There are three fundamental things to get right when building an analytics solution.
Identify and document all data sources used to derive your insight. Understanding the structure of the data as it flows through your solution will allow for easier collaboration throughout deployment and and monitoring. Ensure that any data transformations required by your attributes or models during development will also be available in production.
Using the right tools and technology for the job should mean that the insight you develop should not end up being recoded into another format or language. If your insight is dependent on another team or department, then you raise the risk of being late-to-market and the value of your insight will be diminished.
Don't create separate workflows for batch and transactional jobs. Engineer your solution so that the same artefacts you want to deploy as a transactional service can also be executed by your batch platform.