The owner of an online coffee shop in Almaty forwards you a partner's report: "Twenty customers brought in another twenty. Each one brings one more, so we won't need ads soon." You open the export. A hundred buyers took part, eighty of them invited nobody, and they simply never made it into the calculation. Without them the denominator shrank, the result grew fivefold, and nobody has talked about money yet.
The shop is in the second month of its referral programme. The numbers in this lesson are illustrative. The order economics are the same as in the influencer campaign: a kilogram costs 15,000 ₸, and after beans, roasting, packaging and delivery the shop keeps 6,000 ₸. Both people get a gift for the introduction: the new buyer and whoever invited them.
From a recommendation to the next buyer
The loop goes like this. Ainur received good coffee, told a friend about it, the friend placed a first order and can now invite someone herself. In Reforge's description of growth loops, the output of one pass becomes the input of the next. For the shop, that output is new buyers who can recommend the coffee in turn.
First, the person needs a reason to recommend it. A convenient link makes the offer easy to forward, but Ainur still has to explain to her friend why she would want this coffee. If the delivery disappointed her, the promised gift turns her message into a favour for the shop, and it stops being a personal recommendation. So show the invitation after the buyer has received the order and tried the coffee.
A programme and a loop are different things. You can pay a bonus for a referred friend once, and that is the end of it. You can talk about repeating growth when you see new buyers inviting the next ones themselves. In the AARRR funnel, this is the referral stage feeding back into acquisition.
Agree on who counts first
For the shop, the result is a new buyer's first paid order, delivered and not refunded on the reporting date. Registrations, link clicks and existing customers' orders go into other rows. Rewarding new accounts would be risky here: one person can create several accounts and buy nothing.
The report records three things: how you identify a new buyer, how long an invitee has from invitation to order, and the day the data is ready. In this programme, the first order must come within 30 days of the invitation. Give the last invitee the full 30 days too. If their order is still on its way or a refund is unresolved, mark the result as incomplete. And do not change the rule halfway through to get a nicer number: months measured under different rules cannot be compared afterwards.
For a service used by teams, the unit is different. An employee invited into a company's already paid account is still the same customer; it just has more users now. If the goal is to bring in new companies, the result is a new organisation that completes the chosen action. Employees and companies do not go into the same fraction.
All hundred buyers stay in the denominator
For the first calculation, take a simplified export. A hundred eligible buyers could take part. Twenty of them sent two hundred invitations. The recipients are all different, none had bought from the shop before, and each received one invitation. Every window has closed, and twenty recipients met the first-order condition.
The diagram calls such a buyer an activated user: in this calculation, activation is a completed first order.
Simplified example of one cycle
All observation windows have closed
Original group
100 eligible buyers
20 sent invitations
Invitations
200 invitations
Distinct recipients, one invitation each
First order completed
20 new activated users
Had not bought from the shop before
K for the chosen cycle = 20 / 100 = 0.2. All original buyers stay in the denominator.
| Metric | Calculation | Result |
|---|---|---|
| Buyers who sent invitations | 20 / 100 | 20% |
| Invitations per sender | 200 / 20 | 10 |
| Invitations per original participant | 200 / 100 | 2 |
| Invitees who met the first-order condition | 20 / 200 | 10% |
| New buyers per original participant | 20 / 100 | 0.2 |
The last number is the K-factor for the chosen cycle. In the simplified model, K = i × c: average invitations per original participant times the share of invitees who reached the result. Here, 2 × 0.1 = 0.2.
The partner took ten invitations per sender and got 10 × 0.1 = 1. But that is new buyers per sender, not per programme participant. To bring the whole group back in, multiply by the share who sent invitations as well: 0.2 × 10 × 0.1 = 0.2. For the owner, skip the formula: twenty new buyers per hundred original ones.
Handle duplicates and unknown sends
In a live programme, one person gets links from two friends, clicks on a phone and buys on a computer. Decide before launch who gets credit for that order, and keep the link between invitation, buyer and order. For example, credit the first confirmed invitation within the window. This is your programme's decision: there is no single correct rule here.
Before calculating, deal with existing customers, duplicate accounts, cancelled orders and self-invitations. The clean-up has to run through the whole calculation at once. If you then divide unique buyers by raw clicks, you get a fraction built from two different samples, and you cannot call it the conversion rate of unique invitees.
A "Share" button does not tell you whether the person sent a message. If the channel does not report that, name the event for what it is: opening the share dialog. The shop reliably sees referral-link visits and orders, but may not know how many invitations were delivered. Then leave the invitation-to-order conversion blank. But you can calculate the bottom line: twenty confirmed new buyers per hundred participants.
The buyer forwards the link to a friend, on WhatsApp for example. The shop does not need to collect anyone's address book for a mailing. The recipient opens the offer and decides whether to order.
The shop pays for the gift
Suppose each confirmed first order earns two sample packs: one for the newcomer and one for the person who invited them. A pack costs the shop 1,000 ₸, as in the influencer campaign. Assume all rewards have been handed out and delivery is included in their cost. Handling this batch of orders and gifts costs another 10,000 ₸.
Twenty first orders bring 20 × 6,000 = 120,000 ₸ of contribution before programme costs. Gifts cost 20 × 2 × 1,000 = 40,000 ₸, and with the handling it comes to 50,000 ₸. That is 2,500 ₸ per buyer credited to the programme, and the batch keeps 70,000 ₸ of contribution after those costs. The business's fixed costs are not included.
This is still a report on orders credited to the programme. Some friends would have bought without a gift, and some buyers will forget the link and order directly. So 2,500 ₸ is not the cost of an additional customer. To find that, you need an effect measurement plan, and it has to allow for friends in different groups who may forward the offer to each other and mix the test conditions.
Points instead of gifts do not remove the cost. Describe when they can be spent, what happens to them after a refund and how their cost is counted. If points have already reduced the amount paid, do not subtract that discount from contribution a second time, as in the loyalty programme economics.
How quickly the loop repeats
K = 0.2 over seven days and K = 0.2 over three months give the same number of new participants per original one, but the growth comes at different speeds. So next to the coefficient, write the group's start date, the window rule and the time to the result.
In an ideal model with identical generations, K above one produces growth from generation to generation. In real life, the shop's buyers sit in the same Telegram and WhatsApp chats, gifts get old, and new buyers may recommend less often than the first ones. So the next row of the report is about the twenty new buyers: how many were offered the programme, how many invited others and how many new orders came from them.
Repeat purchases by existing customers belong to retention and do not enter the acquired generation a second time. That way one report holds two questions: how many new people the programme brought and how many of them stayed with the shop.
Exercise: prepare the calculation for your project
Take a business whose customers have someone to recommend the product to. Build one table for a group where you can see every step.
0 of 7 done
To check yourself, here is another table from the partner: out of five hundred eligible participants, fifty sent invitations and a hundred new buyers came. His K = 2 is per sender: 100 / 50. For the whole group, K = 100 / 500 = 0.2. Even with the right denominator, the report is not done until unique buyers, a closed window and completed orders are confirmed.
In the paragraph for the owner, name the group, the period and the buyers per participant, then the costs and the decision. The decision might be to continue a small launch and watch the next generation, or to sort out duplicate accounts first. Do not promise growth without ads from a report like this: with K below one, even in the ideal model each generation is smaller than the one before.
Video: how referral mechanics work
The video shows the mechanic itself. The invitation's result, the period and the reward cost for your own project are yours to define.