Referral Program

The cost of acquiring someone through advertising climbs over time, because more companies keep bidding for the same placements. Meanwhile, what an existing user says to the colleague at the next desk lands better than any ad, since it comes from someone with no reason to praise it.

The problem was that this route had no volume control. Ask new users how they first heard about the product and the most common answer is a document link a coworker sent them. It's the largest source of signups and the company didn't build it — it happened. Double the ad budget and impressions go up; this route has no dial to turn.

A referral program turns that accident into a mechanism. It issues a link that connects who brought whom, and pays out when the connection is confirmed. A handle gets attached to something that was already happening on its own. The handle is small — a link and a payout rule — and it's the difference between a route you report on and a route you can plan with.

It takes two parts: a unique link that ties the referrer to the person who arrives through it, and a reward paid when that connection is confirmed. One without the other doesn't run.

Five things need deciding.

1. Tracking. Every user gets their own link or code. Without a way to connect who brought whom, there's nobody to pay.
2. The reward condition. Does it pay when the other person signs up, or only when they pay? Pay on signup and accounts created purely to collect start showing up.
3. Who gets rewarded. Pay only the referrer and the message reads as an ad to the person receiving it. That's why a double-sided incentive, paying both, is the default.
4. The form of the reward — cash, credit inside the product, or extra seats. Credit that only works inside the product brings people who intend to stay; cash also brings people who collect and leave.
5. When to ask. Not right after signup, but right after the product has shown what it's for. The screen after someone edits a first document with a colleague is that moment.

Results get read through the viral coefficient, the number of new users one user brings in. If 100 users brought 12 people, it's 0.12. Above 1 the user base grows without advertising, and very few products actually produce that number. Somewhere between 0.1 and 0.3 is the common range, which doesn't replace advertising — it discounts it. A coefficient of 0.15 means every hundred users you pay for arrive with fifteen you didn't, which shows up as a lower blended cost rather than as growth on its own.

Cost gets measured on the same ruler as ads. Divide what went out in rewards by the number of people brought in, and that cost per person has to land under what an ad costs to bring the same person for the mechanism to be worth running.

Three settings suit it.

The first is a product that gains value as more people join. A collaborative document tool is half a product used alone, so users already have their own reason to invite a colleague. The reward adds weight to a reason that exists.

The second is a market with expensive ads. Where acquiring one person costs $23, an $8 reward that brings one in is a meaningful difference.

The third is a product where recommendation is already happening. If colleague referral ranks high when you ask how people arrived, the mechanism is attaching to that route. Attach it to a product nobody recommends and only the rewards go out.

Some settings don't suit it. A product used once with no reason to return never reaches the moment of recommending. In categories people don't want to be seen using, a reward won't get a link sent.

Once it's live, abuse management follows. Someone making a second account for themselves, or creating accounts in bulk, always turns up. Set the reward condition at payment, cap what one person can collect, and build a process for reviewing and clawing back afterwards. Put the cap and the clawback in the terms before launch — adding them after people have collected turns an abuse problem into a support problem.

"Referral links accounted for 18 percent of new signups this month."

A route that used to happen by accident is now being counted.

"Paying on signup brought in fake accounts. Let's move it to payment."

The condition is loose enough that the mechanism is being used for something else.

"At a viral coefficient of 0.12 this can't replace advertising."

Treat it as a secondary route rather than a primary source of signups.

At Pagedeck, Ravi attached the referral prompt to the screen after someone first edited a document with a colleague. An earlier attempt on the post-signup screen produced almost no links, because it asked for a recommendation from people who didn't yet know what the product was.

The reward became a free month of a seat for both sides. Cash increases the number of accounts that collect and leave, while a seat tends to get used, and using it often means pulling in one more colleague. After it launched, Mia confirmed that 18 percent of signups came through referral links and decided to track retention for that route separately.

The most common mistake is expecting a referral program to create word of mouth. Rewards moving people is an intuitive idea. But this mechanism can't manufacture recommendations that weren't happening; it counts the ones that were and nudges them up. Attach rewards to a product nobody has a reason to recommend and the people who come are the ones after the reward, and two months later none of them are there. The order is backwards, and what to check first is whether anyone is telling someone else about this product today. If the answer is nobody, the work belongs to the product rather than to the payout.

The second is thinking a bigger reward brings more people. Double the payout and the number of people sending links looks like it should double too. What actually rises is the share of arrivals who came for the reward rather than the product. The count goes up while retention goes down, so cost per person looks lower right up until those people disappear three months later. That's why the decision to raise a reward gets made on the share still around after three months, not on the number brought in.