As stated in your question, the project is focused on the “attrition rate of employees.” At first glance, this would appear to be a problem – not a process. Given the problematic perspective, we must recognize that “attrition rate” is a critical-to-value characteristic (CTV). As such, it should be known as the dependent variable (primary effect). Of course, the associated costs of attrition could be treated as a secondary effect. The point to this is simple – do not become sidetracked by chasing secondary effects.
In this context, we may now equate the model Y = f (X) to the given situation. In this case, we recognize Y as the attrition rate of employees and X as being representative of the independent causes (yet to be determined). Owing to this line of reasoning, you should focus your initial thinking on the development of a detailed problem statement. More specifically, such a statement should statistically describe the behavior of Y. So as to further the problem statement, consider the inclusion of benchmarking data (as related to Y). From such data, you will be able to better comprehend the gap between “what is” and “what ought to be.” It will also allow you to better correlate the primary effect (attrition rate) to the secondary effects (e.g., economic impact).
Until the magnitude of gap is made fully known (or statistically estimated) it will be very difficult (or even foolish) to advance a credible solution. Only when the performance gap has been meaningfully defined can the “vital few” X be rationally postulated and subsequently verified. Once known, each critical X must be realistically toleranced so as to establish the expected limits of performance. Pending these things being done, the control of Y will remain elusive.
To help guide your efforts though the maze of improvement, I would strongly recommend that you follow the process of breakthrough: Define-Measure-Analyze-Improve-Control (DMAIC). A simple interrogation of the Internet will reveal a lot of information on the DMAIC process.