of entry-level marketing professionals, according to the ANA, while they account for just 54.8% of senior management.
Agencies must tackle the diversity issue. That means not just hiring diverse staff but also diversifying at senior levels.
Combine artificial and human intelligence
In advertising analytics, the primary barrier to maximizing the combined value of artificial and diverse human intelligence is multiple-choice-style market research. The advertising industry needs to collect information about customers through more open-ended experiments and surveys that account for nuance. Then, it can use AI-driven text mining and other tactics to transform more unstructured data into actionable insights.
Another area for improvement is the use of analytics to understand customer emotions. Feelings are too complex to be articulated in terms of binary data. Natural language processing tools can equip teams to analyze free-form customer expression on public platforms, assessing such varied questions as how much buzz a campaign is generating, how people feel about a new brand or product, or how customer sentiment has evolved since a major corporate event. Analysis of unstructured data removes bias brought on by humans and enables a more accurate data-driven approach to audience analytics and marketing.
These days, advertising analytics teams and market researchers eliminate human ambiguity from the front end, forcing customers to fit into binaries in surveys that don’t reflect their complexity. Then, when it comes to making decisions based on oversimplified data, we rely too much on human intuition, introducing error and magnifying bias.
Ideally, the opposite would be true. The advertising industry should leave more room for ambiguity in information collection and lean on technology more when it comes to interpreting data, calibrating the media mix, and determining attribution. If we can move in that direction, advertising will be on its way to becoming a more equitable, representative, and truly data-driven business.