In these phases, the Platform is analyzing your Excel model, and extracting a linear, quadratic and/or mixed-integer programming problem from it. For example, the LP coefficients often are not explicit in the spreadsheet, they have to be computed as gradients of the objective and the constraints, through a process called algorithmic differentiation. If you were using a custom modeling language rather than Excel, the modeling language processor would have to do similar work, and this can easily take longer than the actual solution process.
It is possible to write your Excel model in a way that minimizes the setup time, but the tradeoff is that you have to be careful and do more work to explicitly put all the coefficients in cells, and use only certain functions such as SUM and SUMPRODUCT. This is what we call "Fast Problem Setup". This is described in the Frontline Solvers User Guide, in the last chapter "Building Large-Scale Models" and section "Organizing Your Model for Fast Solution".
Finally, you could bypass this setup time if you were to rewrite your model in a programming language (C++, C#, Java, etc.) and use our Solver SDK Platform product. However you have to do more work to write in a programming language, and you give up much of the flexibility and interactivity of Excel.