Measuring the impact of OOH advertising in the Swiss Alps

In 2024, intervista was commissioned by APG|SGA to investigate the advertising impact of campaigns in winter sports resorts in Switzerland. The innovative research design included two essential components. On the one hand, smartphone-based GPS tracking was used to continuously measure contacts with the advertising spaces and calculate campaign reach. On the other hand, a survey was conducted after the campaigns and the survey data was analyzed in combination with the measured contacts. This research approach made it possible to understand the advertising impact in relation to the contact dose. In addition to the methodology, the presentation will also show concrete results.

Goalsetting

The primary goal of this project was to explore the broader use of Generative AI (GenAI) in consumer insights, enhancing accessibility, efficiency, and marketing decision-making. In preparation for Samsung’s Galaxy Ring launch, we developed an AI model with two key purposes:

1. acting as a subject matter expert on the smart ring category, capable of answering business questions, simulating qualitative research, and predicting consumer behavior.
2. generating virtual personas to simulate potential buyers, supporting strategy and GTM activities.

Process

Developed iteratively with a market research agency, the model was trained on proprietary research, stakeholder interviews, global marketing inputs, and social listening data. Validation tested its ability to replay ingested data, simulate buyer personas, and predict selected variables using holdout survey data. After successfully simulating responses on a nationally representative level, further validation focused on key buyer personas.

Being the first AI model of its kind at Samsung, we prioritized reliability and truthful answer behavior, ensuring the model was ready for practical application. It was officially introduced to the Samsung Marketing team in November 2024.

Results

The model showed strong qualitative and quantitative validation (R² > 0.9 in most cases). Confidence in its accuracy led to formal adoption within the marketing team. Ongoing usage is monitored, and findings will be shared at GOR 2025.

Added Value

This session provides a practitioner’s perspective on developing an AI-driven, RAG-based model for marketing decision-making. It contributes to best practices in GenAI by reflecting on development, validation, and implementation. Additionally, it explores how AI-driven insights can transform marketing and consumer insights collaboration, fostering more data-informed, impactful decisions.

Relevance & Research Question:

Dual process theories suggest that the majority of human decisions are made by System 1, which works automatically and largely unconsciously. This system leads to highly habitualised behaviour in which consumption decisions are made relative to other options, regardless of the product category. Therefore, an optimal research design should not only create realistic decision contexts and be based on established experimental methods but also consider consumers’ implicit decision-making processes. We will present an implicit Discrete Choice Model (DCM) that is suitable for modelling System 1-influenced decisions.

Methods & Data:

To this end, we had participants complete a DCM and, subsequently, evaluate the concepts chosen in the DCM tasks (i.e., winner concepts) by means of a Single Implicit Association Test (SIAT). This allowed us to specifically measure the respondents’ response speed when indicating whether the concept represents a purchase option or not through a number of cognitive, affective and behavioural statements towards the respective concept. The SIAT is the most commonly used method for this purpose.

Participants’ responses and response times measured in the SIAT were finally transformed into a linear function of associative strength and used to calibrate the utility estimation of the DCM. By incorporating subconscious associations into the choice model, this approach enhances predictive accuracy, bridging the gap between stated preferences and real-world purchasing behaviour.

Results:

Two online studies – one on “city trips” and one on “smartphones” – demonstrate the impact of integrating SIAT into DCM. Simulations reveal that incorporating measured response times enhances the predictiveaccuracy of purchasing behaviour, capturing decision-making dynamics beyond traditional models. Since DCM is typically influenced by System 2 processes, this integration allows for the inclusion of more intuitive, implicit preferences. The results show that combining SIAT with DCM provides deeper, more meaningful insights into consumer decision-making, bridging the gap between stated and subconscious preferences.

Added Value:

Our approach enhances DCM by integrating SIAT, capturing subconscious biases often missed in traditional models and improving predictive accuracy.