Session Recap: Data & Insights—Insights Amid the Digital Noise

Session Recap: Data & Insights—Insights Amid the Digital Noise

When We Measure What We Don’t Even Want to Measure

During our June session, Dr. Marc Jungblut (LMU Munich, IfKW) dissected a core problem in the discipline: We make decisions based on a systematically biased data foundation. His analogy: Data is not gold, but ore—worthless without processing. Two fallacies were the focus of the discussion.

1. Audience: The digital sentiment landscape is not the same as public opinion.

Those who speak up are not representative. The likelihood of posting depends on status, personality, involvement, and culture; thus, an engagement rate measures only those who are already active. Furthermore, each platform represents its own sub-public. Negative content generates more engagement and is amplified by algorithms, while quiet majorities remain underrepresented. Common metrics weight every voice equally, even though opinion leaders dominate (example: a post only goes viral after Musk comments on it and CNN picks it up). Add to this the problem of authenticity: trolls, bought engagement, agency-to-agency communication, and—increasingly—AI-to-AI communication create false resonance. Thesis: Online publics are a sea of noise in which signals must be identified.

2. Metrics: We rarely measure what we think we’re measuring.

The scientific link between engagement and purchase intent is weak. The key point: we don’t really measure sentiment. The approach was designed for distinct emotions, not for simple positive-negative valence. Tools fail when it comes to irony; even AI only achieves correlations of 0.6 to 0.77 with human judgments, which are themselves inconsistent. Sentiment depends on the topic and platform, and negative sentiment does not necessarily equate to a negative impact: The award-winning Edeka Christmas commercial evokes emotion, which the tool’s logic classifies as negative. Especially on social media, what’s often measured is an artifact. Add to that the shift from SEO to GAIO/GEO: a stable ranking turns into a black box of personalized responses with no access to ground truth.

Solution and Message. Data donations open up this “black box” in a GDPR-compliant way from the user’s perspective, but they struggle with self-selection (participation rates around 20 percent). Jungblut’s recommendation: focus not on how many people interact, but on who does; question metrics; combine methods—surveys aren’t dead. The bottom line: Without data literacy, we’re measuring noise instead of the signal. The key lies with the people who know what a number means—and what it doesn’t.

From the discussion. The “ground truth” question struck a nerve; many no longer trust their human-annotated datasets. On the perennial topic of brand values, the unanimous sentiment was: difficult to measure, often dismissed. The consensus: take a more qualitative approach again. A session with fewer ready-made answers and better questions. That is precisely the foundation of a genuine data culture.