The Power of Data: How AI Creates a Shopping Experience from Product Data

It's no secret that good product information is invaluable. Successful companies understand that it's not enough to have a great product. They need to differentiate themselves from the competition and the crowd. Good product information helps to communicate the benefits and features of the product, giving potential customers a complete understanding of what they are buying. As a result, customers are generally more satisfied with their purchases, leading to higher customer ratings and positive referrals, which directly impact sales.

An essential aspect of marketing is the visibility of a product in search results, where good product information plays a central role. Search engines scan websites for relevant content and rank them accordingly. So, if your company is careful and accurate with its product information, it is more likely to be found by potential customers during their searches. This will drive traffic to your website and ultimately increase sales.

Similarly, companies must ensure that their product content complies with relevant regulations, including consumer protection laws, advertising guidelines, and general legal requirements. Failure to comply with these legal requirements puts companies at risk of legal action and penalties. Unclear or misleading information your company provides creates false expectations and disadvantages customers. Product information must be carefully reviewed to ensure compliance and accuracy. By following these requirements, companies can protect themselves from potential legal consequences and gain the trust of their customers.

Many of the processes described here – from onboarding supplier data to presenting product data in the web shop – often limit retailers' ability to expand their assortment or act as a platform. AI algorithms can help overcome these limitations in several ways.

On the one hand, expert systems can map human decisions directly into algorithms. For example, text can be analyzed for specific patterns to help map supplier data to the retailer's PIM structure and attributes. This system does not necessarily have to make fully autonomous decisions from the start; instead, it can be a decision-support tool for humans. Relevant suggestions can be presented through dashboards, allowing humans to define a variety of mappings in a short amount of time.

In addition, AI systems can autonomously make connections and conclude existing data. For example, if SKUs are already grouped into products in a system, machine learning algorithms can identify and derive common criteria for this grouping. A wide variety of supplier catalogs can be processed without further human intervention.

If data is not yet available, there are also ways to train algorithms with manageable effort. For example, suppose a retailer wants to differentiate between clippings of mood images or technical drawings. In that case, an algorithm can be trained by labeling a few hundred images and then easily trained to label entire image databases containing hundreds of thousands of images with sufficient accuracy. Systematic errors in automatic classification can be corrected by targeted re-labeling.

Significant progress has been made in recent months, especially around large AI language models. For example, providers like OpenAI with ChatGPT offer the possibility to generate appealing product texts from the specification of product attributes – in the desired style and length. However, ChatGPT is only the most visible product at the tip of the iceberg of many different language models.