The debate over AI versus humans is a long and storied one, with endless arguments and evidence on either side. In the last few years, machines have become efficient. Automation has largely replaced rote and mechanically-focused jobs. However, many believe that automation can never take over jobs that require human reasoning and creative thinking. Thus, we find it difficult to envision machines writing compelling content. However, the reality is that machines are already writing content and they are getting good at it. A popular example of artificial intelligence for content creation is Quill, an NLG platform developed by Chicago based company Narrative Science. Using Quill, you can automatically generate anything from stock market reports to sports articles.
Content is what ultimately sells products online. While design can grab buyers’ attention, and attractive prices can influence decisions, customers won’t buy products they know nothing about. Creating product content is the most challenging and time-consuming activity while making your E-commerce website. In general, the product content creation process involves the following steps :
Fig: Product Content Development Process
Wrong or less detailed product content will put off customers. The desire to automate product content creation is thus strengthened because it aligns with the goals of the process i.e., to improve product content quality and to increase quantity.
Every step in the product content creation process presents some unique challenges while automating the process.
In this step, you need to search for products using a unique Product ID on their brand website, retailer websites or marketplaces (in decreasing order of trustworthiness) to gather relevant product data. This raw product data is unstructured and unclean. The challenge here is to make the AI learn how and where to look for product data.
This step involves extracting attribute information from the raw product data and inserting those attribute values in their corresponding fields in the attribute sheet. These curated sheets created here must contain detailed, accurate and standardized attribute values for each product (this is the content that will be displayed on your product page). In practice, per day attribute sheets are filled for thousands of products. Each product has 20-50 attributes. Here, automation is tough simply due to the volume of the products and the number of attributes per product.
Creating quality copy for products is crucial for your E-commerce website as it leads to better user experience and also makes your site more trustworthy. This is absolutely beyond the scope of the current AI developed by Narrative Science as product copy writing involves identification of best product features combined with polished writing to quickly and precisely make the customer understand the “need” of the product.
Images and videos are the most compelling ways to engage customers and can be used to convey vital product information and thus drive sales. There’s enormous scope for automation here as current AI’s are very capable of image editing tasks such as scaling, retouching, color correction, watermark removal, etc.
Now that we are aware of the challenges, let’s take a look at some of the ideas for completely automating each of these steps.
Every idea listed above cannot be materialized with today’s technology. However, some of the aspects of the product content creation process can be automated using current Artificial Intelligence.
You can use a dictionary based approach to group similar attributes. Ngram models are used to extract the attribute value for products.
You must input all the available product attributes to help the AI classify the product into an existing category or create a new one. Currently, AI’s developed in this area can categorize up to half million products per day.
A rule-engine takes product data as input and performs checks such as formatting checks, syntax and guideline checks and consistency checks and highlights the errors so that they can be resolved as soon as possible.
As discussed earlier, automating copy writing is not possible with the current technology but you can use AI for spelling and grammatical checks, presence of restricted words, etc.
Category, size and color are the product attributes that can be identified using current AI’s for image processing.
OCR algorithms are helpful in identifying and removing unnecessary text from product images.
Product content creation is currently beyond the scope of NLG platforms. While complete automation of the product content creation process is difficult, automating certain aspects of this process such as partial attribute filling, product classification, image processing and quality checks can be implemented. Having an optimized semi-automated product content creation process will be beneficial as productivity will improve, accuracy will rise and TAT (Turn-Around Time) will be smaller. It’s time for you to understand that automating product content creation is not a radical concept but something you need to do to make sure your business survives in the long run!