Positive and negative utilities result because we use effects coding for each attribute, which centers utilities at zero. So you will ALWAYS get some positive and some negative utilities.
The best way to measure impact is by running sensitivity analysis via simulations. For example, for one of the concepts in your simulation hold everything constant in three simulation runs. In one run let the salary be low, in one medium and in one high. The simulation will show employer choice probability by salary level. So if your shares for low/medium/high salary are 20%, 25% and 40%, you might conclude that high salary has more positive impact than low salary has negative impact. Or if the salaries were instead 2%, 25% and 30%, you might conclude that low salaries have more negative impact than high salaries have low impact. Showing the sensitivity analysis on a graph would make these relations even clearer.
I should note that you don't have to run these simulations manually - the Choice Simulator available in our software gives you an easy way to run sensitivity analysis.