Deciding to focus on one or the other without data is bad practice. How many defects do you find with each? What is the cost of finding each defect? What is the cost of fixing each defect? What about defects that could have been caught at unit test but were only found at integration test where it typically costs more to find and fix? As a general rule, the better the coding (and code inspection) practices, the fewer defects are found in unit test - so with sufficiently good practices the return on investment of unit tests might make them not worth doing, and I have heard of organizations where this is the case, but unless you have data you are just guessing.