Affiliate program metrics: the numbers to review every week

Three months into a partner program, the question is simple: is it working? A dashboard answers with more numbers than the question needs, and several of them can point the wrong way: an average pulled up by one partner, a total that mixes last week's signups with last year's, a "+100%" built on two extra sales.

This guide picks a few numbers and shows how to read them. The weekly review is for spotting what needs you. The monthly review is for judging whether the program works. There is a worked example, a scorecard to copy and a checklist at the end.

Four rules for reading any partner number

1. Medians, not averages

When one partner sells far more than the rest, the average describes nobody. Take 14 partners who each made at least one sale in a month, with $18,400 of sales between them. The average is $1,314 per partner, and eleven of the fourteen brought in less than that. The median, the partner in the middle, is $420: half brought in more, half less. That is the figure a typical partner can be compared with.

2. Cohorts, not totals

A program-wide figure mixes the partner approved yesterday with the one approved a year ago, and every new approval drags it down before that partner has had a chance. Group partners by the month they were approved and compare one month's group with the next. Affiliate onboarding shows the table for the first 60 days.

3. No percentages on a small base

Going from 2 sales to 4 is "+100%", and it means almost nothing. A first month has nothing to compare with at all. Put the counts beside any change, and leave the percentage out while the counts are small.

4. Know when a number is too small to mean anything

Every rate is an estimate, and the fewer events behind it, the wider the range of true rates that could have produced it. 3 refunds out of 10 sales is 30%, but a partner whose real refund rate is anywhere from about 11% to 60% could plausibly produce it (a 95% Wilson interval, the method in NIST's statistics handbook). With 30 refunds out of 100 sales, the range narrows to about 22% to 40%.

So write the count beside every rate, and before you judge a partner on one, ask how many events it rests on. Affiliate onboarding works through the same arithmetic for clicks with no sales, and affiliate fraud prevention for refunds.

Every week: what needs you

The weekly look should end with a short list of people to contact, not a verdict on the program:

• Applications waiting. How many, and how long the oldest has waited. A partner who hears nothing for a week has had a week to forget why they applied.
• New partners with no click. Approved in the last 30 days and nothing yet. A nudge now costs little.
• Partners with clicks and no sale. Offer help early: a better link, a code, a question about what their audience responds to. Save judgment for when the clicks are enough to tell chance from a pattern.
• Your best partners going quiet. A partner who sold in most weeks and has not sold in the last two or three is worth a message before the silence becomes a habit.
• Anything unusual. A jump in one partner's refunds, clicks far above their normal, code sales with no matching clicks. Affiliate fraud prevention covers what each can mean.
• What is owed. Commission approved and waiting for payday, and anyone who will not be paid and why. How to pay affiliates has the routine.

None of these is a percentage. Each is a name and a reason to write to them.

Every month: is it working?

Six numbers answer it. Keep each one per currency if you sell in more than one, and keep the count beside each rate.

1. Revenue from partners, after refunds, and its share of all revenue

What it is: sales credited to partners in the month, less refunds and chargebacks, and that figure as a share of your total revenue.

The trap: counting what has not been paid. An order placed is not money received, and a deal in a sales pipeline is not revenue until the customer pays. Never add one currency to another without converting, and say at which rate if you do. The share needs your total revenue, which comes from your own books, not from your partner software.

2. Activation by cohort, and the time it takes

What it is: of each month's new partners, the share who made a sale within 60 days, and the median number of days from approval to the first click and from the first click to the first sale.

The trap: one activation rate for the whole program, for the reason in rule 2. Affiliate onboarding covers what to do about each stall.

3. Active partners, and revenue per active partner

What it is: how many partners made at least one sale in the month, and the median revenue among them, with the highest and lowest beside it.

The trap: revenue per partner across everyone you have approved. That figure falls every time you approve somebody, whatever happens to sales. Pick one definition of "active" and keep it, or the number moves for reasons that have nothing to do with your partners.

4. Share of revenue from your top partners

What it is: the share of partner revenue from your single biggest partner, and from your top three.

What it tells you: a program where one partner brings half the revenue is a partnership with one person, plus a long tail. That is not a failure, but it is a risk to manage: talk to that partner more often than anyone else, and know what happens to the month if they stop. A share that falls while total revenue rises means the program is broadening, which makes it less fragile.

5. Refund and dispute rate, by partner

What it is: each partner's refunds and disputes as a share of their sales, beside the program's rate, with the counts.

The trap: small counts (rule 4). One bad month from a partner with ten sales is a reason to watch, not to act. A rate that stays well above the program's for two or three months is a reason to ask.

6. Cost per new customer, beside your other channels

What it is: everything the program cost in the month, divided by the new customers partners brought. Count commission, bonuses, free product or gifts, the software, and an honest estimate of your own time. Then work out the same figure for paid ads, the same way.

The traps:

• Count new customers only. A returning customer credited to a partner is a cost with no acquisition behind it.
• Some credited customers were already on their way. A partner bidding on your brand name, or a code from a deals site used by someone already at checkout, makes the figure look better than it is. Affiliate attribution explained covers how sales are credited, and affiliate fraud prevention covers both patterns.
• Match the months. Count commission in the month of the sale it was earned on, not the month you paid it, so the cost lines up with the customers.

Reading them together: an invented month three

A store selling in dollars, three months into its program. Every figure is invented:

What it says:

• The program is cheaper than paid search, so far. $38 against $55 per new customer, on 92 customers, is enough to keep going.
• Half of it is one partner. The most valuable conversation this month is with that partner. The second is working out why the top three are so far ahead of the other eleven, and whether any of them can be helped to do the same.
• The number to grow is $420. That is what a typical active partner brings in. Activation and help move it. The average mostly moves when the top partner has a good month.
• The partner with 2 refunds in 9 orders is one to watch, not judge. On nine orders, that is consistent with a true rate anywhere from about 6% to 55%. Look again next month.

Monthly scorecard

Copy this into a spreadsheet and fill one column a month. The last column is the one that keeps the others honest.

Metrics checklist

Every week

• Applications answered within two working days
• New partners with no click nudged
• Help offered to partners with clicks and no sale
• Best partners who have gone quiet contacted
• Unusual refunds, clicks or code sales looked at
• What is owed checked before payday

Every month

• Scorecard filled in, with counts
• Revenue counted after refunds, per currency
• Activation read by cohort, not program-wide
• Revenue per active partner read as a median
• Top partner's share noted, and a call booked with them
• Cost per new customer compared with paid ads, worked out the same way

Every time

• No percentage change on a small base, and none on a first month
• No partner judged on a rate before the count can support it

How Refficks shows these numbers

Refficks is a new partner program platform. These are the screens behind the numbers above, and what they leave to you.

• Overview and Reports. Partner revenue, commission, sales and clicks for today, the last 7 or 30 days or dates you choose, and 12 months on Reports, with a chart of every day in the period, "including the quiet ones."
• One currency at a time. Revenue and commission are read in one currency at a time and never added across currencies. When a period holds more than one, a picker chooses which leads, and the overview lists the others on their own lines beneath it.
• No percentage on nothing. Where the days before had nothing to compare with, the overview names them and says so ("Nothing in Sep 1 – 30 to compare with") instead of printing a percentage.
• Activation and cohorts. The activation funnel shows how far partners get, names who is stuck at each stall, and gives median days to a first click and a first sale: "The middle partner, not the average one." A cohort table, "Do the partners you recruit keep earning", follows each month's new partners' revenue month by month.
• The share of your sales. "Where your revenue comes from" says what share of your sales reached a partner, and what the rest was, in words such as "No partner ever touched this customer".
• Concentration. "How much rests on how few" shows the share of partner revenue from your top five partners.
• What the commission brought back: revenue for every unit of commission, with the caveat on screen: "Not profit: nothing here knows what the goods cost to make." Cost per new customer stays yours to work out, because only you know the other costs. Each partner's page shows the customers they brought, their average order and revenue per customer.
• Refunds, as the commission they took back. Partner revenue is shown as sold, with commission clawed back by refunds beside it, overall and per partner. For revenue after refunds, use your store's or billing tool's refund figures.
• Small numbers held back. Some judgments wait for enough to go on: "Too few clicks to judge the conversion pattern." below 20 clicks, and "Refficks needs 4 customers from a partner before what happened to them says anything."
• Scores you can read. Each partner's score is "Their rank in your programme, not a grade", worked out nightly, with every factor, its weight and a sentence of evidence.
• Results, not claims. The Results tab shows what happened in the 30 days after you acted on a recommendation: "Things that happened after. Never proof that acting caused them."
• A weekly or monthly email, if you switch it on: clicks, sales and new customers, revenue and commission per currency, and the three partners who earned the most.

Start free. No card needed.

Sources

All read on October 16, 2026.

• NIST/SEMATECH, e-Handbook of Statistical Methods, 7.2.4.1. Confidence intervals: the Wilson interval for a proportion, used for the ranges in rule 4.