Product information management (PIM) is one of the most important data management tasks for ecommerce and online retailers. Let’s look at what product information management is and how we can use artificial intelligence (AI) to optimize PIM tasks.
Product information management is the process of structuring and managing complex product information such as product content, product taxonomy, and other product related assets. PIM systems focus on creating a single touchpoint for all product data processes and easily synchronizing selling across multiple platforms or marketplaces. The goal is to improve selling performance and give ecommerce companies more structure for managing the products they sell.
The best practices in product information management come down to what best fits your organization regarding everything from business product structure to employee commitment. Understanding what goes into successful product information management and being able to quantify that into resource deliverables is the key to PIM being able to show ROI.
Breaking down the PIM task hierarchy into specific roles is a great way to understand everyone's job and what deliverables come out of each one. This lets you ensure you have coverage of the most common PIM tasks such as publishing products, updating inventory numbers, syncing products across channels, and updating product taxonomies.
Many PIM systems let you create various roles with different access levels to various operations. This lets you control risk and understand who has access to what operations.
Making changes and optimizations to product information is a process that is nearly impossible to avoid. Instead of trying to reduce the number of times you have to perform the most popular PIM tasks, why not optimize the process of performing those tasks? The goal of PIM is to make those processes easier to perform. Popular PIM tasks such as creating new products and updating product data are not going away but can be optimized with better SOPs.
Creating a centralized system for product information allows you to keep a birds-eye view of your entire product catalog. This allows you to have better insight into the quality of your product data across one or multiple channels and implement changes more efficiently.
Centralizing your product information makes it easier to manage data quality as well. The standardization that comes with fitting product data to set fields leads to a better viewpoint of the product catalog data as a whole which makes it easier to find holes in the product information. A research report from Forrester found that poorly architected sites sell 50% less than organized sites with solid product information structures.
Modern product information tools focus on improving the product information structure for ecommerce brands and retailers. The goal outcomes are generally described as increasing clarity and streamlining product information tasks for employees, which allow businesses to optimize required product management tasks.
But a critical downfall in modern product information management is the reliance on manual employee effort to actually manage the product information. These tasks although optimized through PIM workflows still require humans to:
1. Perform the tasks
2. Monitor product information repositories
3. Build support style teams to work through constant updates
4. Have a deep understanding of product taxonomy
Let’s look at a few key points that make product information management a challenge for growing ecommerce brands even with modern PIM software.
Not only do these tasks have to be performed manually, but the number of resources required to complete them increases as the size of the company grows. Categorizing 1,000 products into 20 categories is a much different workload than 50,000 products into 5,595 categories (like the Google Product Taxonomy!). More products, more categories, and the velocity of product creation lead to a sharp increase in required product information work.
Even small changes to product information require a decent amount of work to complete. Adding a new category “Cat Car Items” to the “Cat Supplies” category would require you to go through all surrounding categories and determine which products belong in the newly created category.
Most PIM software today makes the management of these tasks easier and more efficient but doesn’t eliminate the need for manual effort.
Product information tasks such as categorization or taxonomy structuring require industry specific skills and knowledge from the people working on them. Product taxonomists that work at a cat supplies ecommerce company need to understand the product enough to decide if a new product fits in “non-prescription” or “prescription” in the above taxonomy. This means experts are required which can be more expensive.
Optimizing the shopping experience for customers is a key goal of product information & catalog management. Customers want to be able to reach the product they’re thinking about in as few searches as possible, and 47% of users give up on their search after one attempt. Structuring your product data and categories in a format that best aligns with your customer's intent helps to reduce bounce rate and keep potential customers on your site and find the exact product they’re looking for.
Customer intent optimization is a difficult task that requires knowledge of both in-house metrics and general ecommerce sales optimization. Your employees that work in product information management need to understand how their backend work improves conversion rates. The required skill for PIM is more than just monitoring product data.
The key takeaway from the difficulties and problems with modern PIM tools is the reliance on a serious manual workload and expertise to follow the best practices. PIM tools streamline many old school processes but still rely on manual teams to perform these new tasks, and many grow in workload over time.
Artificial intelligence (AI) and natural language processing (NLP) have pushed many industries forward with data focused automation. Data focused automation allows many industries such as ecommerce, legal, accounting, and more to remove huge amounts of manual workload required to perform data tasks. These tasks can also be automated at the same or higher levels of accuracy than human effort.
Ai is doing the same thing to PIM and product data management. Models learn relationships between various product data points to allow these models to automatically perform various tasks. These relationships can be between product data from the same field or relative to other fields depending on the task. Once data relationships are learned we can start leveraging them to make comparisons, predict outcomes, and generate new data such as product descriptions. Stacking these simple building blocks on top of each leads to powerful Ai pipelines that can perform tasks.
Let’s take a look at a few quick examples of how Ai can be used to enhance PIM solutions.
Automated product categorization allows you to automatically fit product data such as title, description, and image to your internal product taxonomy. These ai models learn the relationships between this product data and taxonomy trees based on previous and current catalog relationships. You can use product taxonomies as small as 10 categories or as large as Google Product Taxonomy (5800+ categories). These automated product categorization models categorize products 17x faster than manual effort can and becomes even more efficient as the amount of data increases.
These Ai pipelines allow you to generate metadata such as categories and product tags from product data. This is a great way to add to your product information and improve your search engine's ability to find best fit products and lower bounce rates.
You can also take raw unstructured product data and extract metadata such as care instructions, size, color, descriptions, and tags. These two Ai models allow you to automate huge parts of the onboarding process for new products.
Product matching allows you to compare product data such as titles, descriptions, and images for similarity. These models have learned a deep relationship between products based on the goal underlying metric. This allows you to automate tasks such as:
- Multi-seller SKU standardization: Online marketplaces that sell the same product from different sellers need to standard the incoming products to an internal SKU. These products come in from different sellers with high variation in the information provided in titles and descriptions. Product matching Ai models allow you to compare these incoming products to the existing products and match them to the correct SKU.
- Real-time product categorization: Compare existing products from the same category to new products to understand the best fit based on your current catalog.
- Product listing optimization based on competitor analysis: Automate your competitor analysis by comparing products for similarity. You can take this further with deep learning models that extract key information from listings and find gaps in yours.
- Automated product taxonomy tree optimization: By understanding the similarity between a set of products in a category you can evaluate outliers to create deeper categories. Given a set of products in the category “Accessories”, if 10% of them are deeply similar to each other and are all related to “sunglasses” you can create a new category “Accessories > Sunglasses”.
We’ve built a SaaS product Pumice.ai that offers a number of Ai based endpoints that allow you to automate all these PIM tasks in a matter of seconds. Let’s take a look at how you can get started automating PIM tasks with Ai.
Understanding & gathering product data useful for PIM tasks is the first step in using ai to automate. Most PIM tasks require data such as:
- Product data fields. Fields such as attributes, features, price, model etc can either be added as optional fields or appended to the product description. Product data quality is important when using machine learning or ai. Having accurate product data will improve the results you get.
- Taxonomy tree or categories list. This can be as simple as just the names of the categories and the tree layout. You should include all levels to the tree as the ai models use all information to perform taxonomy tasks.
- Existing product to category matches. These relationships can be used for fine-tuning the baseline models for better performance.
This product information data can be exported from your PIM system or internal product database. Many ecommerce platforms such as Woocommerce and Shopify support quick data exports.
There are two ways to interact with the models and endpoints in Pumice.ai - file upload in the dashboard or direct connection to the APIs. When using the dashboard you’ll be required to format your data into the acceptable format based on the endpoints you’re using. Enterprise customers with custom integrations to external databases or ecommerce platforms can leverage their own upload workflows.
The API connection allows you to skip the dashboard and interact with the Pumice.ai endpoints directly. You can easily build endpoints into PIM workflows adding automation to the tasks without needing manual effort to start ai tasks. The raw API connection also makes stacking multiple ai tasks on top of each other much easier to create fully automated PIM workflows.
Pumice.ai offers a number of built in ai endpoints that allow you to automate different PIM tasks. These ai endpoints offer baseline models built for general use and can be fine-tuned to improve performance for your specific use case. Fine-tuning allows these models to learn a deeper relationship based on your data which “steers” the understanding towards your data.
All of the PIM tasks outlined above as well as many more can be automated with Pumice.ai endpoints. Let’s take a look at a few offered in the baseline version.
This pipeline focuses on categorizing product data into a taxonomy tree of categories. The dynamic nature allows you to fit products to a taxonomy tree without the models being trained on the specific categories. This feature lets you start categorizing products quickly without needing to train a model specific to your categories. You can use taxonomy trees as large as Google Product Taxonomy.
Natural language processing pipeline focuses on understanding the relationships between products.
Ai endpoint focused on extracting metadata from unstructured product data. Fits metadata into the Shopify metadata format.
Generate new metadata such as categories and tags. Can be used to generate “higher order” metadata such as “Season: summer” for swimwear.
Once you’ve connected your product information to the ai endpoints you can officially say you’ve started in automated information management! The level of manual effort after this is up to you. You can create fully automated pipelines (seen above) with custom integrations. The most common integrations we’ve built for customers are:
- Sales Layer PIM
- Your favorite PIM tool
- Google Sheets
- Other digital asset management tools
- Other Ai models (At the end of the day, we are still a development company!)
Pumice.ai is a PIM enhancement platform that leverages our proven models to automate product information tasks and easily setup digital commerce workflow automation. With the help of our Width.ai development services, you can leverage custom models and integrations built directly into the software. Set up a demo today to learn more about how you can automate your product management and get hundreds of hours back in a few minutes.