I applaud your interest in AI. The strong suit (pun intended) available to you at the moment would probably be to use ML/AI in your logistics chain. If you are already in the position to acquire inexpensive materials you have a step up, but what data and forecasting about the buyer market? I would first concentrate on getting a structured pattern learning system up and running. Look for insights into structured data by converting it to triples and graphing it, looking for anomolies and trends along the edges, especially newly discovered ones.
Once you get the hang of this, research into applying ontological reasoning to the data (as the Internet is still yet a semantic search world). Once this is achieved, look into OPPL (ontological pre-processing language) and SWRL. Then you can ramp this data up byapplying Logic programming and reasoning to it. I would recommend using OPPL to cycle SPARQL queries into variabled Epics for your industry domain. It will help keep digital overhead down, but beware, it can also cause overfitting by the reasoner and rabbithole you in the wrong direction and waste your time and resources if not supervised.
If no overfitting is obvious, then move it through a neuralnetwork and joila... you find out that selling jackets for pet pigmy pigs in Nebraska is the next big fashion trend....
Most of this is already being done with structured data in MySQL with ML tools like Jena connected. I like shiny things too, but AI might be overkill, especially in the Pre-Training data stage. Simple RDF graphing analysis might get you where you need to be.
