How can we make sure error budgets don't optimise for the wrong things

Coming from the discussion in gitlab-com/www-gitlab-com!77903 (comment 541679505).

I pointed out a project by the registry team, which would tighten up SLIs. This would mean that they would spend more of their error budget, but gain visibility into the user experience of the service.

How do we ensure that projects like this don't get put on the back burner because it would negatively affect this measurement?