I think the ability to interpret data from this pilot is severely limited by the way it was structured. With a fixed size pool that was divided up based on years registered, the RoI for participants was very uncertain, which will lead to very different behaviour from participants. Early in the program with high RoI, rational participants may have registered and renewed names in the expectation of a high discount (even potentially >100%), meaning we can’t extrapolate from their behaviour here to how people would behave with an actual referral program.
I think any subsequent pilot needs to operate on a more realistic basis: x% (say, 10%) of revenue from a referrer returned to that referrer, with the pilot terminated when the funding is exhausted or the program expires.