How You Can Improve Product Data Quality by Using AI

AI is capable of automating many repetitive processes. Especially resource-consuming tasks, such as generation, validation, and the correction of product descriptions, qualify for automation. It's crucial for companies to weigh the options contemporary AI solutions can provide.

What suffers the most when eCommerce needs accurate product data description?

One of the most critical issues that can arise when eCommerce businesses need proper product data descriptions is a lack of customer trust and satisfaction. Accurate and detailed product descriptions are essential for customers to make informed purchasing decisions. With this information, customers may feel confident about the product's quality, features, and benefits, which can lead to a lack of confidence in the business. This can result in a higher rate of returns, decreased customer loyalty, and a lower overall conversion rate. Additionally, if product data is complete or accurate, it can lead to issues with inventory management and order fulfillment, leading to additional costs and operational headaches. Furthermore, poor product data can lead to lower visibility and fewer sales on search engines and marketplaces, as the search algorithm might not be able to match the product with the customer's search queries.

Why use AI to enrich your product description and what are the benefits?

There are several reasons why using AI to enrich product descriptions can be beneficial. One reason is that AI can help improve product information accuracy and relevance by analyzing data and identifying patterns. This helps ensure that the product descriptions are accurate and up-to-date, which can increase customer trust and satisfaction. Additionally, AI can automate the process of creating product descriptions, saving business time and resources. Furthermore, AI can be used to generate personalized product recommendations for customers, which can increase customer engagement and sales. Finally, to improve the customer experience and reduce the risk of returns, AI can help identify and correct product description errors, further enhancing the customer experience and reducing the risk of returns.

"What if I told you there is a copywriter available 24/7 who can complete missing descriptions and improve the data quality of your data objects?"

 This copywriter is OpenAI GPT-3. Let's see how we can integrate it with Pimcore.

What is a GPT-3 and how can it be used with Pimcore?

According to OpenAI GPT-3 itself:

"OpenAI GPT-3 (Generative Pre-trained Transformer 3) is a state-of-the-art language processing model developed by the company OpenAI. It is pre-trained on a massive amount of text data and can generate human-like text, complete tasks such as translation and summarization, and respond to prompts in a conversational manner. GPT-3 has been trained on a diverse range of internet text, allowing it to have a wide range of knowledge and understanding."

- OpenAI GPT-3

With such powerful capabilities, GPT-3 can be used to generate product descriptions or assist with content creation automatically, and it can also be used to assist with content tagging and categorization. Today, we'll talk about how we used it to generate missing product descriptions and generate articles.

Improving data quality by providing missing descriptions

Everyone of us encountered a situation, where a lot of descriptions were simply missing. But, if the rest of the product data is there, we can generate them using AI.

The Basic idea was:

We use attributes stored in Pimcore, such as color, material, dimensions, and intended use, to create a prompt used by AI to generate text.
It can be done automatically if the description is missing, or on demand by using a button in the data object.
Generated description can be used, modified, or updated by a human editor.
We can also generate text by entering prompt manually.

There are a variety of options to do it automatically such as Event Listeners or Value Providers. Still, we decided to go with the 'on-demand' option, so we created custom buttons and models on the data object view.

After clicking on one of them, we can configure the expected length of response (max. 4000 characters). In Article Generation, you can manually enter a prompt. In the description generator, it is created automatically using filled-in product attributes.

Then in case of description generator we are simply asking AI over an API with a generated question based on product attributes:

"Create a description for a product named Cobra 427 which is a Sports car, production year is 1966, country of origination is GB, and body style is a 2-door roadster. It has 2 doors, 2 seats, rear-wheel-drive, 8 cylinders, a front located engine that generates 305 kW of power."

To make it convenient, make it asynchronous.

Receiving a response from GPT-3 can take some time, so we decided to incorporate Symfony Messenger into our functionality. If you don't know what Symfony Messenger is or how to use it with Pimcore, check out the article by Mateusz Soroka about it.

Symfony messenger allows us to complete operations asynchronously, the user gets a notification that his text will be ready in a few seconds.

When the response is ready, whether it is a description or article content, object fields will be automatically updated by Symfony Messenger handler.

Are humans obsolete? Do we still need them in the process?

Like it was already said, AI technology comes with loads of benefits, but we still need to consider some downsides and limitations. AI will not always understand every nuance, feature, or function of your products. Although it's trained on vast amounts of text data, it is not tailored to individual customers' and companies' specific needs and preferences. If the training data is outdated or not comprehensive, the generated information may not be accurate or complete. AI can also have problems understanding very specific technical language and terminology, for some product descriptions can be too repetitive, while others can be so different. Human editors should still check texts generated by AI and redact them if needed, but we can already reduce the amount of work needed to provide complete product information.

Summary

We can use AI for a number of processes, which most of them can be automated, such as generation, validation, and correcting of descriptions. Even if we only want to use AI descriptions to provide drafts of descriptions, it's a great tool. It's important for companies to weigh the benefits and downsides of using generated product information and to take steps to ensure the information provided is accurate, complete, and up-to-date in order to provide the best possible customer experience. It's worth noting that it is only the beginning of this already amazing technology; we can't wait to see what the future will bring to us in this matter and how we will use that with Pimcore!