Well-maintained product data is decisive for sales. The days, when customers relied on a single shopping channel, are over. Nowadays, they are reading up on products online and are researching carefully before they decide to buy. This is why it is so important to work with optimised product data. Especially in ecommerce, the information must therefore be meaningful, detailed and always up-to-date.
Information crucial to success
Regardless of whether it’s the possibility of even finding a product, information in a personally designed newsletter or an appealing product description – meaningful product and category information helps the customer to make decisions.
In product data marketing, vast amounts of product information are often processed and managed, used for campaigns and product text and transformed into personalised messages for customers. This also makes product data interesting for automation, because if you have a good data structure and data quality, you can also use it for automated texting with the text robot.
But not all product-related data that a company possesses is suitable for machine evaluation. In order for the Text Robot to actually use the data feed, it must meet certain criteria. Consequently, the product information must be prepared accordingly in order to obtain attractive and meaningful product descriptions by means of automated text generation.
Product data and what it can say about an item
First, let’s look at the data that most companies have anyway and allow for optimization. Often, existing master data from a PIM system or other product data management tool is well suited for automated text creation. Master data is simple, descriptive information such as product names, manufacturers, dimensions etc.
A structured overview of this data (according to the model: product=x, manufacturer=y, length=a, width=b, height=c) already enables the creation of simple sentences, which may even be used for all products of a supplier. We can already create a short text like:
The smartphone from x is available in the size a x b x c.
The bottom line is that it would of course make more sense to specify a screen diagonal here, but you understand the principle. By the way, the screen diagonal – or any other surface – could also be calculated by the Text Robot from the dimensions, thus creating completely new possibilities. In the next step, we collect data on properties and features. Structuring these requires a little more effort and this is also the point where it often gets stuck.
Product features as structured data
Most companies have a wealth of data on product features and properties. This is important, because who buys a smartphone with the above product description?
But how can the data be structured to be machine-readable? Since the assignability of data plays an important role for the Text Robot, it is important to split the product data into unique data fields. This could be, for example:
- camera: 8MP
- data_transmission: LTE
- operating_system: Android
- dual_sim: false
- touchscreen: true
These product data can then be used to create a few more or more meaningful sentences:
The Android smartphone with a touch screen has an 8MP camera. With LTE data transmission, you will be quickly connected when you are on the move.
The individual data fields may apply to several products and may contain data of completely different products for another data set. For example, for a desktop computer, the data field “operating_system” could contain the information Windows 10.
Meta data for an interpretation of information
Of particular interest to large ecommerce retailers is also the so-called meta-data, which enables the Text Tobot to work interpretatively. Such meta-data can be, for example, sales figures, return rates or evaluations of other buyers.
Accordingly, a low return rate, for example, suggests that customers are very satisfied with the quality of the product, which allows the text robot to write a sentence from this data feed such as:
A particularly popular product that is guaranteed to provide you with joy for a long time.
Other interesting sources of information for automated copywriting can also be store info and manufacturer data, which offer additional added value and therefore optimization in automatically created texts.
However, it is and remains important to structure the data optimally so that it is machine-readable. Data fields that contain several badly separated attributes are often unusable and also fields with continuous text cannot be evaluated by the Text Robot.
Product data optimisation for automated content creation
But what does the optimal and purchase-relevant data for the automated creation of product descriptions look like? If I want to prepare and optimise my product data for content automation, there are a few things to consider. Specifically, this means that I have to take a close look at my data and separate it if necessary.
In many cases, online shop owners have, for example, a product description or a product name that already contains the most important features. In the name of a sofa, for example, the colour, material and perhaps even the dimensions are already included. Even though this is practical for the header and the customer thus can see immediately all important information about the product, it is useless for the Text Robot.
Basically, such a data field contains a lot of wasted information – simply because the attributes from this data field are not machine-readable. But if I have a separate data field for each attribute (“colour”, “material” or “dimensions”), I can generate a separate sentence for each attribute. If these are then displayed in different variants and sequences, variance arises and duplicate content is avoided.
The keyword here is granularity and the more granular the data, the more flexible and multi-faceted the text can be.
Frequency of occurrence or level of completeness
Of course, there is no point in creating hundreds of data fields, which then only affect one product in thousands. For automated content creation, it is a matter of weighing up the pros and cons: Which data fields describe most of my products and what is the level of completeness when I create specific attributes?
This train of thought is not one-dimensional, because in some cases it can make sense to use an attribute that is only slightly filled in/used – namely if exactly this attribute is a special feature that I want to emphasise in my automatically generated text.
The application with shimmering pearls makes this T-shirt a real eye-catcher.
Even if there are very few T-shirts with such an application in the shop, this category (T-shirts) is especially important to me.
Granularity and structure
To begin with, create 20-25 attributes and increase their number to the rule-of-thumb value of about 90 attributes for the complete text project. With this amount, a significant variance can already be created. The more attributes that are filled in correctly I have, the more flexible and detailed will the automated text I can generate be.
It may take a little effort to prepare the existing product data and product information for automated texting by a text robot, but it’s worth it! Because with well-structured and optimized data, great search engine optimized product texts can be created for all channels that appeal to and excite customers.
Video: Types and use of data
The CEO of AX Semantics, Saim Alkan, has summarised the topic of product data very nicely in a video.