Referral programmes are cheap to launch and frequently produce nothing. The difference between the ones that work and the ones that do not lies in the structure rather than the reward.
The product has to be shareable first
Referral works where people already discuss the product in ordinary conversation. A scheme amplifies existing word of mouth; it does not create a reason for anyone to talk.
Products that are used privately, that are awkward to mention, or that carry no social signal generate few referrals regardless of how generous the incentive attached to them is.
Which is why the first useful diagnostic is whether unprompted recommendations already happen. Where they do not, a programme is usually paying a considerable sum to discover that.
Rewarding both sides changes the ask
A one-sided reward asks the referrer to trade a friend's attention for personal gain, which a large share of people are reluctant to do openly.
A two-sided reward reframes the same act as passing on a benefit. The referrer is offering something rather than extracting something, which removes most of the social cost.
This distinction matters more than the amounts involved. Modest two-sided offers frequently outperform considerably larger one-sided ones for exactly this reason.
Timing determines participation
The moment a customer is most likely to refer is shortly after a good experience: a successful first use, a problem resolved well, or a result they were hoping for.
Programmes that surface the invitation at those moments capture far more activity than ones relying on a permanent link in an account menu that almost nobody visits.
Asking too early, before the product has demonstrated anything at all, produces low participation and can damage the impression of the product being recommended.
Attribution decides what gets paid for
Many schemes reward introductions that would have occurred without any incentive, which simply converts organic acquisition into paid acquisition with no gain in volume.
Distinguishing the two requires comparing behaviour against a group not exposed to the offer, which is rarely done because the headline numbers look good without that comparison.
The honest measure is incremental customers rather than referred customers, and in most programmes those two figures are a long way apart.
Abuse scales faster than the programme
Any reward attached to a repeatable action attracts people optimising for the reward itself, through self-referral, fabricated accounts or organised networks of participants.
Controls such as requiring qualifying activity before payout, limits per account and manual review of unusual patterns add friction that also reduces legitimate participation.
Finding the balance is continuous rather than settled, because the methods used to exploit a scheme change as soon as the initial controls are put in place.