In January 2025, seven things were slowing down my AI-based HS classification project: the cost, the lack of a published benchmark to learn from, the absence of classification know-how in public AI training data, investors’ reluctance to fund research, the gap between how developers and classifiers talk, money, and my own limits in communication and sales. This article explains each one as I experienced it.
Where the project stood
As I described in my earlier article on whether AI can do HTS classification, I have worked as an HS classifier since 2017. Over time, curiosity led me into research on using artificial intelligence for HS classification. After I published that article and shared it on LinkedIn, the response was far more positive than I expected.
Startup founders, CEOs and people from the industry wrote to me, and some were interested in the project. That was humbling and it kept me going. At the time of writing I was in discussions with a few founders about working together.
I had also found several problems that were slowing the work down. With respect to everyone involved, here they are.
Seven challenges in building an AI-based HS classification tool
1. The cost of the project
My estimate for making the project market-ready was around INR 50–75 crore. For most startups that is a very large first investment, and when funding has to be justified by quick returns, investors are hard to convince.
2. No published benchmark to learn from
I could find no published benchmark or proven workflow for this kind of tool, so I had nothing to follow. That made the prototype harder to build. It also left room for a new approach, but the uncertainty made investors hesitant.
3. Limits of current AI technology
AI systems are trained largely on open internet data, which is plentiful for general topics. The specialised theory and working methods of HS classification are mostly missing from public datasets. Training a model to classify with precision is therefore much harder than training it for general tasks.
4. Reluctance to invest in R&D
Startups tend to put the business first. For many founders, research and development feels risky because it does not promise quick returns, and that reluctance to commit resources was a real obstacle for this project.
5. Communication gaps between developers and trade compliance specialists
Developers and trade compliance specialists come from very different professional worlds and often misunderstand what the other needs. I ran into this when explaining my ideas to a developer with more than 16 years of experience. He knew his field well, but we used different professional languages, and the gap was hard to close.
6. Financial constraints
A project of this size needs money at every step, from prototyping to testing, and my limited funds slowed progress. I tried to raise money myself, but crowdfunding platforms such as Kickstarter are not available to creators in India, which made it harder.
7. Communication and marketing skills
I studied in Bengali-medium institutions and English is not my first language, so explaining my ideas to CEOs and investors is not easy for me. I also stutter in conversation, which adds to the difficulty. And with little experience in sales and marketing, I find it hard to persuade people of the project’s potential.
What I am doing next
I am not giving up. I keep working on the project with the resources I have, and I am working on my communication and marketing skills so that I can explain the idea better to investors and collaborators. I believe the hard times will pass. (Update: the tool was launched in July 2025 as Tariff Wolf, an AI-powered HS classification tool.)
Support the future of AI-driven trade compliance
Demo of the AI-powered classification tool
Thank you
I am grateful to everyone who has supported me, read my articles and encouraged this work. Your encouragement keeps me going when things get hard. Thank you for reading. Jai Hind!