This may be field-dependent; I am answering from the point of view of an HCI-related CS subfield that is frequently in touch with other fields (for different use cases etc.).
Always choose your venue (concrete journal or conference) first before writing the manuscript.
The underlying core information you want to convey will remain the same no matter what. Hence, you may prepare a very abstract outline first. However, writing the paper before choosing a venue does not make much sense:
- Important if you are in a field where manuscripts are formatted with the final styleguide: Each venue will require a particular formatting styleguide. While you should largely keep contents and format separate in your document, there are still more than enough detail decisions that require changes in your document according to the styleguide. For instance, if you use LaTeX, depending on the venue-specific document template used, usually only one out of three or so alternative existing packages for a specific feature (e.g. subfigures, special tables, ...) will be compatible with the template.
- Different venues have different audiences: Even when presenting the very same procedure and findings, you will have to add a different set of explanations, references, etc., depending on the primary audience of the target venue. Likewise, depending on the audience, sometimes you are better off starting with formal definitions and then presenting a concrete use case, whereas other communities prefer starting with a concrete problem and a case-specific solution description, and then formalizing that to a generic model.