Now that the reviews are automatically moderated at high quality, and the vast majority are published in near real-time, the customer experience is significantly improved. The automation has also led to reduced costs making the project a clear success.

The challenge:

• Millions of customer reviews received across 33 languages
• Every review needs to be moderated before it’s published
• Moderation is a tedious, time consuming and labour-intensive process
• Decathlon hadn’t found a solution that could analyse all languages at a high quality

The solution:

• Utopia AI can filter out unacceptable content in any language with high accuracy
• Manual moderation has been reduced by 92,8%
• The automation has led to decreased costs
• The customer experience has significantly improved

‍
‍

“We have been very satisfied with Utopia AI Moderator, and have now deployed it in several markets. Adding new languages has been fast and straightforward process, and the quality has proven to be very good regardless of the language. In the first few countries alone, human moderation has been reduced by 92,8% and the work effort of 4 full-time employees has been saved. These resources are now focused on keeping customers satisfied: The DNA of our company.”

Thomas Dujardin
Product Manager – Reviews at Decathlon

‍

1. Challenges

Decathlon values customer feedback. Each year the world’s largest sporting goods retailer receives millions of customer reviews across 33 languages. The reviews are about products, services, stores and overall customer service experience. Most of them are relevant and to the point but some contain inappropriate content such as personally identifiable information (PII), foul language, spam, irrelevant information, etc. For that reason, every review needs to be moderated before it’s published. Whilst manual moderation ensures relevance, quality, and legal compliance, it’s a tedious, time consuming and labour-intensive process.

2. Earlier approaches

Each local customer relations center handles the moderation for their country, in addition to their other duties. Due to a large number of reviews, despite best efforts, sometimes it can take even days before a review is published. This is a challenge as it can affect the customer experience negatively. Decathlon has tested automated moderation solutions to ease the human moderation burden but the challenge has been finding a solution that can analyse a wide range of languages at a high quality.

3. Objectives & proof of concept

Decathlon set out to find an effective, efficient and reliable moderation tool that can handle all the required languages. Their objective was to improve the publishing speed of reviews while ensuring high level of quality, and to reduce the overall moderation cost through automation. When they discovered Utopia, they quickly realised that Utopia AI Moderator could be what they were looking for: A high-quality, language-agnostic solution.

Decathlon wanted to be sure that Utopia could deliver what they promised in terms of moderation quality (accuracy, precision, recall and other KPIs). They asked Utopia to build two separate AI moderation models, one for French reviews and one for German reviews. The process of training the AI models took only two weeks. Decathlon had an opportunity to test and validate the performance of the models, and to quantify the benefits that Utopia’s automated moderation could bring to them on a larger scale.

4. Results

Decathlon was impressed with how well the tailor-made AI models performed, especially taking into account that the reviews often contain spelling errors, informal language, special characters etc. Essentially, Utopia AI learned to mimic their moderators’ decision making. It is now able to filter out unacceptable content like humans do, but at scale and in real time.

Importantly, the moderation process is also fair and consistent, and that’s why Decathlon didn’t hesitate to assign the routine work to the AI. Based on the great results of the pilot, Decathlon decided to expand the AI moderation to several other countries and languages.

On practical level, Decathlon concluded that the moderation algorithm can autonomously publish but can’t refuse a customer review without a human check. A very small part of customer reviews is still managed by manual moderation, and these human decisions are used to retrain the AI models on a regular basis.

About Decathlon

Sportspeople first, world’s largest sporting goods retailer second, says Decathlon in its slogan.
‍

Decathlon is a family-owned company, founded in 1976, and inspired by the belief that the best sports products should be accessible to everyone. What started small is nowadays an international sports retailer with 2080 stores in 56 countries in 5 continents.

But, with close to 100,000 employees Decathlon is not just a reseller. They have over 20 in-house brands, each dedicated to a specific sport or practice. Decathlon builds on sustainability. They are e.g. committed to eco-design, giving millions of plastic bottles a new life in their products.Through Utopia’s AI-based analytics Decathlon learned that, in general, the company’s customers behave responsibly and provide invaluable feedback.
‍