Identify flop products before they produce costs

Finding the Rotten Apples in your Innovation Pipeline

Survey-based Opinion Mining

Tchibo, a prominent multi-channel and multi-category business, faces the unique challenge of creating an all-new product assortment in the NonFood sector every week. Products that do not achieve a specific sales target result in significant costs. To mitigate these costs, Tchibo has implemented a proactive strategy: testing new product sketches through online surveys before even beginning production and planning quantities. This helps gauge potential customer interest and reception. By utilizing an advanced predictive model developed by Cauliflower, Tchibo can assess and anticipate the success of its products. The model takes into account historical sales data, as well as valuable customer feedback in the form of open responses on likes and dislikes. Thanks to the insights from the Cauliflower model, Tchibo can now identify potential "flop" products and pinpoint their problematic features, all before committing to full-scale production. This strategy has proved instrumental in cutting unnecessary costs and refining product offerings according to customer preferences.


Online Survey / Market Research



Tchibo operates around 900 shops in 8 countries, over 24,300 retail depots and national online shops. In addition to coffee and the single-serving systems Cafissimo and Qbo, the company offers weekly changing non-food ranges and services such as travel and mobile telephony via this multichannel distribution system.

/ Background

Path to success

Tchibo faced a unique challenge in its NonFood sector: managing and optimizing a product assortment that changes on a weekly basis. At the foundation of addressing this challenge was a rich dataset consisting of historical sales numbers across different assortments and detailed results from online surveys. These surveys offered a deep dive into the respondent's demographics, their buying habits, and feedback on tested product sketches. Notably, participants evaluated product sketches based on their likeability, purchase intent, and provided open-ended feedback highlighting specific likes and dislikes. To bridge the gap between predictive analysis and real-world outcomes, the actual sales figures of these tested products were also incorporated. Drawing from this information reservoir, a sophisticated predictive model was devised. Its core capabilities centered around detecting specific characteristics of products that historically underperformed and evaluating the probability of a product failing in specific sales channels. The true efficacy of this model was measured by comparing its forecasts with the actual sales data, ensuring its predictions were not only accurate but also actionable. This approach provided Tchibo with a powerful tool, sharpening its product strategy and mitigating the costs of potential underperformers.

/ Added Value

The prediction model is applied for each new assortment with its products. The results are channel-specific predictions of product success and a summary of the likes and dislikes as concrete feedback on the most relevant positively and negatively-perceived characteristics of a product. The results are presented in an intuitive dashboard solution and are the basis for decisions regarding product design, sales volume, placement in the sales channels and product promotion. The single-minded indicator concerning product success combined with actionable qualitative feedback creates the winning solution for product management. While the statement about the product success defines the framework for action, the results of the text analysis capture the free associations and improvement- wishes of the respondents and transform them into actionable guidance for optimisation.

Conclusion: The Competitive Edge in an Evolving Marketplace

In today's fiercely competitive landscape, having a distinctive edge is paramount. For Tchibo, that advantage has been realized through the meticulous product testing and accumulation of knowledge over the years. Every product evaluation and every piece of feedback has been a building block, systematically contributing to a vast reservoir of knowledge. The Opinion-Driven Prediction Model, with its advanced embedding driven classification model, is not merely a technical marvel; it's a repository of this vast wisdom, encapsulating years of consumer insights and preferences. By continuously refining and embedding this wisdom into the model, Tchibo has positioned itself uniquely, ready to predict, adapt, and respond to ever-shifting market demands. As competitors grapple with the challenges of the marketplace, Tchibo stands a cut above, fortified by a legacy of deep consumer understanding and a future-facing tool that leverages this knowledge. The result? Enhanced product offerings, attuned to consumer desires, and a formidable stance in a turbulent competitive field.