DeepProducts AI-Based PIM Evolves Into A Future-Ready Data Generation Solution 

product data generation - new release

DeepProducts—Or How To Speed Up Product Data Creation

We’ve come a long way since we launched the first version of DeepProducts back in 2017. We designed it as an AI-based image annotation tool, that would speed up product data generation for online fashion retailers. The solution analyses product images using AI and Deep Learning techniques and outputs a set of relevant hierarchical categories and attributes. 

product data generation - new release

While brands and retailers have started realizing the inefficiency of spreadsheets, more requests have been coming in regarding the connectivity of our platform. They want to know how to plug it into their eCommerce platforms and ERP systems in order to generate product data and properly manage whole catalogs. 

During the past year, we’ve worked out the details to connect the solution to retailers’ channels and platforms and be able to fetch and push data by means of an API for more seamless and efficient user experience. 

But we’ve also recognized the need for the tool to become —and be named as sucha proper Product Information Management platform. Even though we had never named it that way before, it became clear that it was solving the same pain as a PIM system. But with even better functionalities! With that idea in mind, we decided to build a new, improved version. 

The Idea Of A Future-Ready User Interface

We have spotted a few flow and feature improvements on the journey. Therefore, the logical path towards an improved Deep Products version included establishing a series of requirements for the new version of our product, which would have a more intuitive, improved UI to cater to the needs of every apparel e-tailer.

It would not only help them generate their data using AI to speed up the process by 10 times but it would also help them better organize collections, filter items by attributes and categories, perform bulk edits…

Our team has worked hard to achieve this goal and we’re proud to announce the release of the latest version of DeepProducts. But how have we built it? Let’s tap into that.  

Interface Improvement Process

Slide to see the before and after of DeepProducts interface


1. Industry And Customer Feedback Regarding DeepProducts

We’ve been diving into the fashion industry for some years now. We’ve had many discussions and meetings with segment leaders regarding online retail processes and main pains and challenges when managing product information for small to large online catalogs.

Therefore, in order to better plan what the new DeepProducts version would look like, we have held interviews with our clients and with industry experts. The aim was to see what was working for them and which areas could be improved. These are the areas of improvement we tapped into: 

  • Working flows: Adapting features in order to adapt the software to typical retail workflows. 
  • UX: Modifying buttons, filters, etc. to eliminate tasks that consumed time, for instance, navigation or complex searches, as well as more intuitive use of icons for intuitive concepts such as gender or age.

2. PIM System – UX Research and Improvement

Keeping in mind those conversations, our UX and front-end team has conducted thorough research in order to bring the platform to the next level, following best industry practices and performing usability tests in order to provide the best possible navigation and user experience. 

The first prototype of the new version was tested and validated internally. Then came the first design proposal with basic capabilities. Our team iterated on it and performed user tests inside and outside the organization, with both people working on eCommerce merchandising and more general users trying it out.

Heatmap of the PIM prototype

Heatmap and users flow within the first prototype of the new DP version

Everything from design to UX is now data-driven, in order to optimize time spent annotating products and shorten time-to-market. And we have also organized navigation to resemble that of an eCommerce site.

The benefits of all this are:

    • Higher product lifecycle traceability: users can now see whether the status of an item within the product cycle (whether it has has been completely annotated, or pushed to the eCommerce platform, etc. 
    • Better annotation progress visibility: progress bars and filters make collaboration between annotators easier, as they show the annotation progress of a batch of items or the collection itself, while also showing whether an item is still pending or already approved.
    • Improved filters for product search: it is now possible to filter items not only by categories, status or item name but also by finer-grained attributes, like the type of sleeves or patterns. 
    • Improved efficiency: new flows and features allow for faster annotation processes.

The result is an intuitive, user-centered solution that speeds up product data generation.

3. On Point Design

In terms of design, our team wanted to bring to users a sleek, clean interface to simplify the way product information is created and updated. Therefore, we’ve used the principles of Material Design, a design language developed by Google in 2014 that uses grid-type layouts, transitions, and animations, and tries to mimic depth effects using lights and shadows. 

Nowadays, Material Design is present in almost all Google Applications. Why? The psychology behind this design language creates an illusion of “material” reality. For instance, shadows are used to convey hierarchy among elements. Whitespace, on its turn, draws user focus to a specific element.

Another area we’ve worked in is iconography. We’ve turned some basic attributes into easy-to-read icons. Therefore, when using the application, eCommerce content managers and merchandisers can see a sample of the garment color, which would look like this:

Alongside the color sample, we can also see whether the garment is intended for a man, a woman or both and whether the garment is fitted for adults, children, or babies.

Main Highlights and Features

After conducting several rounds of tests among existing customers and potential users of DeepProducts new version, we have been provided with some positive feedback. Besides our flagship feature of automatically generating product attributes, the main highlights that test users have regarding our final Product Information Management software included:

bulk and sequential edit

  • Bulk or common edit: Some of our retail customers were struggling with repetitive actions required to change a feature from different items to a single value. With our new bulk edit feature, they can, for instance, select the items they want to change the color for, and assign a new color to the whole batch of items. 

sequential edit

  • Sequential edit: Our previous version of DeepProducts already had an interface that would show the list of products to annotate and every time you would edit an item you had the option to save it or to save and go to the next item. However, what our users seem to like in this new version is that it allows you to see the progress of the annotation batch while working on it.  

Backend Improvement Process

Microservice structure

We couldn’t improve our solution without rethinking the whole infrastructure behind it. That is why, while building the new PIM version, our team decided to divide the architecture in what is known as microservices, a software development technique organizing an application as a group of “loosely coupled services”. These services are connected through API and allow for better scalability of each service.

For customers, this translates into the ability to react and adapt more quickly to changes in catalog volume.

Internally, this helps us maintain and introduce improvements in the system in a cleaner and faster way, for instance, adding functionalities in the form of new, connected microservices.

Latest programming language version and fastest servers

The thorough update we are releasing is also built with the latest versions of the programming languages used (Python), which will help maintain the code in the long term. Moreover, our Amazon Web Services servers allow for the fastest speed.

Curious to Try it Out?—Shoot Us A Message!

We can speak about the wonders of our brand-new version of DeepProducts, but why just take our word for it? Take a look at this short video demo and see how it works! We also invite you to shoot us a message so you can try it out with your products!