For AI applications built for businesses that would be enhanced by industry vertical specific knowledge, which would be the better approach - finetuning the model on ground data or using retrieval augmented generation?

Would like to hear the community's thoughts re: benefits and concerns.