An assistant generates fifty replies in a day. That number says something about output, but little about whether the user’s communication became easier or more accurate.
Consider the work around each draft. Did the user need to correct invented details? Did they spend longer reviewing the suggestion than writing the reply? Did the tool identify a missing decision or simply generate another plausible sentence?
Useful evaluation can include the effort needed to produce an acceptable message, the kinds of corrections made and whether important follow-ups remain visible. Choose a method that fits the task and describe its limits.
For example, compare a set of fictional coordination messages with and without assistance. Record what had to be changed and why. Do not present a small internal exercise as proof of results for every user or relationship.
Some benefits may involve fewer messages. A clear summary can prevent repetition. Recognizing a completed exchange can avoid an unnecessary reply.
Before calling a communication tool successful, identify the burden it is meant to reduce. More text is easy to count. A more useful question is whether the person can complete the actual communication task with less avoidable effort while retaining accurate facts, appropriate tone and control over the final action.