Payments, retention and operations
Refund rate: how to measure it honestly, find what drives it and fix it at the source
Read 30 days in, the example cohort in this post looks like a 2.9% refund rate. Once its guarantee window has closed, it's 5.3%. Here's how to measure refund rate so it doesn't flatter you, what it does to your break-even, and how to trace refunds to their cause and fix them where they start.
On this page
Refund rate is the share of orders you refund, or the share of revenue you pay back. Measure both, count each refund against the month its order was placed, and read a month only after its guarantee window has closed. Then split the rate by bundle, traffic source and VSL version, because the average hides the cause. The fixes start upstream, in the promise, the delivery and the billing.
How to calculate refund rate without flattering yourself
There are two formulas and one rule about timing:
Refund rate, two ways
Refund rate by ordersorders refunded ÷ orders placed
Refund rate by revenuemoney refunded ÷ revenue
Cohort refund raterefunds on one month's orders, whenever they arrive ÷ that month's orders or revenue
Illustrative March cohort, read after its guarantee window closed: 53 of 1,000 orders refunded = 5.3% by orders; $6,747 of $106,500 refunded = 6.3% by revenue.
The rate by orders tells you how many buyers were unhappy. The rate by revenue tells you how much money went back, and it's the one your margin math needs. The two differ whenever big orders refund at a different rate from small ones. Count partial refunds at their amount in the revenue rate; in the order rate, pick one rule, either any refund counts as a refunded order or a partial counts as a fraction, and keep it.
The timing rule matters more than either formula. The shortcut that misleads is dividing this month's refunds by this month's orders. Refunds arrive weeks after the orders they belong to, so a growing offer looks better than it is, because its newest orders haven't had time to refund, and a shrinking one looks worse.
Count each refund against the month its order was placed, a cohort, and don't treat a cohort's number as final until its guarantee window has closed. If your store lists refunds by the date they were paid, match each one to its order number first. Here's one month of orders filling in:
One month of orders, refunds still arriving
Illustrative March cohort: share of its orders refunded, by days since the order
Days since the order
Show the numbers
| Days since the order | Orders refunded |
|---|---|
| 0 | 0% |
| 10 | 0.9% |
| 20 | 1.9% |
| 30 | 2.9% |
| 40 | 3.8% |
| 50 | 4.6% |
| 60 | 5.1% |
| 70 | 5.3% |
A new offer is the extreme case. In its first weeks almost no order is old enough to be refunded, so its refund rate looks wonderful, right up until the cohorts mature.
What a refund really costs
A refund costs more than the money you send back. The product and the shipping of a refunded order are already spent, and the payment fee may not come back either, so check what your processor does on refunds. Nor do returned bottles go straight back into stock: FDA's manufacturing rule for supplements says returned dietary supplements must be quarantined until quality control reviews them, and destroyed or otherwise suitably disposed of unless that review approves salvage or reprocessing (21 CFR 111, subpart N).
Take one of the example cohort's three-bottle orders, at $129. Kept, it leaves $100.84 after the bottles ($15), the shipping ($8) and the payment fee ($5.16). Refunded, it leaves you $28.16 out of pocket: a swing of $129 from the kept order, with nothing left to resell. If your processor returns the fee, the loss is $23.
Refunds feed straight into break-even. The example cohort's orders average $106.50, carry 2.55 bottles at $5 landed ($12.75), cost $8 to ship and pay 4% in fees. Its refunds by revenue, 6.3%, go in the refunds line:
Calculator
Break-even ROAS and CPA calculator
Your numbers per order. Below the break-even ROAS, every order loses money before fixed costs.
- Profit per order before ads
- Contribution margin
- Break-even cost per order (CPA)
ROAS = revenue ÷ ad spend. Refunds are counted as revenue you give back; the product cost and shipping of a refunded order are already spent.
At 6.3% the average order leaves $74.78 before ads and breaks even at a ROAS of 1.42x. If refunds reached 12% of revenue, it would leave $68.71 and need 1.55x, so every campaign would need about 9% more revenue per ad dollar just to stand still. Use the revenue rate from a closed cohort here, not last week's number.
Why a fast refund beats a chargeback
A refund costs you the order. A chargeback costs you the order too, plus fees, and it counts against you with the card networks:
- Fees. On a US Stripe account, as of October 2026, each dispute costs a $15 fee, and for disputes opened after June 17, 2025, countering one costs another $15, returned if you win (Stripe). Other processors set their own fees.
- The count. Card network monitoring programs track disputes, and Stripe's guide says they don't consider refunds when identifying disputes, or dispute outcomes (Stripe). A refund issued after the buyer has gone to the bank doesn't take the dispute off the count.
So the cheapest refund is the fast one: answer requests within a business day and refund the same day. Making refunds hard to get lowers your refund rate on paper and can push unhappy buyers toward their bank instead, where each one costs more. How the networks count disputes is in chargeback ratio, and the chargeback reduction post takes each dispute reason in turn.
Split the refund rate to find the cause
An average blends very different groups of orders. The March cohort's 5.3% looks like this by bundle:
Refund rate by bundle, March orders
Illustrative cohort, share of orders refunded
The six-bottle pack refunds at two and a half times the single bottle's rate, and because each of those refunds is $219, the six-packs account for nearly half the money refunded. That's a question for the page, not the buyers: what did the offer promise the people who chose six bottles, and did they get it?
Split the rate the same way by:
- Traffic source and ad angle. A source with cheap orders and a high refund rate can cost more per kept customer than an expensive one. Cost per kept customer = CPA ÷ (1 − refund rate): a $60 CPA with 20% of orders refunded is $75 per kept customer, more than a $70 CPA at 5% ($73.68), and that's before the product and shipping the refunded orders burned.
- VSL version. A version that promises more can win on revenue per visitor and lose once refunds come in. In RevvyScales, folder splits, and variant tests on the Growth plan and up, report orders and revenue for each version (reading your analytics); add each version's refunds from your store before you call a winner.
- Days to refund. When a refund arrives says almost as much as why. Refunds before delivery point to delivery or billing. Refunds in the first weeks point to a gap between the promise and the product, or a charge that surprised someone. Refunds near the end of the window point to buyers who didn't see results, or who used the guarantee as intended.
Code every refund reason
Free text can't be counted, but a short list of reason codes can, by bundle, source and VSL version, week after week. Ten codes are plenty to start with: not as described, no results yet, arrived late or not at all, unexpected charge, didn't know how to use it, damaged or wrong item, ordered by mistake, changed mind, reaction or side effect, and other.
Keep the reason optional for the buyer and apply the code yourself. If "other" starts to fill up, read those requests and add a code. And a reaction or side effect is more than a refund code: pass it to whoever handles adverse event reports for your brand.
The whole loop, from request to fix, takes four steps:
From refund request to fix
1Make asking easy
What happens: One email address or form, answered within a business day, asking for the order number and an optional reason.
Why it matters: A buyer who can't find how to ask can go to their bank instead.
What goes wrong:
- A refund policy that lives only on the terms page
- A required reason that reads like a hurdle
2Code the reason
What happens: Turn the buyer's words into one code from a short list, and note the order's bundle, source and VSL version.
Why it matters: Free text can't be counted. Codes can, against the orders they came from.
What goes wrong:
- Thirty codes that nobody applies the same way
- A reaction or side effect filed only as a refund reason, instead of passed to whoever handles adverse event reports
3Refund the same day
What happens: Pay the refund the day it is approved and confirm it by email, with the amount and the date.
Why it matters: A slow refund doesn't save the money. It gives the buyer time to dispute the charge.
What goes wrong:
- Arguing with buyers inside the guarantee window
- Asking for a return the policy never mentioned
4Fix the biggest code
What happens: Each week, count refunds by code and by cohort, pick the largest code and fix its cause.
Why it matters: The codes point at where the problem starts: the promise, delivery, billing or what buyers know after the order.
What goes wrong:
- Counting refunds without the orders they came from
- Fixing the loudest complaint instead of the biggest code
What causes refunds, and the fix for each cause
Each cause leaves its own signature in the codes and the timing, and each has a fix that starts before the refund request.
The promise is bigger than the product
Signal: "not as described" or "no results yet" in the first weeks, and higher rates on the most aggressive angles. The ad, the VSL or the page promised a result, a speed or a certainty the product can't deliver to most buyers.
Fix: claims you can substantiate, a realistic picture of what buyers will notice and when, a plain line on who the product isn't for, and the same claims everywhere: ad, VSL, page, label and emails. The VSL compliance guide covers the claim rules. The honesty that keeps you compliant also keeps expectations where the product can meet them.
Delivery is slow or silent
Signal: refund requests before the parcel arrives, "where is my order" emails, and "arrived late or not at all" codes.
Fix: state realistic shipping times before checkout, send tracking as soon as the parcel ships, and tell buyers about any delay before they ask. The FTC's Mail Order Rule sets the floor: you need a reasonable basis to ship within the time you state, or within 30 days if you state none, and when you can't, you must send a delay notice that lets the buyer cancel. If the buyer cancels, the refund for an ordinary card purchase is due within seven working days; one billing cycle applies only when you extended the credit yourself (FTC business guide). This isn't legal advice.
The charge is a surprise
Signal: "unexpected charge" or "ordered by mistake", and refunds that arrive right after a renewal or a post-purchase upsell.
Fix: state subscription terms before the buyer enters payment details, keep add-on boxes unticked, make "No thanks" on an upsell as easy to find as "Yes", and remind subscribers before each renewal. Stripe's dispute-prevention advice suggests reminders typically 7 days before a yearly renewal and 2 to 3 days before a monthly one (Stripe).
Buyers don't know how to use it
Signal: "no results yet" or "didn't know how to use it" late in the window, from buyers who stopped after the first week. A supplement taken the wrong way, or for a week instead of the period it's meant for, disappoints.
Fix: tell buyers how to use the product, what to expect and when, in the days after the order, then check in. The post-purchase email flow lays out those emails.
The wrong buyers, or the same buyer again
Signal: one source or angle refunds far more than the rest, or the same names keep coming back.
Fix: judge sources by cost per kept customer, not CPA. Rework angles that over-promise. Track refunds per customer, and if you limit the guarantee, to one refund per customer for example, state the limit next to the guarantee before anyone pays. Never quietly refuse a refund your page promised. The money-back guarantee post covers terms and limits.
Two more causes are easier to spot: damaged or wrong items, which point at packaging and picking, and orders placed by mistake, which point at unclear quantities or duplicate submissions at checkout.
What to do this week
Start with the measurement, then the biggest cause:
Refund rate: what to do this week
0 of 8 done
Frequently asked questions
What is a good refund rate for an online store?
There's no honest universal number. It depends on the product, the guarantee, the price and the traffic, and figures quoted without those details don't transfer. The useful comparison is your own rate against your own margin: work out how high refunds can go before an offer stops making money, then track your cohorts against that ceiling and against your own history.
Do refunds count as chargebacks?
No. A refund is money you return on your own terms; a chargeback is a payment reversed through the buyer's bank. Card network monitoring programs count disputes, and Stripe's guide says they don't consider refunds when identifying disputes. A refund issued after a buyer has already disputed doesn't take that dispute off your count, so refund quickly when a request arrives.
How long after a purchase do refunds come in?
For as long as your guarantee allows, plus the delivery time. With a 60-day guarantee counted from delivery, refunds on an order can arrive more than two months after it was placed. That's why a month's refund rate should be read only after its window has closed, and why a new offer's early refund rate looks lower than it will end up.
Can I refuse a refund if the bottle was opened?
Only if the terms you showed before the buyer paid said so. If your offer promised a refund regardless of use, honor it. If refunds depend on returning unopened bottles, that condition belongs next to the guarantee, where buyers see it. Quietly refusing a refund your page promised invites a dispute and a complaint, both of which cost more than the refund.
How fast do I have to refund a cancelled order?
If an order is cancelled under the FTC's Mail Order Rule, for example because you couldn't ship on time, the FTC's business guide says the refund for an ordinary card purchase is due within seven working days. One billing cycle applies only when you extended the credit yourself, such as a store account. Your own guarantee can promise faster, and a faster refund is cheaper than a dispute.