When a referral program is just a coupon

What makes a referral program a coupon?

The main variable. A referral program's job is to shorten the distance between a customer wanting to tell someone and them telling them. A coupon's job is to change a price. When the biggest lever in your program is the size of the reward, you have built a price promotion with a social story on top of it, and it will behave like a price promotion: a spike, a plateau, a cohort that is worth less than your average customer, and a quiet decision eighteen months later that referral does not work in your category.

I want to be precise about why this matters, because the usual version of this argument is moral and the real problem is arithmetic. A discount is paid once and shows up in margin. A referral reward is paid per acquisition and shows up in CAC, and it is defended in the board meeting as a cheap channel. If the referred customers are no better than your paid ones, and some share of them would have arrived anyway, the reward is a rebate you are accounting for as acquisition. That is the part that goes unnoticed for a year.

The appetite for recommendation is not the problem. Nielsen's survey of more than 29,000 consumers across 58 countries, fielded in February and March 2013, found 84% completely or somewhat trusted recommendations from people they knew, well ahead of any paid format. McKinsey Quarterly in April 2010 put word of mouth as the primary factor behind 20% to 50% of purchasing decisions. When a program underperforms against a market like that, the mechanism is almost always what failed.

How do you tell which one you built?

Three checks. None of them needs a vendor and all three can be done with the data a running program already produces.

The holdout check. Exclude a random 10% of eligible customers from the program entirely for 60 days. Nobody in that group sees the ask or the reward. Then compare the two groups on referred signups per hundred customers. If the holdout produces a similar number, you are paying for referrals you were already getting, and the program's real contribution is the difference, not the total. This is the check that almost never gets run, because running it requires being willing to see the answer.

The cohort quality check. Take referred and non-referred customers who signed up in the same weeks, and compare their 90-day retention and revenue per customer. The reference point from research is clear: in a German retail bank studied by Schmitt, Skiera and Van den Bulte and published in the Journal of Marketing in January 2011, referred customers churned about 18% less and were worth roughly 16% more over six years. If your referred cohort retains worse than your non-referred one, that is the signature of a reward attracting people who want the reward. It is the single most diagnostic number in the whole exercise, and you can only produce it if the referrer was stamped onto the referred customer's record, which is covered in the referral moment.

The concentration check. What share of successful referrals comes from your top 1% of referrers? Some concentration is normal and healthy, because enthusiastic customers exist. But a very steep distribution combined with a reward that has cash value is usually arbitrage: a small number of people have found that your program pays, and they are working it. Look at those accounts by hand. You will know within ten minutes.

Which mistakes cost the most?

In order of what they cost, which is not the order in which people worry about them.

Launching before retention flattens. The most expensive by a wide margin, because you pay a reward for every customer in a cohort that was going to leave. Worse, the German bank data published in the Journal of Marketing Research in 2018 showed that referred customers' retention advantage held only while the referrer was still a customer, so referral on a leaky product produces churn in correlated clumps rather than spread out. Draw the retention curve first. It is 30 minutes.

Putting the ask where the moment is not. Account settings, the footer, a monthly email. Each of these is cheap to build, which is exactly why they get built, and each is somewhere a person is not currently feeling anything about your product.

Paying the referrer only when the referred person purchases. In a business with a long sales cycle this means the referrer does something in March and hears about it in June, which breaks the loop as a learning mechanism. Split it: acknowledge at the invite, reward at the referred person's activation, and add the bigger reward at purchase if you want one. The acknowledgment matters more than the money.

Making the reward cash or near-cash. Cash is worth the same to everybody, which means it attracts people with no particular attachment to your product. Dropbox's two-sided reward was storage, which is to say the product itself, given to two people who both wanted more of it. Drew Houston's 2010 deck reported that program permanently increasing signups by 60%, with 35% of daily signups coming from it. The reward being the product is not a detail of that story.

Measuring it with a "how did you hear about us" dropdown. Asked at signup, before the person knows what to call you, with your existing channels as the options and no open text box. It will under-report word of mouth every time, and then the under-report becomes the evidence for killing the program. Measuring word of mouth has the question that works and the timing.

Is a discount ever the right answer?

Yes, and this is where I want to be fair to the thing I have been criticizing. Price promotions work. They clear inventory, they pull purchases forward, they win a competitive deal, and in some categories they are the main lever anybody has. If what your business needs this quarter is a price promotion, run one, call it one, and measure it as one: incremental orders against a holdout, margin per order, and what happens to the following quarter.

The damage is not in running a discount. It is in calling it referral. A mislabeled program produces a conclusion that is wrong in a specific and lasting way: that your customers do not recommend you. That conclusion closes off the one channel that gets cheaper as it grows, on the evidence of a mechanism that was never testing it.

And there is a version of this that is worse than useless. A reward large enough to be interesting on its own, with no holdout and no cohort comparison, run for a year, teaches your customers to attach a price to the act of recommending you. That is difficult to undo, because the next time you ask without a reward, nothing happens, and you now have evidence for the wrong conclusion.