The error bound I got does not seem very tight with respect to the actual max abs error. Is this to be expected in some cases?
Not very tight indeed, but it depends how that is “defined”.
Bounds that are at most one order above the true error are considered still informative.