How MIC is using AI to improve the next generation of trade management software

News 24 June 2024

Find out how MIC is developing new AI-enhanced solutions to improve the efficiency of critical customs management tasks.

Artificial intelligence (AI) is set to be one of the biggest trends of the coming years across many aspects of Software-as-a-Service. We recently looked at what advancements in Generative AI (GenAI) are likely to mean for trade management software, but we also wanted to share a few of the things that MIC is doing in this space to improve the offerings for customers and make life easier.

These efforts are being led by MIC's Data Science team, which has been a dedicated division of the company since 2020. Its Team Lead Clemens Kriechbaumer explained that the goal of this is to build a platform to "develop, train, and serve AI models, supporting various use cases in customs and global trade on our MIC cloud platform".

So what does this mean in practice? Here are a few of the steps MIC is currently undertaking to integrate AI into its operations and better service customers.

Using AI to meet customer demands

Clemens noted that there has already been significant interest from MIC's customers in AI, and in particular the opportunities offered by Large Language Model (LLM) based Generative AI (GenAI). For instance, one area in which the Data Science team has been working is implementing this technology into MIC’s customs tariff classification solution (MIC CCS).

Assigning the right customs tariff number for every product in each country a company operates in is critical for calculating duties and correctly applying free trade agreements. Even with the right software to assist, this can still be a highly complex process. However, with AI, these activities can be greatly streamlined.

Clemens said: "Recently, we trained AI models for one of the largest German eCommerce providers using 650,000 items and over 3.1 million product images, employing state-of-the-art techniques such as fine-tuning multi-modal foundation models."

Elsewhere, MIC is also developing chatbots that can enable interaction with customs tariff content. These are able to assist users with product tariff classification using natural language to improve productivity.

"It is also conceivable to have AI-powered co-pilots for customs declarations, supporting the processing by providing suggestions and instructions for various fields," Clemens continued.

What tools are MIC using to implement AI?

MIC's AI platform and infrastructure is based around Kubernetes, which includes dedicated hardware resources such as GPUs to support the training, development, and serving of AI and machine learning models. This gives MIC a long-term foundation to support and integrate AI into our products as the company looks to identify more potential use cases and enhance them with AI.

Clemens noted that it isn't only in customer-facing software that MIC is embracing AI. The technology is also being used behind the scenes to aid in software development. He said that tools such as AI co-pilots have helped significantly reduce the time needed to solve problems during development, greatly increasing the team's overall productivity.

A roadmap for the future of AI at MIC

In addition to the development of an assistant for customs tariff classification, which is similar to ChatGPT but focused on customs and global trade applications, MIC is exploring how existing open-source LLMs can be fine-tuned for use in customs-specific applications, such as handling legal documentation.

"Creating a domain-specific foundation model while maintaining its innate capabilities is an innovation we will implement sooner rather than later," Clemens explained. "Fine-tuning an LLM in a casual language modeling sense for a classification task might allow us to successfully train a model on mixed data from multiple customers."

The next step will also include educating users about what AI is capable of, and what it needs in order to be successful. At the moment, this knowledge is still limited, so Clemens noted it will be vital to improve on this to ensure the success of future AI initiatives.

"It's crucial to explain and educate our customers about AI for our business and the importance of having extensive historical data available to train the models," he stated.