Hold scope constant and compare three things: the prompt or keyword set and how it was built, the engines or channels covered, and what appears first in the monthly report. Totals differing by 3x usually reflect different scope, not different margin.
Ask both what they cannot control. An answer with no limits is a warning.
Ask who does the work, and check whether those people are on the calls by week six.
Ask what would tell them it is not working.