Why is AI particularly useful for PIM (product information management)?
Around 70% of all product data is text-based. With large product volumes there’s a lot of text to create, update and maintain. This is usually not a task that your staff is eager to take on. AI tools leveraging large language models (LLM) and natural language processing (NLP) technologies are now capable of generating product data at a good enough quality level to replace humans manually registering and updating information.
The good enough quality level is in fact another reason why AI is a perfect fit for generating product information. Often, you don’t require texts of the highest quality for your product descriptions. It needs to be accurate and grammatically correct of course – but not top-notch.
I’ve spoken to businesses who claim they probably will get better quality and consistency when an AI tool generates the product descriptions because the option was to have a large number of freelancers filling out product data fields for products they were not experts on and struggled to keep a consistent tone of voice.
Which organizations can make use of AI for PIM?
I think any organization could make use of AI in their PIM workflow, but probably in a few different ways. Large international organizations are the ones that will gain the most short term. These handle huge volumes of product data from lots of different sources with large variations in format and quality. Thanks to AI, these businesses can go from poor (or missing) product data to a pretty good baseline quickly.
Plus, businesses who want to localize their offering should use AI translation tools. Here, AI truly outperforms the previous online translation tools because it understands the context, nuances and learns how to interpret text from large datasets. This way, companies can avoid the embarrassing ‘lost in translation’ product descriptions that we’ve all seen hilarious examples of in the past.
What other use cases do you see for AI and PIM, beyond generating text?
Well, the sky is the limit…! I see AI as a solution looking for a problem to solve. That means we’ll discover a lot of new use cases in the future. And many of these are already possible today.
One example is the usage of AI plugins. These extend a chatbot’s capability by connecting it to specific data sources you control. Here at Bluestone PIM we’ve given ChatGPT access to our entire documentation. Our goal is that businesses can use the AI chatbot tool to create integration scripts and quickly find answers without searching through hundreds of pages of technical documentation. In addition, because our solution is headless build on MACH architecture our customers could use ChatGPT as an interface layer on top of the PIM solution. This way, AI can help both analyze product data quality and take action to update the PIM when discrepancies are identified.
What are the risks of using AI to create your product data?
Depending on your industry, you need to identify which product data you can let AI run and where you need humans to verify data accuracy. For example, if you work with food, you know you need to get information about allergens correct, so I wouldn’t let AI touch that at all. The same goes for industries with lots of legal requirements on product data. In those cases, you shouldn’t rely solely on AI but make sure you have processes in place to ensure correct information for those business-critical data points.
On a high level, I think that many companies will establish their own AI strategy and a dedicated AI team within their organization. Not only to manage all the AI tools available but to develop their own AI that they can train, control, and use to represent the company. One of the core tasks for this function will most likely be to set the boundaries for where and when using AI is acceptable.
