A well-set-up and managed product information management (PIM) system is crucial to the success of your eCommerce business. You won’t get far without a centralized structure that stores the product descriptions, photos, categories, attributes, prices, and inventory data and then distributes them to the other systems that make your store tick.
But there’s a problem with PIM systems: it takes a lot of time to supply the system with all this information. It’s not a very engaging job, and because of that, it’s highly prone to mistakes as you repeat the same tasks over and over again. Outside of writing the main product descriptions, there isn’t much room for creativity. Fill out the attributes, set the price, select the categories. Then, try to repeat that for every product in your inventory.
Luckily, it doesn’t have to be like this. The recent boom of artificial intelligence (AI) and machine learning (ML) solutions has revolutionized what’s possible in terms of product information automation. According to our eCommerce Trend Radar, it’s one of the quickest rising trends in eCommerce at the moment. Introducing AI to your PIM system can save you and your eCommerce staff hundreds of hours of dull work and, at the same time, have a positive impact on your business’s bottom line. Here are seven areas you can successfully deploy AI and ML solutions.
Merging data from different sources
Say you’re starting a dropshipping eCommerce business. You’ve set up the back end along with some content and a beautiful front end. Now all you need to get going is the inventory.
Traditionally, this is where problems would start. Each of your suppliers will have different methods of delivering product information. Some will give you CSVs, others XLSXs, some JSONs. They’ll all have different conventions of organizing data, like column and row names. Some won’t have a database format at all and will give you catalog PDFs or website links instead. In short, there would be a lot of tiresome manual work ahead of you to import this data into your PIM.
The introduction of artificial intelligence can make the data cleansing and editing process much smoother. You’re no longer bound by the precise format that your PIM supports to import the data correctly. Whatever data structure or file format you throw at the AI-powered system, it will figure it out and import the data automatically, even from non-standard formats like photos and PDFs. If the system is not 100% perfect on the first try, ML comes into play. You’ll need to show the system the correct way of interpreting problematic data only once, and it will apply this rule to all future imports. It’s like Pimcore’s already impressive Data Importer extension but dialed up to 11.
Content writing and editing
In the previous point, we looked at the scenario of processing product data received from suppliers. But what if you’re setting up a store for your own products?
Based on our experience, it takes about 30 minutes on average to manually fill out the description and attributes of a single product. Multiply that by the number of products you’re going to keep in your PIM, and you’ll probably get a headache-inducing amount of time.
AI is a complete game changer in this area because it allows you to reduce the time you spend creating those descriptions to just a couple of seconds per product. You can feed it information you already have about the product, such as technical specifications, talking points for your salespeople, or just random unstructured notes. That’s really all it needs to automatically create full-blown descriptions that will neatly fit into your store’s structure. If you’re thinking of going global, AI can also help you with translating the generated descriptions into other languages with a single press of a button. This was exactly the approach taken by Mondial Tissus and Akeneo, and it significantly sped up the store’s time to market in terms of localization. Of course, you can’t blindly trust content generated by AI and you will need to verify it. But this still means that what once was a job for multiple dedicated writers now requires just a single person.
Attribute mapping and classification
One of eCommerce’s greatest tragedies is a store that carries the products that clients are looking for but makes them hard to find due to chaotic attributes and categories. So close to conversion, yet so far. Luckily, this is another area where AI-powered PIM systems can help you increase your profits.
Sorting products into categories
One way is to start with your hand-created attributes and categories. Then, you can employ AI, like Pumice.ai, to analyze and map the products you carry and automatically sort them based on their title, description, or images. This approach might be helpful if you’re sure that the categories and attributes themselves are in order and that the problem lies in product assignment.
In this case, AI can help you with tasks that require a lot of dull work but also those that need specialized knowledge and understanding of the product. If, for example, you’re running a store with aftermarket car parts, you’ll want to assign products to specific car models that they fit. In a traditional PIM system, you’ll need either a specialist who knows all of this by heart or hours upon hours of researching official product specifications and other information sources. Of course, AI won’t replace the knowledge and experience of an expert. It can, however, handle looking that information up in the documents you’ll feed it and save you a large amount of time.
Creating categories from scratch
On the other hand, a reverse situation might also be the case: the problem could be that the categories and attributes themselves are at fault. They might be unintuitive, cryptically named, mapped in a confusing hierarchy, and so on. In this scenario, it might be a better idea to start from scratch. You can let AI propose a new, unified category and attribute structure for your inventory, like with the solution offered by Width.ai. The categories can, for example, be based on product taxonomy standardized by Google to ensure maximum SEO-friendliness.
SKU matching
A fantastic use case for AI and ML in PIM systems is finding and automatically matching SKUs of duplicate products. For instance, Unite.ai offers an AI-based natural language processing (NLP) solution that only requires product titles to perform accurate matching. This is especially useful in marketplace eCommerce platforms, where many different sellers can offer the very same product.
Consolidating SKUs will benefit both you and your customers. On your end, it means a better organized and leaner database that will run quicker and be easier to maintain. This will also improve your ability to moderate the marketplace. Imagine 10 sellers offering the same product but one of them at only 30% of the usual price. Unless they’re running a crazy sale, there’s a good chance something suspicious is going on, and you might want to look into it. An AI-powered PIM system can find those outliers in real time and suggest them for human review.
Matching SKUs with AI also means you can implement several quality of life improvements for your customers. For instance, you can give them the ability to compare all the different listings of the same product and suggest the one that best matches the customers’ priorities. Those can be the best price, quickest shipping, or most generous returns policy.
Dynamic pricing
Product pricing is yet another area where the introduction of AI in product information can revolutionize the way you approach your business. Solutions like PriceIntelligence AI already let you monitor your direct competition and adjust your prices accordingly. Currently, this can happen as often as every 10 seconds. But there’s still more potential here.
You could, for example, task AI with monitoring and analyzing customer behavior and real-time global trends and use this data to suggest optimal pricing strategies. Is there a shortage of a given product on the market? Might be a good idea to bump your price on this item. Weather forecasts suggesting a particularly cold and wet summer? Better get a sale going on your outdoor sports-related categories. These are the kinds of analyses and decisions you can now successfully delegate to AI.
Demand forecasting
Demand forecasting, on the other hand, is about monitoring your store’s internal data as much as the outside world. Machine learning systems, like Garvis AI, can observe that, for example, you tend to run out of stock of bicycles year after year around June. Naturally, for the next year, you should be trying to get a bigger stock to maximize your profits.
But how much bigger? You don’t want to overshoot it either, after all. AI can take into account aspects like:
- Your own data to see if sales were already cooling down or just picking up steam when you ran out of stock.
- Website analytics to gauge how many users visited the product pages only to find the item already gone.
- Global reports on the bike market.
- Seasonal weather forecasts.
- Data on your clients’ overall financial situation.
Based on those criteria, the system can come up with an educated prediction on how many more bikes you should get for the next year. Then, it can automatically adjust your orders from the suppliers.
Image recognition and tagging
In the “Attribute mapping and classification” section, I briefly touched on how AI can analyze images to classify products. That remains true, but there’s much more potential for image-related tasks, where introducing AI in PIM systems can save you countless hours of work.
Photo-based tagging is the perfect area for AI to shine and has been successfully implemented in Pimcore with the Automated Image-Tagging bundle. Imagine you’re running an eCommerce business that sells decorative pillowcases. Based on just the pictures, AI will be able to determine most of the crucial tags and attributes that make your store easy to search. It can classify the color, shape, pattern, like floral, geometric, etc., and maybe even the fabric with some training. All you need to do is feed the system the appropriate product photos. AI-powered image tagging is already a fairly large industry. This is great news for you because you’ll be able to find solutions specially crafted for certain uses or product types. Pixyle.ai, for example, is dedicated to tagging fashion products.
Another field where AI can save you from mind-numbing work is filling out image metadata. Nobody enjoys entering image titles, file names, and alt texts for every photo, yet they’re necessary to make your store accessible and SEO-friendly. Luckily, this is an ideal job for an automated system. It’s a straightforward process where not much can go wrong because it boils down to “describe what you see.” A piece of cake for the AI, and a huge weight off your shoulders.
Bottom line: investing in AI pays off
The introduction of AI and ML solutions is the next logical step of PIM systems’ progression. There’s so much to gain in terms of efficiency and profitability, the decision seems to be a no-brainer for most eCommerce business owners. Automating the tasks described above can save you countless hours spent on dull tasks and, at the same time, increase your profits thanks to systems like dynamic pricing and demand forecasting.
Artificial intelligence is also not a monolith, and there’s no need to go all in. If some of the described systems sound exciting to you but you’re skeptical about others, you’re free to mix and match them in any configuration that works for your business. Since the popular PIM systems don’t offer AI functionality out of the box yet, setting up such features will be a matter of connecting the AI to the PIM system’s API. Pimcore’s robust integration capabilities are a good example of what you should be looking for to ensure the integration is successful.
Of course, AI will be more effective the more you use it. When you first employ automated systems to the tasks described above, you can’t expect completely perfect results. Each store is unique in one way or another, and it takes some time to teach the technology the logic behind your business. However, when you spend some time with the systems and feed the algorithms enough data, you can expect AI to be more and more precise with every new item you’ll ask it to process.