3rd May 2023
Ewa Solarz, Global Channel Solutions Lead, Mindshare
So whilst there has been a recent boom in AI large language models, such as Chat GPT, it’s pretty clear we’re not quite at the singularity yet. However, the ‘rise of the machines’ is inevitable and so more and more businesses are looking at ways that AI can be implemented into their business processes. Doing this correctly promises huge opportunities, massive efficiencies and savings and another way to beat the competition. Doing it badly however, could cause long-term damage to your brand and business.
So how do you avoid the most common pitfalls when implementing AI for advertising?
If your data set is largely skewed to towards certain demographics, over-indexing on specific audiences it is most likely excluding those who ‘don’t fit’ the AI-based profile. For example, you may end up targeting only wealthy, middle-aged Caucasians for luxury cars. Diversity, scale, and quality of input data are fundamental characteristics that determine the predictive effectiveness of AI models.
You should therefore assess and evaluate any outliers, subcategories, or data points you do not want to bring into your systems that may introduce bias - such as unfiltered user-generated data could lead to inappropriate language used in content or data correlation-based assumptions could result in oversampling and affect targeting.
Even if you have a global presence, you should always choose localised vs global targeting to ensure a nuanced approach to marketing your products and services.
AI won’t understand cultural sensitivities or differences e.g. targeting a product that includes animal-based ingredients to people from cultures that are plant-based. A global approach can translate a global bias to local delivery, resulting in an insensitive approach, even if unintentional (and also resulting in wasted marketing spend as your product is not relevant to the audience).
Identifying and mitigating bias within our creative and content-based communication is key. Optimising messaging-based engagement across a broader subset of the audience will not only help avoid inaccurate representation of your target audience and so damage to the customer relationship but it is also likely to encourage the target audience to be more connected to the brand.
In research conducted by Microsoft advertising almost two-thirds (64%) of people say that they were more trusting of the brands that represent diversity in their ads. Mindshare’s Empathetic Executions, informed through Precisely Human Intelligence, brings together accuracy and empathy to deliver more relevant messaging and media activation, better creative storytelling across social media platforms, more precise targeting and is proven to increase creative effectiveness for brands.
AI can be a tremendous ‘shortening’ technology, something that can take care of things that would otherwise take humans much longer…but it is always better to have a human being with oversight.
Make sure you deploy diverse platform, targeting and media buying strategies, report over and under delivery against certain audiences to publishers and revise your exclusion and inclusion lists regularly.
Work with your media agency to engage with the big tech players offering AI services. Help to explore if there are opportunities to influence how models are trained, away from just sensitive variables, such as race, age or other identity-based attributes.
Open discussions about fairness testing, sensitivity analysis, and counterfactual analysis. Ask to participate in the development of new tools and tech built specifically to reduce bias. e.g. Meta's new AI-based Variance Reduction System (VRS), designed in collaboration with the US Department of Justice in the wake of a discrimination complaint, aims to reduce bias and increase the equitable distribution of ads for housing, employment and credit. Only by working as an industry will we get the AI that we all need.
One of the biggest threats that face the development of AI in advertising is that we transplant human bias into the AI machine. As in any algorithmic model, the quality of data and its source is paramount to the optimisations within the model. Follow these five rules and you will go a long way towards developing AI in a positive way for your business and also helping to create a better use of AI for the wider industry and ultimately, the world.