Rep Data's In-Survey Data Cleaning Helps Media1 Protect Stiegl's Brand Tracking From Systematic Bias

For Stiegl, one of Austria's most traditional private breweries, brand tracking is not just a routine obligation, it is a strategic management tool. At regular intervals, the media and market research agency Media1 conducts brand tracking studies to measure key indicators such as brand awareness, slogan recall, and advertising effectiveness. However, with the increasing prevalence of survey fraud in the industry, signs of problematic data quality began to accumulate, including unusual responses, suspicious patterns, and early indications of automated participation. For Media1, it was clear that they did not want to leave data quality in studies for Stiegl and other clients to chance.

Until now, data cleaning at Media1 was carried out manually after fieldwork was completed. Project managers reviewed response times, read through open-ended answers, and flagged suspicious cases for removal, a time-consuming and subjective process that took roughly two hours per wave.

On average, around 6 to 7% of interviews were excluded through this manual process. However, more and more cases began to raise red flags. Responses appeared that didn't match the language or tone of the target group, closely resembled content from Wikipedia or Google, or showed unusual repetition patterns. The suspicion was that a portion of the sample might include dishonest or even fraudulent participants, but there were no objective tools in place to confirm this. Uncertainty grew, as did the concern that flawed data might slip through unnoticed, potentially leading to misguided strategic decisions for Stiegl.

Media1 activated ReDem within the survey software Keyingress, where ReDem is integrated via API. From that moment on, every incoming survey participant was checked in real time, and responses of low quality were not counted as completes. Thanks to real-time quota management, re-recruitment became unnecessary.

Each participant is sent from Keyingress to ReDem for a real-time quality evaluation, and ReDem returns a pass or fail result back to the survey software. The evaluation covers several checks.

• Open-ended response analysis
• ReDem Quality Score
• Response time checks
• Duplicate detection within and across interviews

The use of ReDem delivered two key advantages.

• Time savings of 75%. The manual cleaning effort dropped from around two hours to less than 30 minutes per wave, freeing the team to spend more time analyzing and interpreting results.
• A significant improvement in data quality. With ReDem in place, 16% of interviews in the Stiegl brand tracking study were identified as low-quality responses, more than twice as many as had previously been detected through manual checks. In other projects since adopting ReDem, removal rates have ranged from 10 to 18%.

Had these low-quality cases remained in the dataset, they would have significantly distorted key metrics. Aided brand awareness would have appeared 20 percentage points lower, a clear deviation from the steady trend observed over recent years. Aided slogan awareness would have falsely suggested a drop in advertising effectiveness, showing a decline of 18 points. Consideration, the willingness to purchase or try the product, would have been overstated by 10 points. The most extreme distortion would have been in TV spot recognition, which would have shown an increase of 26 points, completely disconnected from the actual campaign performance.

These deviations clearly demonstrate that low-quality data doesn't just create noise, it introduces systematic bias that can fundamentally distort the interpretation of a brand tracking study. For Media1, data quality isn't a nice-to-have, it's a strategic imperative.

"The risk of systematic bias was significantly lower in this year's brand tracking study due to the use of ReDem. We are pleased to have implemented the tool, not only to obtain more reliable data but also to ensure plausible results in our longitudinal comparisons."