More accuracy is better, but it may not be a good idea to keep working on a model if you are expecting negligible improvement or cost of accuracy exceeds financial gain. The sole purpose of a data science job is to create financial value and minimize loss by building more accurate models. The guiding regulatory rules say say that if your model is having a negative impact on a customer then it must explain why an individual was so rated. This is a classic tradeoff between accuracy and interpretability. In a regulated industry if someone suffers from your decision and you can’t explain why the prediction model worked that way, your technique is not allowed. A good story telling using data visualization might help you to convince management. Some techniques like Penalized Regression, Generalized Additive Models, Quantile Regression can provide better accuracy and maintaining interpretability. Deep Neural Networks have also proven a successful approach to solve this problem.
You can read in more detail at : http://www.datasciencecentral.com/profiles/blogs/deep-learning-lets-regulated-industries-refocus-on-accuracy