GOOD PRACTICES of AI Audience Targeting
This is part 3 of 3 in this article series about Artificial Intelligence for Audience Targeting. Read part 1 here and part 2 here.
Since the launch of ChatGPT by Open AI in November 2022, artificial intelligence functions and systems have penetrated different aspects of digital life, and audience targeting is no exception. It is debatable if the benefits of AI for audience targeting outweigh the challenges, limitations and concerns. Those pros and cons were addressed in the first two articles of this series. However, AI has come to the point of being seemingly inescapable, even for audience targeting, so it’s time to turn to the measures that can mitigate the risks.
Here are some good practices to apply artificial intelligence to target audiences as safely and effectively as possible:
RISK MANAGEMENT. The first step is raising awareness of the risks of using AI, in general and specific to the subject matter. In the second article of this series, we discussed some of the broader challenges, limitations and concerns, as well as risks more specific to the usage of AI for audience targeting. Keeping up to date with the risks managing them appropriately is a responsibility from business leaders and from any professionals making use of these technologies. A cross-department effort is the best approach: technical teams must work together with the legal and compliance teams to ensure proper risk management. This includes mapping and monitoring the risks, and ensuring that AI usage for specific purposes and with specific data sources is authorized in contract with all the parties involved. A clear risk management is essential to address challenges of data quality and data protection. Although audience segmentation is usually done using anonymized data, AI platforms used for audience targeting may scrape personal data without proper informed consent, besides other risks mentioned in the second article of this series.
AUGMENTED INTELLIGENCE. Artificial intelligence is best used to enhance human cognitive functions rather than replacing them. This means creating processes and standards where humans rigorously program artificial intelligence and validate its results. Keeping humans in control of the final output is key to minimize cultural bias and confirmation bias, as well as limitations of context interpretation. Without careful training of the AI models and validation of the output by humans, AI may inadvertently sneak harmful prejudices in the audience configuration or cultural worldviews incompatible with the real audience that the user aimed to create.
Get Ingrid Pino’s stories in your inbox
Join Medium for free to get updates from this writer.
OPTIMIZED TECHNICAL SETUP. Risk mitigation should be done in every step of the technical setup of AI for audience targeting.
Preparation. Start by auditing meticulously the data sources and artificial intelligence systems chosen, and mapping the data flows. Avoid branded data and personal data (note: the legal definition may vary by country). Beyond learning how to write effective prompts, understand how instructions hard coded on the training model can minimize risks. Consider addressing biases (confirmation and cultural), potential misunderstandings (e.g. brands and product categories with ambiguous names), data quality issues (plan for transparency and accountability) and data protection violations (minimize risks of improper personal data usage). Evaluate case by case the level of comfort with AI making inferences. Especially when trying to address data gaps, reduce the level of assumptions that lead to unreliable outputs, also called ‘hallucinations’. Most large language models (LLM) allow controlling the level of ‘hallucinations’ accepted. This improves the robustness of the model, and the accuracy and sophistication of the responses.
Refining. When using the trained model, it is crucial to have experienced humans monitoring the outputs and refining the instructions. Users should never assume that the AI knows how to interpret the source data, the instructions for the interpretation should be explained in detail. It is a good practice to always ask and insist on confirming the sources of specific information given by AI, even when the model was rigorously trained by the user. Be aware that AI is programmed to be agreeable, even if the user trains the model to not agree with everything, so ask questions and interpret answers carefully.
TRANSPARENCY. It is advisable to identify clearly which outputs were created with AI. This values both the usage of artificial intelligence, showing the adoption of advanced technology, and the human efforts and creations that did not depend on AI, highlighting expertise and craft. For example, when using AI for audience targeting, it is best to disclose and specify which report pages or which audiences were created with the assistance of AI technologies. Similar practices can help enforce trust in how artificial technologies are being adopted.
THOROUGH COST-BENEFIT ANALYSIS. Managing the risks to maximize the benefits of using AI for audience targeting, as well as for any other purposes, requires an honest look into whether the benefits compensate for the costs and risks. Using AI for audience targeting may help increase productivity and efficiency, speed up data projects, fill data gaps, reduce costs, build strategic advantage and grow revenue. It might have high initial investments in infrastructure, long hours preparing the data and training the model, and ongoing expenses to monitor and update the AI models, besides hiring or training qualified people to overlook the process. There are hidden costs to fix AI inaccuracies. Moreover, it can be costly to deal with risks such as data protection violations, ethical and environmental impact, consumer trust, and the predicted quality decline of AI technologies.
AI is a tool
If there can be a magic formula to use AI, it would be: Human Intelligence + Data + Artificial Intelligence = Insights. In that combination, AI is the least important. It is the human intelligence who guides the utilization of data and of artificial intelligence.
Artificial intelligence, as any tool, is meant to be wielded by humans. Some tools can be used well or be misused depending on how we choose to operate them. It’s the same with AI. We owe it to humanity to use it responsibly, no matter in which area we are applying AI, and audience targeting is no exception.
