Referral

For a long time the way to add users was to buy them. Buy ads, buy search placement, buy partnerships. The trouble with that model is that the price keeps climbing: as more companies bid on the same keyword the cost of one user rises, and the moment the spend stops, so does the traffic.

Meanwhile the path where an existing user brings in a colleague went uncounted for years. If the acquisition table has three rows — ads, search, direct — then everyone who arrived through an invite link a coworker sent lands in the direct bucket. Nobody can tell how much that path turns, or where in the product it turns.

There's a Pagedeck meeting where this surfaced. Grace asked where the month's 4,000 new users came from, and Ravi answered: 2,600 from ads, 900 from search, 500 direct. Grace asked how many of those 500 arrived through an invite, and there was no answer, because the invite links carried no tag to separate them. That is why referral occupies one of the five AARRR boxes. Unless it's counted on its own, you can't tell whether the path exists at all.

Three numbers do the work in this box.

1. The share of users who have sent at least one invite. If 120 of 1,000 surviving users sent an invite, that's 12 percent.
2. The share of sent invites that turn into a signup. Ninety signups from 300 invites is 30 percent.
3. The two multiplied together: people brought in per existing user. In the example above, 1,000 users brought in 90 people, or 0.09 each.

That third number is the report card. Above 1.0 a user base doubles itself, and very few products ever cross that line. Moving from 0.09 to 0.25 is worth having on its own: it means 360 of a month's 4,000 new users arrive for free, and that many fewer have to be bought.

Two things set this box apart from the ones before it. First, the population is small. Only people who entered through acquisition, got through the activation stage, and survived into the retention stage ever reach it. On a funnel that starts with 10,000 visits, roughly 300 people arrive here. Polish the invite screen all you like — if an earlier box is blocked, the absolute number won't move.

Second, there's a lag. Ad spend lifts visits the same day; invites don't. The moment a user gains a reason to invite someone might be two weeks after signup, and the recipient may click the link several days after that. Dividing this week's invites by this week's signups reads low. Group invites by the week they were sent and follow them for about four weeks instead.

The difficulty of this box varies enormously by product type. A collaboration tool that only pays off with several people in it has invitations built into the feature set: sharing a document requires inviting someone, so invites go out without anyone thinking about marketing. A notes app used alone has no in-product reason to invite anybody, so the incentive has to be manufactured with a reward — which is why the same referral tactic can read as a runaway success at one company and as wasted budget at the next.

Referrals turning doesn't lift the revenue box by itself. Three things have to line up first.

1. The people brought in have to clear the earlier boxes. If the 90 invited signups leave without creating a document, this box rises and nothing behind it does. Split invite traffic out and read its activation and retention rates separately; when they sit below the numbers for paid traffic, the problem is in how people are being recruited.
2. The reward can't eat the revenue. Giving both sides a free month per invite lifts invite volume quickly, but it also pushes back the moment payment would have happened. If the free time handed out to win a user exceeds what that user will pay, the trade loses money.
3. The people being brought in have to sit near purchasing authority. In a collaboration tool, an invite that pulls in five teammates turns one team into a paying candidate. An invite that pulls in five people from five different companies produces five free users. Same count, far longer road to revenue.

Compressed into one line: the goal of this box is not invites sent but paying teams arriving through invites. That reframing also changes what gets built. A share button that reaches the widest audience serves the first goal; a flow that makes adding the rest of one team easy serves the second.

"Referral went from 0.09 to 0.21. Fifteen percent of new users now come through invites."

Each user brings in more than twice as many people, so the dependence on paid acquisition dropped.

"Invites are up but they're all people at other companies. This won't turn into upgrades."

Volume rose without spreading inside the team, which is the unit that pays.

"Let's look at retention before we attach a referral reward. Right now the bucket has a hole in it."

Attaching invites to a product people don't return to just means the invited people don't stay either.

Last quarter, 90 of Pagedeck's 320 users in this box were sending invites. When Ravi tagged every invite link and separated the traffic, what showed up was that almost all 90 senders belonged to paying teams. The free plan capped sharing at three people, so inviting a fourth required a payment. Raising the cap to five lifted invite volume 40 percent and left upgrades flat, because the extra invited users became next quarter's upgrade candidates.

The most common belief is that a reward makes this box turn. Campaigns paying roughly $8 per invite get retold often enough to make it sound reliable. A reward speeds up a wheel that is already turning; it doesn't start a stopped one. Recommending something to a colleague puts your own reputation on the line, and few people will do that for money about a product they stopped using. Left uncorrected, a team spends the reward budget on top of weak retention, and the people it brings in leave for the same reason everyone else did.

Managing the box with a single number is the other mistake. Told to raise people-brought-in-per-user, a team finds the cheapest way to raise it — reading the whole address book and blasting invitations out. The count goes up, the recipients call it spam, and enough spam complaints mean the product's other mail stops arriving anywhere. Read invites sent alongside how recipients responded, or the number tells you nothing.