Generative AI and PIM: 5 Measures to Improve Your Marketing Skills

Product Information Management (PIM) systems help you manage product data centrally. As a result, they are always up-to-date and complete, so you can play them out on your channels in real time.

With the help of generative AI models, you can now use this data even more efficiently. In this article, you will learn from our guest author Arndt Kühne about five exciting use cases of how you can scale your marketing faster with Generative AI and product data.

What is PIM and why is it so important for companies?

Product Information Management (PIM) systems allow your company to manage all product information in one central location. The PIM serves as a “single source of truth” for all product information, providing the data to all relevant channels through interfaces, such as your online shop or your website.

A PIM makes it easier to keep product data up-to-date and complete across all teams and touchpoints. This reduces errors, speeds up time-to-market, and thus provides a better product experience.

5 Areas of Application of AI in Product Information Management (PIM)

Data is the foundation of every AI model. Data is needed to train AI models and provide them with context. Just as your PIM delivers product data to your online shop, it can also send data to AI models, like to generate or enrich automated prompts.

1. Generating (product) texts

When it comes to Generating texts, two approaches can be distinguished.

• Automating with Large Generative AI Models
• Automating with rule-based AIs

Both approaches have advantages and disadvantages, making them more or less suitable for certain applications. But you don't have to decide on an either-or choice. Many rule-based tools integrate Large Generative AI models, such as ChatGPT, and can therefore combine the best of both worlds. Alternatively, you can, of course, also integrate both and choose the appropriate tool depending on the use case.

Large Generative AI Models (LGAIM)

Large Generative AI Models, like ChatGPT, have long since become a part of our everyday lives. Until the launch of Threads (Meta), OpenAI's tool was the fastest app with 100 million users.

Large Generative AI Models are therefore very popular for creating texts. With the right prompts, texts can be created in just a few minutes that often hardly differ from average manually created content.

However, there are two common problems with models in text creation: hallucinations and “bias”.

Under hallucinations we understand the property of generative AIs to invent false information. For example, if you let ChatGPT write product texts, even with data directly from a PIM, it is not uncommon for the AI to invent additional information that it considers relevant. This leads to texts suddenly containing product features that do not exist or that you might not want to highlight.

Automated text generation in with , Source: Pimcore, applied here by Basilicom

Large Generative AI Models are therefore excellent for processing creative content like blog texts or integrating in chatbots and/ or Chatbots Tools.

However, for content with less creative leeway, such as product texts, they are less suitable. Here, rule-based tools come into play.

Rule-based Automation

The advantage of rule-based text generation is that you can create templates for certain texts. In these, you can define rules, as the name suggests. For example, you can define placeholders for certain product information such as weight, which are then automatically filled with information from your PIM.

You can also specify whether your texts should contain certain words or explicitly not contain them, for example, to comply with your brand guidelines.

Unlike ChatGPT & Co., the templates and rules must first be defined, but the texts do not need to be edited afterwards.

Rule-based content AIs are therefore ideal for generating content automatically and in consistent quality, and producing it in large quantities. This is fast and scalable. In combination with product information, this makes sense especially for product texts and other product-related content such as FAQs.

2. Generating product images

Getting good images is a big challenge in marketing (23.7% of all content marketers see visual content as their biggest challenge). Generative AI is changing that. Gartner predicts, that by 2027, 70% of all design tasks will be automated with AI.

You have the choice between three areas of application. You can use AI to generate completely new images from prompts, you can train models with your own images, or you can use existing images and have them altered with the help of AI.

Likewise, Generative AI models like , or , are well suited for creating visuals for creative content, but are probably not capable of actually producing useful product photos from the data in your PIM.

If you have a large number of product images available, you can also train a model to create product photos for you. However, most companies do not have the required number of photos to achieve photorealistic results. Current models should be trained with datasets of at least 200 (Stable Diffusion) to at least 500 (Runway) photos.

However, the images in a PIM or DAM are very well suited for modifying existing images. You can, for example, change the background to show the same product in different environments.

This has several advantages:

• You save costs because you only have to photograph the product once.
• You can increase the number of product images quickly (for 75% of all buyers, product images are decisive when making a purchase).
• You can personalize product images by, for example, showing the same product for different segments in different environments.

For example, we have integrated Pebblelly and Stable Diffusion into Pimcore to generate authentic product photos for a shop from a single image. Adobe Firefly now also offers a similar tool.

Automated image generation and editing in Pimcore (Bundle from Dall-E (OpenAi), DreamStudio (StabilityAI) and a local set-up via Automatic1111 and StableDiffusion-API), Source: Pimcore, applied here by Basilicom

Automated product image creation in Pimcore with Pebblely, Source: Pimcore, applied here by Basilicom

3. Generating ALT Tags

The opposite of text-to-image is image-to-text, i.e., image recognition. Machine learning for image recognition is not new. Facebook (now Meta) already announced in 2010 that it would use face recognition for tagging people..

However, while Meta has already discontinued the function, AI for image recognition in connection with your PIM or DAM is more useful than ever. Two practical examples that can save you a lot of manual work.

• You can use image recognition to suggest tags or automatically tag images.
• You can use the AI's image description as an alt tag.

You should particularly keep an eye on alt tags at present, as the European Accessibility Act will enter into force in 2025. Then (almost) all e-commerce sites would have to be accessible without barriers. However, currently 55% of all images on the Internet are not provided with an alt tag. 

Image recognition in Pimcore with Astica, Source: Pimcore, applied here by Basilicom

Automated image descriptions in Pimcore with Astica

4. Hyper-Personalization

Personalization in marketing is not a new idea, OMR Reviews lists more than 40 different tools for website personalization (July 2023). In addition, there are tools like CRM systems, DXP or CEP platforms, which aim to create personalized customer experiences.

Personalization is then usually based on rules and segments.

The next stage of evolution is hyper-personalization. AI enables you to create content individually for each user, play it out in real time and thus personalize the entire user journey.

The product data from your PIM can make a valuable contribution to this. Together with the user data from your CDP or CRM, you can use them to adapt texts and images exactly to the wishes of your customers.

However, this requires a fully integrated MarTech stack with tools that offer sufficient personalization options. For example, many CRM tools currently do not offer the option to include placeholders for personalized images.

5. Better results in internal search with AI

43% of all users initially use the search function of an online shop. However, 42% of all websites offer a poor search experience, for example because they do not recognize synonyms for their product categories. Many companies still rely on classic full-text search in their tools and on their websites. This works well if your users already know what they are looking for, but it produces poor results if they are looking for concepts or context.

Results on imdb.com for the search “detective london”, Source: imdb.com

Vector-based search engines can be a solution. A vector represents the distance to a feature. A simple example of vectors are coordinates. If you search for Berlin, Hamburg is closer than London. You can apply the same principle to any feature. For example, OMR is closer to the Marketing feature than CEBIT.

Large Language Models use the same principle to understand language. If you get ChatGPT to analyze a text, for example a product description, the AI automatically generates vectors. You can also store these vectors in a vector database via the (Embeddings-) API and make them searchable. This works with data in your PIM as well as with texts on your website. As long as you have content that an AI can analyze, you can build your own AI search.

Advantages of vector-based search engines:

• Better search results
• Content and products can also be searched in context
• Content and products can be automatically searched in any language with which the LLM has been trained (even if the content is not available in that language)

Semantic search over a vector database in Pimcore, Source: Pimcore, applied here by Basilicom

This also works in principle with multimedia content like videos. However, the current generations of the necessary multimodal AIs are not yet developed enough to deliver good results.

What are the benefits of integrating AI into your PIM system?

Integrating an AI tool into your PIM can help reinforce many of the known benefits of a PIM and simultaneously opens up entirely new possibilities that were previously technically unachievable.

AI tools can help speed up or completely automate manual tasks such as tagging content, entering alt tags, or writing and translating product texts. This saves you work and further shortens the time-to-market for new products or markets.

Some measures, such as hyper-personalization, especially in omnichannel campaigns, are only scalable through the use of AI.

In short: Almost every process and measure involving the processing of data from your PIM can be automated or at least made more efficient using AI.

What should you consider when integrating AI into your PIM system?

Before using AI strategically, you should first check the availability of your data. This applies to applications with your PIM as well as all other systems. For your PIM, this means that your data model should reflect all relevant product data and you need high quality data.

If a large proportion of the products in your PIM are missing information, you will have trouble producing usable content for them.

In addition, your PIM should already be connected to all relevant output channels. It makes no sense to generate product texts for thousands of products if you then have to manually transfer them to your CMS or shop system.

Last but not least, you should think about whether and what information you want to share from your PIM with an AI provider. Your AI and data strategy should therefore also contain clear compliance guidelines.

Which tools are suitable for integrating AI into your PIM?

The most important tool initially is not the AI model you want to use, but your PIM. Your PIM should be open to interfaces, because then you have free choice of which AI models you want to integrate, as long as they also have a corresponding API. Open Source PIM systems have the added benefit of allowing you to use applications from the community.

Depending on which other tools and data you want to use, it is also advisable to use AI models that you can install locally. This is, for example, the case with the models from Stability AI, such as ,. This makes it easier to use sensitive product information or customer data because they never leave your own system.

In the end, of course, it depends on what tasks the AI should take over for you and where you can achieve the greatest efficiency gain.

Conclusion: Can Generative AI revolutionize your PIM system?

Generative AI can make many processes in your PIM much more efficient and significantly reduce the time expenditure. However, the greatest value of AI lies in how you can use the data in your PIM to improve your marketing and sales activities in other channels.

So AI will not revolutionize your PIM system itself, but together with your PIM, AI can revolutionize your marketing and your conversion rates.