How Can Artificial Intelligence Help a Retail or Distribution Company to Improve Their Online Success?

One sector of the retail or distribution industry that has seen an astronomical boom during the pandemic is eCommerce. Industries that support eCommerce have likewise enjoyed a tremendous surge, such as those creating and deploying intralogistics technologies for distribution centers (mobile robotics, unit sorters, pouch sorters, etc), last mile delivery companies, and even software companies such as those developing PIM (product information management) systems or MDM (master data management) systems. All these technologies have the same underlying goal in common: to help eCommerce companies scale and deliver orders on time and with accuracy.

Scaling can present many challenges, whether it is an eCommerce company (B2C or B2B) or the online sales of a brick-and-mortar retailer. One critical area that companies often discover (sometimes late in the game) they need urgent help with is their eCommerce catalog. Without well-structured data being pushed to the online store, they could be missing out on the opportunity of a lifetime. How So?

There are a few reasons why scaling an online catalog with thousands of products might be challenging, but at the same time it presents an opportunity to achieve better and more spectacular results than a corporation could traditionally expect from online sales.  Of course, I am not talking here about the multi-billion-dollar corporations that have already realized this long ago. They are already well prepared and taking advantage of the situation. I am talking about the next tier of online stores where the annual revenues range approximately from a hundred million to a billion+ USD.  For these companies, artificial intelligence can help them to properly scale and cement a strong position within the market they operate. This will allow them to explore new markets without complications.

Remember, If the customers visiting an online store struggle to find what they are looking for, perhaps never finding it at all (even when it could be there), they will leave and likely hesitate to come back again, finding perhaps a better experience with the competition. If they do find an online store that is user-friendly and resourceful, that business will gain and retain the loyalty of their customers.

THE LARGE ECOMMERCE CATALOG PROBLEM

When the online eCommerce catalog of any corporation surpasses about 50 thousand SKUs (and in some cases even less), it becomes particularly challenging to keep the costs down on the work of onboarding products to the catalog managed by the PIM or MDM system that will eventually make the product data push to the online store. Why is this the case?

Each retail or distribution company has its unique product classification and product taxonomy. The way the data is provided by vendors typically brings PDFs product data sheets with half-filled info, some titles, maybe some marketing copy, images, etc. That data is not ready for eCommerce. It does not match the way the retailer chooses to display products online whatsoever.

There are CSP (content service providers) providing this service. These are companies whose main task is to collect information about products, give them a sort of structure, and push them to the companies subscribed to their service. However, the product taxonomy of the corporation selling online is different from the CSPs own taxonomy. Also, the data obtained from the CSPs is usually incomplete. The subsequent work therefore must be done manually, which creates an entirely new set of problems. More people will have to be brought in-house just for this task. That will include reviewing and structuring the data, cleaning it up, reformatting it, and enriching the images. Also, it will encompass the creation of metatags for search engine optimization purposes and for their own site search and customer experience improvement. Maintaining the accuracy in this work and keeping the error avoidance consistently at a high level is a daunting task. Expect the accuracy to be 70% or less.

Another way of handling the formidable job of inputting data is to outsource the work to a BPO (business process outsourcing) company fully dedicated to this task and typically located in some of the lower-wage areas of the planet such as Vietnam, Philippines, India, etc. They will have to fill in the templates given to them. This is typically done with spreadsheets with many columns that need to be filled out. That brings another expense.

The first step is typically product classification. The product hierarchy at this point commonly has four, five, six or more levels, which makes it tricky. However, once it is classified, then the content needs to be enriched. The BPO needs to make sure the content is aligned with the corporation’s requirements and validated. A visit to the vendor or manufacturer’s website will probably also happen to ensure the information is correct, and to download missing warranties, product manuals, etc. Then this information is fed into the PIM system. From the PIM system it goes to the online store, printed catalog, smartphone app, etc.

I simplified the process described above a bit, but in generic terms this is what happens. The process typically takes from 4 to 6 weeks. Since most of the work was done by humans, there are a lot of errors induced. And if something happens along the way that changes the product taxonomy or the process, people will have to be trained yet again. Scaling the online catalog literally means hiring more people and increasing costs even more.  There is no scale benefit when operating this way. If anything, the opposite is true. Maintaining and updating the products will just add more headaches to an already painful operation.

THE AI SOLUTION

AI really shines in this scenario and is the perfect solution for any corporation with a large online catalog. AI models can be created to classify and categorize products automatically based on the information extracted from spreadsheets, scanned documents, PDFs, images, etc. Yes, it is not science fiction or black magic! Now with AI all the processes mentioned here can be automated.

Even with just a well-designed and professionally trained classification algorithm, a company can benefit a lot. So imagine what a full AI-driven catalog onboarding solution can do! Some catalogs can have thousands of leaf nodes. In other words, a product must be placed into one of thousands of different places. How easy would it be for a mistake to happen if this were done by imperfect human employees? How patient can a person be classifying a product this way? Well, in some companies they have another team just dedicated to correct where the products have been initially classified. Many of them end up in the “miscellaneous” bucket because people did not want to spend time finding the correct place.

A well thought out AI system that has already classified a product will then automatically proceed to define its attributes with more machine learning algorithms that will extract some attributes from documents and images, etc. A proficient AI system will also automatically generate titles, subtitles, product description, etc. Moreover, some good AI systems will also have algorithms that will also automatically update the taxonomy. This is especially useful on apparel related products where the way customers search for them is constantly changing. The manual process related to taxonomy updates is usually very convoluted and not clean. Now with AI this problem is a thing of the past. 

It makes sense also to update the metatags to match new keywords. The AI system will continue to make the customer experience better and better by analyzing the data from the Google analytics account connected to the online products and deciding what are the best keywords to adopt.

For example, if a trend develops in Online searches for “Ferrari red” t shirts, AI will automatically detect the trend by reading data from Google analytics. It responds by propagating the appropriate metatags and adding them to any of the shirts in the catalog that have a red color. A wide range of reds from dark to light could be included, since Ferrari has changed their red hue over time. But it doesn't stop there. AI would not just add “Ferrari red” metatags to your red t shirts, but also to all related products, such as all red shirts, caps, socks, ties, etc. You get the idea. Companies can remain in sync with current trends and ride each wave to the next one, staying ahead of the competition.

Additional benefits of the AI solution include its ability to reduce the product catalog onboarding process from 6 weeks to 1 week and even in some cases to 24 hours or less. It all depends on the products, the data, and the taxonomy structure. How about the accuracy? The accuracy with AI will be much better than with humans. As mentioned above, humans do this work with an accuracy rating of about 70% on average. With AI, 90% is typically the average on the lower end. If the training of the ML models properly takes place, an accuracy rating in the mid-nineties shouldn’t be a surprise.

THE TRAINING STAGE

With any AI model, data must be processed for it to be trained. In the case of auto classification, the input to the system might perhaps be a little description that the vendor provides, and the output to the PIM is the classification bucket. Humans might have already done this with many products. Those products could be good examples used to train the ML models. When the precision exceeds 90%, then it can be said it's trained. Otherwise, the process of information feeding continues until the desired accuracy rating is achieved. Once it is trained, the algorithm is ready to be used. The same is true with any other AI algorithm. Once in place, the AI system will deliver the information to the PIM system to be pushed to the online store.

THINGS TO CONSIDER

A good AI solution will not be a plug and play system. If anyone promises you something along those lines, run for your life! All AI systems need to be finely tuned to the needs of the corporation investing in it.  If it is done properly, then it will not happen overnight. Time is a factor to consider. Also consider that ML algorithms have an intrinsic performance degradation tendency. AI algorithms require proper monitoring and maintenance. A good AI vendor will discuss this with its customers.

It is imperative to consider how you will transition from a manual process to an AI system without disrupting business operations. The easiest way to do it is by using a gradual transition, but since there are different players or companies involved, this could be tricky.

Retail or distribution corporations will also want the transition to take place without having to pay twice for it, both to the team working on manual processes and to the AI company. If your AI vendor does not have a clear path or solution on how to accomplish this, then perhaps it is better to reconsider. Otherwise, it could turn out to be a messy AI experiment.

BENEFITS SUMMARY

With an AI system that automates the catalog onboarding process for any eCommerce effort,  the benefits could be summarized as follows:

• Faster process from 6 weeks to a few days or even hours.
• Improved accuracy from 70% to 90% +
• Smaller investment as the catalog grows
• Easier to scale
• Customer experience optimization
• Product matching recommendations
• Taxonomy and audit optimization
• More

If you want to know more about how an experienced AI company can help you with an automated catalog product onboarding solution for your online store, feel free to contact me.