How To Make Your Product Data Fit: 2 Experts Provide Tips
Product data is the be-all and end-all in e-commerce business. The quality determines whether the purchase in an online store will be a success or a failure. High-quality data with complete product information ensures that the online shop can unfold its strengths. Poor quality product data are the reason why many companies are left far behind their capabilities. Online retailers should have the utmost interest to optimally prepare their product data for the online store. It is important to not optimize the data now and then, but continuously, in order to provide the potential customers better purchase advice.
Rosella Wenninger, CEO at uNaice, and Stefan Sprenger, founder of DataCater explain in the interview why high-quality product data is important in the e-commerce sector and which role does the currentness play. In addition, they explain what does a successful data preparation comprise and consists of and how meaningful data structures look like. Furthermore, they also discuss which role do optimized product data play for automated content generation and to which degree can the software and technologies help shop operators in creating better content.
What is structured data, and what do we need it for?
Rosella Wenninger: “In the context of automated content, structured data means clean data that is made available for the content creation process in such a way that it is tangible and can be processed with little rules and programming requirements. It is important to note that only one statement may be contained in an attribute at a time. All data must also be presented in the same grammatical form. For example, if the data field material is filled with wood, all other data fields must also be filled with nouns and not adjectives or verbs, such as wooden or plastic.
Structured data also includes a clean separation of information. Let's take an example from the fashion industry: if the data is "delicate silk," delicate and silk must first be separated before they can be processed. Structured data is the basis for automated content generation. The more structured the data is, the less complex the programming and the more flexible the design is for us."
Stefan Sprenger: "Structured data is data in a format that can be processed automatically by algorithms.
In the context of search engine optimization, website content, for example product information, can be made available as structured data so that search engine crawlers can enrich the indexed content of a website with further information, such as the price or rating of a product, and present it accordingly in the search results.
Structured data typically leads to better search engine rankings (e.g. google), more traffic, and also higher sales conversions."
Why is the quality of product data important for e-commerce sales?
Rosella Wenninger: "The quality of the product data determines the quality of the product descriptions. There is a lot of competition, especially in e-commerce. The more accurate the statements in the content are, the better a customer can make a decision. The customer's decision-making process can be positively influenced by good product data by providing helpful information, e.g., whether a washing machine is loud or quiet.
The expressiveness of the texts depends on the product data, which is the basis of the texts. USPs can also be better texted with good product data. Good product data also plays a role in visibility, as search engines honor the added value in terms of content. Of course, the prerequisite for this is that all texts must scale at a relatively high level."
Stefan Sprenger: "The higher the quality of the product data, the better search engines can index the corresponding products. Poor product data leads to products not being found at all in search engines, or only with considerable effort. Up-to-date and correctly entered product data also leads to fewer returns and higher customer satisfaction in the online store, as customers experience fewer surprises when unpacking the goods. The quality of product data is very relevant not only for consumers, but also in the B2B sector, where inadequate product data affects the entire supply chain."
What are the reasons for poor product data in e-commerce?
Rosella Wenninger: "Often the cause of poor product data lies in the different sources from which the data is taken. It often happens that our customers' data is provided by different manufacturers. As a result, the data is not uniform and must first be merged. However, non-structured data does not necessarily have to be bad. In other areas, other requirements are placed on data, so that unstructured data can also be sufficient.
Often, historically grown data is not structured because there was no need for it. However, as a basis for automated content creation, this is not sufficient, and therefore inferior. On top of that, there is a large volume of data that many companies can't handle because of the very high manual effort."
Stefan Sprenger: "Often, data is already delivered by the manufacturer of the product in a bad format and then has to be laboriously prepared. In addition, many e-commerce companies face the challenge of preparing and integrating product data from not just one, but numerous manufacturers and delivering it in a uniform format. One example of this is shoes, where size specifications are often available in different formats and have to be adopted by the store operator."
What role does the up-to-dateness of product data play?
Stefan Sprenger: "The up-to-dateness of product data is immensely important. We have been used to being able to access all information on the internet at any time for many years. Unlike traditional retail stores, online stores are open around the clock. Therefore, up-to-date product data should also be available at any time in order to offer the best possible buying experience."
Rosella Wenninger: "Up-to-dateness of product data can play an important role in the food sector, for example. If ingredients are changed and not updated, this not only leads to incorrect product descriptions, but can also lead to legal consequences. Regardless of how up-to-date the product data is, my tip is to always replace product descriptions on a regular basis. If we understand topicality in the sense of completeness, this of course plays an essential role. Only with complete data, we can generate great content."
Where does Product Information Management (PIM) come up against limits?
Stefan Sprenger: "Without going into technical differences and specifics of different products, I see challenges primarily at the organizational level. High prioritization of product data is crucial for the success of PIM systems."
Rosella Wenninger: "Every PIM system has different prerequisites and thus also comes up against different limits. The same applies to API connections. So it is mostly a matter of individual limits that cannot be generalized. This is often also related to the know-how or the volume-time budget of the IT people who are responsible for the PIM system."
What does data preparation for store operators involve, and how does it work?
Stefan Sprenger: "Basically, data preparation is about converting raw data into a high-quality, uniform, and cleanly structured format so that the best possible data is available to downstream applications. Specific tasks include normalizing product attributes, for example, merging different spellings of product colors, correcting formats, fixing errors in the data, or enriching product data with additional information.
Data preparation can either be performed manually by the store owner, which is not only time-consuming and costly, but also error-prone, or automated, with the help of a data pipeline."
Rosella Wenninger: "Data preparation is individual for each store owner. Store owners who do not have structured data, for example, must first determine which content they want to generate and "create" the data for it accordingly. This can be done either manually or through an automated process. Automation of data preparation is possible if the store operator can provide product information (such as manufacturer data for products) in the PIM/ERP or via other sources, and the product information can be read out for each product and enriched to the desired attributes. For example, "red trousers from XY" then becomes "color: red; product type: trousers.
The data preparation is always based on a data model (i.e. the attributes selected by the store operator, which he needs for his target text).
If permanent data preparation is necessary, I always recommend automating the whole process to save time and therefore money. For this, the store PIM has to be connected via API to the data preparation software, which in turn is already connected to the API of AX Semantics. Then, only a configuration of the API to the respective customer project is needed."
Which data structure makes sense for product data and why?
Rosella Wenninger: "The data structure depends on the requirements for the content. A very detailed product description with many attributes also requires more attributes than a short text. The prerequisite for the data structures is that they contain data specifying a category. Both the main categories and the subcategories must be present. For example, the main category for Fashion would be Clothing, and Dress would have an associated subcategory.
The data structure must also contain the main attributes needed for a text. If the material is to be highlighted in a text, then, of course, the data structure must also contain the attribute Material. I always recommend a manageable number of attributes. If there are 300 attributes, then 250 attributes are too many. However, the cost-benefit ratio must always be weighed here. If a long, detailed text is needed, then, of course, many attributes can be useful."
Stefan Sprenger: "This depends very much on where the product data is used. While PIM systems often manage product data in ontologies, many other systems work with tabular structured product data."
Why is high-quality and optimized product data important for the automated creation of product descriptions?
Stefan Sprenger: "The more up-to-date and high-quality the product data is, the better product descriptions can be generated. In this context, you also hear more often about "garbage in - garbage out". Similar to a search engine crawler, an NLG application, such as AX Semantics, is software that processes data in an automated way and benefits from high data quality."
Rosella Wenninger: "With high-quality and optimized product data, product descriptions with added content value can be created. The more concrete and high-quality the data is, the better the texts will be. Let's take another example from the fashion industry: if only one data field is available for a dress, only one text with one attribute can be generated from it, such as: "The dress is made of cotton". However, if additional attributes are available, such as "Indian" and "organic", the text can also be filled with more details, and it receives a greater added value.
With good product data, programming and conception can also be made more efficient, which also leads to cost savings."
How do uNaice and DataCater support store owners in data preparation for automated product descriptions?
Stefan Sprenger: "Thanks to DataCater, up-to-date and high-quality product data is available to the AX Semantics NLG Cloud at any time. Thereby, not only high-quality product descriptions are generated, but also an excellent user experience is offered by a high topicality of the product descriptions.
DataCater is the no-code platform for streaming data pipelines and allows users to connect a variety of data sources (online stores, PIM systems, etc.) to AX NLG Cloud. DataCater detects changes in sources in real-time and immediately transfers them to AX NLG Cloud. Data can not only be transferred, but also prepared in real-time while streaming. More than 50 no-code transformations, as well as Python-based data transformations, allow for the time-efficient implementation of any product data preparation requirements."
In this Meetup video, Stefan Sprenger from DataCater and Alexandra Waldleitner from AX Semantics use the concrete example of a furniture online store to show you how data and content can be kept up-to-date without manual effort and how content creation can be automated.
Rosella Wenninger: "uNaice supports store owners throughout the entire data preparation process. We have developed a structured process in which the data is first analyzed and checked by our data experts. In a joint feedback meeting, we provide the customer with maintenance recommendations and tips on how the data should be prepared for optimal formatting. The briefing or the objective of the content is always in the foreground. We always pursue the goal of achieving the best possible result with the minimum amount of effort.
If the customer has major deficiencies in his data or does not have the right attributes, we create a data model together. Some customers also obtain their data from many different sources and are unable to perform the data preparation themselves. In this case, we take over the entire data preparation.
We automate everything from data preparation to content creation: data is provided by the customer via API and the generated texts are sent back to the customer via a REST API - in real-time if desired so that the manual effort required by the customer is kept to a minimum. Once the project has been successfully set up, however, our collaboration does not end there. We continue to work closely with our customers afterwards and support them with customizations and changes."
In the Meetup video, uNaice reports on the optimization potential offered by automated category pages descriptions, how all relevant keywords can be automatically integrated into category pages with the help of an SEO tool, and how better search engine results can be achieved with an intelligent ontology and integrated flexible keywords.
With high-quality product data, you improve the quality of product descriptions to a great extent and generate content with added value. Not only are high-quality product descriptions generated, but the user experience is improved enormously by keeping the product data up to date. This ultimately leads to an increase in the conversion rate of your website.
If you also benefit from the advantages of automated product descriptions, then the texts can be created particularly quickly and for a large product range. The automated content generation with the natural language generation software from AX Semantics is data-driven - each text is based on a data set with structured data. This product data with the product information forms the basis for the high-quality product descriptions.
The user creates logic, statements, and variances in the desired style and the software uses these to create natural language content. Thousands of unique and optimized product descriptions are created in no time by automating the repetitive parts of the writing process.