Why referral programmes fail and where to look first
A referral scheme gone quiet? Check your measurement, count the four steps, form hypotheses including the reward, and choose one test.
Your referral scheme launched, then went quiet. Before you change anything, find out where people are dropping out, then test what might be causing it. This guide sets out a diagnostic method.
The numbers are invented for illustration and are not benchmarks.
First check you can measure the steps
A quiet scheme is not always a failing scheme. Before diagnosing, check what each of your counts actually measures.
- Who was shown the offer, and how do you know?
- How do you track a share: a link click, a code, a form?
- What happens to referrals that arrive without a code, for example when a customer simply messages a friend, and the friend then mentions the customer's name?
- Can you see private invitations at all, or only shares made through your own page? If you cannot observe them, say so and treat the share count as a lower bound.
- Do your counts use the same group of people and the same follow-up window?
Denominators matter. A drop from 150 to 60 means something different if the 150 includes people who never saw the offer. If your tracking misses whole routes, the scheme may be working better than the numbers say. Fix the measurement first, or at least write down what it misses.
The four steps where people drop out
Follow one group of referrers through the steps. Pick a cohort of unique eligible customers who were shown the offer in a chosen period, use customer IDs so nobody is counted twice, and follow them for a consistent window long enough for a referral to mature. Then count, within that same cohort:
- Customers shown the offer.
- Of those, customers who opened the share page or got their link or code.
- Of those, customers known to have shared (private invitations may not be visible to you).
- Of those, referrers who generated at least one qualified friend purchase.
Illustrative referrer cohort (invented numbers, with a 90-day example window):
| Step | Referrers | Share of previous step |
|---|---|---|
| Shown the offer | 1,000 | n/a |
| Opened the share page | 150 | 15% |
| Known to have shared | 60 | 40% |
| Generated at least one qualified friend purchase | 9 | 15% |
The last row counts referrers with at least one qualified referral, not friends or purchases. Friends, purchases and revenue can differ from the referrer counts, because one referrer can bring several friends or none, so count friends and sales separately. Do not label friends divided by referrers as a same-person funnel conversion rate; report any friends-per-referrer metric separately. The drop between steps, with its denominator, tells you where people leave. It does not yet tell you why.
Possible causes at each step, including the reward
Treat every explanation as a hypothesis to test, not a verdict.
| Step with a big drop | Hypotheses worth testing |
|---|---|
| Shown the offer to opened the share page | The ask is hidden, badly timed, or sent to people who have not had a good experience yet; trust in the offer; the referrer's reward is not worth the effort |
| Opened the page to known to have shared | Too many fields or steps; unclear what to say; discomfort about asking friends; the referrer's reward |
| Shared to at least one qualified friend purchase | The friend's offer is weak or unclear; the friend does not trust the business yet; the product does not fit the friend; the friend's reward |
The reward can matter at any step. A weak reward for the referrer can stop sharing before any invitation is sent. A weak offer to the friend can stop the purchase. So do not assume the reward is irrelevant until a friend has bought, and do not assume it is the cause either. List each hypothesis, then test one.
Choosing what to test first
The biggest proportional drop is one input, not the whole answer. Also weigh:
- Denominators. A big percentage drop on a tiny count may be noise.
- Business value. Which step, if improved, matters most to you?
- Effort. What can you change this week?
- Constraints. Is there anything you cannot change, such as a legal wording or a system limit?
Change one thing, set a review date and count again with the same kind of cohort. If you change several things at once, you will not know which did anything. Even then, a change followed by a different count is a diagnostic signal, not proof that the change caused it: season, volume and other changes can move the numbers too. This is practitioner judgement, not a proven rule, and your own results may point somewhere different.
Initial diagnostic checklist
Where Zenko fits
If you want to look at rewarding something other than a sign-up, our guide to verified real-world actions covers that idea. To ask about a campaign, contact Zenko. Availability and terms are confirmed for each campaign.
Related referral guides
- How to set up a referral programme one decision at a time
- Referral programme rewards: what each type really costs
Frequently asked questions
Why has my referral programme gone quiet?
There is no single cause. Check that your measurement captures every route, follow one cohort of referrers through the four steps, and test one hypothesis at a time. A weak reward is one possible cause among several.
Should I raise the reward if my scheme is quiet?
Not before you know where people drop out. A bigger reward may help if the drop is at the sharing step, but a hidden offer or a confusing share page could also be involved. Treat each as a hypothesis, test one change and compare, remembering that a difference is a signal rather than proof.
How long should I wait before recounting?
Pick a review date in advance that gives referrals time to mature and enough volume to compare, which depends on your business. With very low counts, treat the result as a prompt for the next question, not as proof.
Sources
- Zenko, How to prevent reward farming without punishing real customers: https://www.zenkoprotocol.com/insights/prevent-reward-farming/
- Zenko, Verified real-world actions: https://www.zenkoprotocol.com/insights/verified-real-world-actions/
General information. Worked examples use invented figures and are not benchmarks.