Affiliate programme launch kit: the payout day is computed from the refund window

---
name: affiliate-program-launch-kit
description: Produces an affiliate or partner programme from nothing: the commission structure chosen from the product's repurchase shape, a payout day computed from the refund window plus settlement and published as a worst case, the ten clauses that have to be written before anyone is recruited, a five-tier recruitment order that puts published advocates ahead of strangers, the two-stage partner ask, and a media kit of finished assets with a tracking link unique to each partner. This skill should be used when an affiliate, referral or partner programme is being set up, when a commission rate is about to be published, or when partners are about to be approached with an ask.
---

# Affiliate programme launch kit

## The claim this skill is built on

An affiliate programme is a set of financial terms with a marketing kit attached, and almost every team ships the marketing and leaves the terms until the first dispute.

The obvious approach is to pick a percentage that sounds generous, install a tracking plugin, put a signup form in the footer and email some people with audiences. It fails in three specific ways, and none of them is visible in the first month.

It pays commission on money that later reverses, because the payout schedule was inherited from the finance calendar rather than computed from the refund policy. It cannot attribute anything, because everybody was sent the same link. And it produces partners who said yes and never posted, because they were sent a brief instead of assets, and a brief transfers the work to the busiest person in the chain.

The decision that settles the economics is the payout day, and it is normally decided by whoever runs payments, on the grounds that thirty days is what everything else is. The decision that settles participation is what lands in the partner's inbox on the day they say yes. Both are made late, casually, by people who are not in the meeting where the rate is argued about.

Everything below is written in the order the decisions have to be made, because several of them are one-way doors.

## Part one. The commission structure comes from the repurchase shape

Do not start from what competitors pay. Start from how your product is bought a second time, because that is what decides which behaviour you are paying for.

**One-off purchase, no meaningful repeat.** A flat rate per sale. The only question is whether the rate survives the arithmetic in part seven. Simplicity is worth real money here, because every exception you write is a support conversation later.

**Repeat purchase where you want new customers rather than larger orders from existing ones.** Two tiers: a higher rate on a first-time buyer, a lower rate on a returning buyer. A useful shape is roughly three times the returning rate for a new customer, for example 30 per cent and 10 per cent. Without the split, a partner's cheapest route to commission is to point their existing audience at a purchase those people were going to make anyway, and you pay acquisition prices for order padding. With the split, the partner's incentive matches the reason you opened the programme.

**Subscription.** A recurring share of each payment, capped at a stated number of months or running for the life of the account. Recurring share buys you a partner who cares whether the customer stays, because a customer who churns in month two stops paying them too. A one-off bounty buys volume regardless of fit, and a partner paid a bounty has no reason to qualify anybody.

**Mixed or unclear.** A bounty on a qualified action you can price, such as a first paid invoice, rather than a percentage of a revenue stream you cannot yet forecast.

Two structural points that get missed. A recurring share is a permanent claim on your best cohort unless it is capped, and the cap is far cheaper than it looks because most accounts do not survive it, which part eight demonstrates with numbers. And whichever structure you choose, write the rate you can afford to raise. A published rate is close to a one-way door: raising it is an announcement, lowering it is a pay cut delivered to every partner on the same morning.

## Part two. The payout term rule, and the arithmetic behind it

**The rule: the payout term must be at least as long as the refund window plus a settlement buffer.** Not the same as. At least.

```
payout day = refund window (days) + settlement buffer (days)
settlement buffer = 7 to 14 days, covering the gap between a refund
                    being requested and the money leaving, plus your
                    own reconciliation
```

A sixty-day refund policy therefore produces a payout day of about 67. A fourteen-day policy produces about 21. A thirty-day policy produces about 37.

### What breaking it costs

Take a product at 100 units, commission at 30 per cent, a sixty-day refund policy, payouts at day 30, and 400 affiliate-sourced sales a month at a 12 per cent refund rate on that traffic.

```
monthly affiliate revenue      400 × 100        = 40,000
monthly commission             400 × 30         = 12,000
refunded sales                 400 × 0.12       = 48
refunded revenue               48 × 100         = 4,800
commission already paid on it  48 × 30          = 1,440 per month
                                                = 17,280 a year
```

That 17,280 is not a loss yet. It is a recovery problem, and recovery only works against a partner who is still earning. Affiliate revenue is normally concentrated in a handful of partners with a long tail of people who make a few sales and stop, so a realistic third of that figure has no future balance to deduct from. Call it about 5,800 a year that is simply gone, and every attempt to recover the rest is an unpleasant conversation with somebody who did nothing wrong.

Moving the payout to day 67 costs nothing at all, provided it was published before anybody joined. That is the whole point of the rule. It is free before recruitment and expensive afterwards, because changing a payout term later is indistinguishable from a term being taken away.

### The detail that turns the rule into a broken promise

Most programmes pay in monthly runs, not per sale. A run described as "the 15th of each month, for sales whose refund window closed before the 1st" adds up to another month on top of the computed day.

```
sale on 3 March, 60-day window closes 2 May
2 May is after the 1 May cut, so it falls into the June run
paid 15 June, which is 104 days after the sale
```

**Publish the worst case, not the rule.** A partner told 67 and paid on 104 has been told the terms changed, and they are not wrong. Write it as "paid on the 15th of each month, for sales whose sixty-day refund window closed before the 1st of that month, so between 74 and 105 days after the sale". Nobody argues with a number they were given in advance.

### Chargebacks are a different clock

Card scheme dispute windows are considerably longer than most refund policies. As of August 2026 the widely cited figure for most consumer dispute reasons is around 120 days from the transaction or from delivery, and it varies by scheme, by reason code and by category. Verify the current position with your own payment processor rather than taking that number from here.

You cannot hold every payout for four months without killing the programme, so the resolution is a split of responsibilities and it should be written down as one:

- **The payout term covers refunds.** It is long enough that ordinary refunds never produce a clawback at all.
- **The clawback rule covers the chargeback tail.** A chargeback reverses commission whenever it arrives, deducted from the next balance, and a partner whose chargeback rate exceeds a stated threshold is suspended pending review.
- **A reserve covers a genuinely high-risk category only.** Holding a percentage of every payout for ninety days is a real cost in partner goodwill, and most programmes do not need it.

Two more payout mechanics that produce complaints if left unstated: a minimum balance before a payment is issued, so a transfer fee never exceeds the commission it carries, and an explicit rule that balances below the minimum are still payable when a partner leaves or after twelve months. Silent forfeiture of small balances is one of the most common public complaints about affiliate programmes and it is entirely avoidable.

## Part three. The clauses, all of them, before you recruit anybody

**A clawback rule written after the first refund is a negotiation, not a rule.** The same is true of every line below. Each one exists because somebody discovered it during an argument.

1. **Attribution window and model.** How long the cookie lasts, whether it is first click or last click, and what happens when an affiliate link and one of your own paid ads both touch the same sale. Say which wins, in writing, before a partner asks.
2. **Commission basis.** What the percentage applies to: the invoice total, the amount net of sales tax, VAT or GST, the amount net of payment processing fees, or the amount net of all three. "Thirty per cent of the sale" is ambiguous, and on cross-border invoices tax can be a fifth of the number. State the basis with an example calculation.
3. **Self-referral.** Whether a partner may earn commission on their own purchase, their employer's purchase, or purchases by accounts they administer. The usual answer is no, and it has to be written or the first case sets the precedent.
4. **Coupon and discount stacking.** Whether an affiliate code combines with a site-wide sale, and whether commission is calculated before or after the discount. After, almost always, and say so.
5. **Refund, downgrade and chargeback treatment.** A refund reverses commission in full. A downgrade adjusts recurring commission from the next billing period rather than retrospectively. A chargeback reverses commission whenever it lands. All three, named separately, because they behave differently.
6. **Prohibited practices.** Bidding on your brand name in paid search, which is the single largest source of affiliate disputes and the one that quietly raises your own advertising costs. Also: cookie stuffing, distributing codes you never issued to coupon aggregators, unsolicited email, misrepresenting what the product does, and paid social advertising that impersonates your brand.
7. **Disclosure.** The partner must disclose the commercial relationship. Requirements vary by jurisdiction and by platform and have been tightened in several places, so state the obligation, supply the exact wording you want used, and check the current position rather than relying on any wording, including this file's, that has a date on it.
8. **Trademark and asset use.** What they may use, what they may not modify, that logos may not be recoloured or stretched, and that permission is withdrawable.
9. **Termination and pending balances.** On termination, accrued commission on sales that have cleared their refund window is paid. Forfeiture applies only to defined fraud, and the definition is in clause 6.
10. **Programme change notice.** How much notice before rates or terms change, with thirty days as the customary floor, and an explicit statement that balances already accrued are paid at the old rate.

Add to that the operational lines nobody thinks of as clauses: which jurisdictions the programme is not available in, who is responsible for the partner's own taxes, what identification or tax documentation is required before the first payout, and the named human a partner emails when something is wrong.

## Part four. Recruit in this order, and the order is not politeness

**Tier one: people who have already published praise in public.** Reviewers who left detailed reviews, customers who wrote a post about the product unprompted, anybody who answered a testimonial request with specifics rather than pleasantries. They have demonstrated two separate things, enthusiasm and the willingness to write, and the second is much rarer than the first.

**Tier two: power users identified from usage, invited in the product.** Invite at a success event rather than in a monthly email, because the moment somebody has just got a result is the moment they can imagine recommending it.

**Tier three: the always-on routes.** A signup link inside the product and an invitation inside the account-creation email sequence. This is the part that keeps recruiting when nobody is working on it, and it is normally implemented as a footer link that only existing partners ever find.

**Tier four: adjacent professionals.** Consultants, agencies, implementers, trainers, course sellers, community moderators. Their audience already pays them for advice in your category, so a recommendation costs them nothing and earns them something.

**Tier five: publishers, comparison sites, deal sites and cold outreach to strangers.** Last.

Three reasons the order changes the outcome rather than merely being courteous:

- **The second stage of the partner ask needs names.** Social proof is what stage two attaches, and if you recruited strangers first there is nobody to name.
- **The first version of your kit is wrong.** Discovering that with an advocate costs one email. Discovering it with a publisher costs the relationship, and publishers talk to each other.
- **Early data from advocates tells you the programme's real conversion rate.** Once deal and coupon traffic arrives, click-to-customer rates blend into an average that cannot be interpreted, and you have lost the baseline you needed in order to judge tier five at all.

Expect concentration. Affiliate revenue is heavily skewed almost everywhere: a small number of partners produce most of it, and the rest produce a handful of sales each. Treat the top few as accounts with a named human relationship, and treat the tail as a self-service programme that must not require your attention.

## Part five. The two-stage partner ask

The general one-to-one outreach ladder, the research gate and the reply branches belong to a dedicated outreach method and are not restated here. What follows is the part specific to a partner ask.

**Stage one is a small ask, and it is deliberately not the thing you want.**

Five sentences or fewer. Who you are in one line. Who your audience is, with a number attached. What the programme pays, as a number and a term, never as the word generous. One sentence on why their audience specifically would care. Then a single question answerable yes or no in one breath: may I send the one-page terms, or do you want to be on the partner list for the launch.

Never ask for a list send, free product, a webinar or a review in the first message. An opening email that asks a stranger for their audience is asking for budget, and the answer to that is silence rather than no, which is worse, because silence cannot be distinguished from a message that never arrived.

**Stage two runs only after a yes, and carries the three things stage one withheld.**

- **Social proof, by name.** Who else has joined. This is why the recruitment order exists.
- **The mechanics.** Rate, term, attribution window, payout schedule stated as the worst case, what the kit contains, and the dates.
- **The real ask.** The specific thing: post on these two dates, a section in the newsletter of the 14th, a comparison page, a webinar, a review.

The small yes changes what the larger ask is. It is now a request to a participant about something they already agreed to be part of, rather than a request from a stranger for something valuable. That is a different message with a different reply rate, and it cannot be manufactured by writing a better cold email.

**Addressing.** Send to the person who decides and copy the person who would execute, never the reverse, because two people who can each assume the other owns the reply produce no reply at all. The fuller argument for that convention sits in the giveaway campaign method and is worth reading once.

**The tripwire.** If a stage-two message goes to somebody who never answered stage one, you have not run this method. You have run a cold big ask with extra paragraphs, and it will perform like one.

Keep one record per partner: organisation, decision maker, executor, date of first message, date of yes, the specific commitment, the tracking link, what they actually did, and what it produced. That record is your next campaign's shortlist and it cannot be reconstructed afterwards.

## Part six. The kit is finished assets, not a brief

**Anything a partner has to write is a thing that does not go out.** A brief looks generous and is a work transfer to the busiest person in the chain, who is usually a marketing manager with four other launches that week.

The kit is a single page or folder, versioned and dated, containing:

- **Their tracking link, at the top, before anything else**, plus a way to append their identifier to any page on your site. Deep links convert better than the homepage, and the homepage is where attribution goes to die.
- **Post copy for each platform, two variants each**, written to the platform's cut rather than its ceiling. For a short-post network, under 280 characters including the link. For a professional network, the first roughly 140 characters carry the whole offer, because the feed collapses everything after that behind a see-more control. Two variants so that two partners posting on the same morning do not post identical text.
- **Two email swipes**, an announcement and a reminder, each under 150 words with one link. Assume a narrow mobile client shows roughly the first 40 characters of a subject line, so the product noun goes first.
- **A newsletter paragraph** of 60 to 90 words that can be dropped into an existing issue without editing, which is the format most partners actually have room for.
- **A spoken version of 30 to 45 seconds**, for anyone with a podcast or a video channel, written to be read aloud rather than to be read.
- **Images at the sizes actually used:** 1200 by 630 for a link preview card, 1080 by 1080 square, 1080 by 1920 vertical, and 1600 by 900 for a blog header, exported as PNG, each supplied with and without a price or discount overlaid, because some partners are not permitted to publish a monetary figure.
- **One plain text file containing every piece of copy**, so a partner on Windows or on a Mac can use the whole kit without a design tool or a specific font installed.
- **A one-page facts sheet:** what the product does in one sentence, three claims you can substantiate, the three claims they must not make, the price, the discount code and its expiry date if there is one, the exact disclosure line, and a named person to email.
- **Dates**, including one date when everybody posts together if there is a launch moment.

**Version the kit and put the version number on it.** Partners keep assets for years, and an expired code in an evergreen post produces support tickets long after the campaign is forgotten.

### The tracking link requirement is more specific than it sounds

One link per partner, and where practical one per placement, so that newsletter, social post and comparison page are separable. The ratio of clicks to customers per placement is the number that tells you what to ask for next time, and it is unrecoverable if everything shares a link.

The partner identifier has to be carried through checkout and stored on the resulting account, not held only in a browser cookie. Client-side cookie lifetimes have been shortened repeatedly by browser vendors since 2020 and the position continues to change, so treat the cookie as a convenience and the identifier stored against the account as the record. Offer a coupon code as a second attribution path as well, because a customer who reads on a phone and buys on a laptop is invisible to cookie attribution and entirely visible to a code.

Never send a partner's audience to your homepage.

## Part seven. The decision rule

The question is not what rate is normal. It is whether you can afford this structure at this payout term.

```
contribution per referred customer
  = price × gross margin × (1 - refund rate) - commission
```

- **Contribution is comfortably positive and the payout day clears the refund window plus settlement.** Launch. Publish the terms, open with tier one, and hold tier five until the kit has been tested on people who will forgive it.
- **Contribution is positive but thin**, meaning under roughly a fifth of the price for a one-off product, or a payback beyond twelve months for a subscription. Launch only with a capped recurring share or a fixed bounty. An uncapped lifetime share on a thin margin is a permanent claim on your best cohort, and it is the term nobody notices until the programme is three years old.
- **Contribution is negative.** Do not launch a percentage programme. Raise the price, cut the rate, or pay a fixed bounty on a qualified action you can afford. A programme that buys revenue at a loss will look like growth for two quarters.
- **You cannot tell, because the product is new and you have no refund rate and no cohort data.** This is the common case and it has its own procedure. Do not publish a public rate. Run a closed pilot with five to ten partners under a written agreement that names an end date and states plainly that the rate is a pilot rate. Collect one full refund window plus thirty days of data, then set the public terms. The reason is the one-way door: a published rate cannot be lowered without every partner experiencing a cut on the same day, whereas a pilot that ends on a stated date simply ends, and the partners in it are the first people you invite into the real programme.

## Part eight. A worked example, compressed

A time-tracking product for small building contractors. Sold as a subscription at 24 units per seat per month, average account six seats, so 144 units a month. Sixty-day refund policy. Gross margin 82 per cent. Measured self-serve churn 6 per cent a month.

**The proposal on the table:** 30 per cent recurring for the life of the account, payouts net 30, one shared link on a partners page, recruit by emailing twelve industry newsletters.

**Run the arithmetic.**

```
commission per account         144 × 0.30                = 43.20 a month
mean account life              1 / 0.06                  = 16.7 months
lifetime revenue               144 × 16.7                = 2,405
lifetime gross profit          2,405 × 0.82              = 1,972
lifetime commission, uncapped  43.20 × 16.7              = 721
contribution after commission  1,972 - 721               = 1,251
```

Positive, so a recurring share is affordable in principle. Three things still fail.

**The payout term.** Net 30 against a sixty-day refund policy means every account refunded between day 31 and day 60 has already had one commission payment made on it, sometimes two. At a planned 120 referred accounts in the first quarter and a 14 per cent refund rate on partner traffic, that is about 17 refunded accounts carrying an average of roughly 1.4 payments each, so around 1,030 units of commission paid on revenue that was returned, in one quarter, with a meaningful share of it unrecoverable because the partners who produced it have stopped selling.

**The uncapped share.** Cap the recurring commission at twelve months and the expected cost falls by more than the cap suggests, because most accounts do not reach it.

```
expected months paid under a 12-month cap
  = (1 - 0.94^12) / 0.06 ≈ 8.7 months
capped lifetime commission     43.20 × 8.7               = 376
saving per referred account    721 - 376                 = 345
across 120 referred accounts                             ≈ 41,000
```

The partner is still paid across the entire period in which they would notice, and the programme stops carrying a permanent claim on accounts it acquired three years ago.

**The shared link.** With one link, the twelve newsletters cannot be told apart, so nobody can be paid correctly, nobody can be reported back to, and the second campaign re-approaches all twelve with the same message regardless of which one produced everything.

**The revised programme.** Commission 30 per cent recurring, capped at twelve months. Payout published as "monthly on the 15th, for sales whose sixty-day refund window closed before the 1st, so 74 to 105 days after the sale". The ten clauses written and linked before any recruitment. Recruitment opens with 40 customers who left detailed public reviews and 6 who posted about the product unprompted, then an in-product invitation at the success event, then the twelve newsletters, then a trade directory. Every partner gets their own link, plus a second link for a newsletter placement. The kit ships as finished assets, versioned, with a plain text file included.

**The two-stage ask in action.** The largest newsletter is approached with five sentences to the owner, copying the editor, ending with "may I send the one-page terms". Four of the twelve say yes. Stage two goes to those four with three customer advocates already live and nameable, the full mechanics, and one specific ask: a placement in the issue of the 14th.

**Verdict: launch, with the cap and the published worst-case payout day, and do not send stage two to the eight who never replied.** The commission rate never changed. What changed the economics was the payout day, worth about 1,000 units a quarter in commission not paid on refunded revenue, and the twelve-month cap, worth about 41,000 units over the first quarter's referred accounts. Neither of those would have come up in a meeting about the rate, which is the entire argument for doing the arithmetic before publishing anything.

## Part nine. After the campaign

**Report back to every partner within three working days, with their own numbers:** clicks, customers, revenue, their commission, their share of the programme total, and one sentence on which placement performed best. Almost nobody sends this, which is precisely why it works, and it is the message that produces the yes for the next campaign.

**Never hand over the contact data.** Their return is exposure and commission, and that was the deal. Beyond the deal, passing customer data to a third party not named at the point of collection is a plain breach in consent-based jurisdictions, and the complaints come from the customers you just acquired.

**Reconcile quarterly and sunset the dormant.** Deactivate links with twelve months of zero clicks so the reporting stays readable, honour any accrued balance regardless, and tell the partner before you do it rather than after.

**Close with the invitation to the next one**, and put the date in it. A partner who has been paid, reported to and invited again is a partner who does not need recruiting twice.

## Failure modes

**Paying inside the refund window.** Commission goes out on day 30 for sales refunded on day 45. From the outside it appears as a reconciliation gap three months in, and as an awkward conversation with your best partner, who did nothing wrong and will remember it.

**The cold big ask.** The first email requests a list send. Replies are near zero and silent rather than negative, so the team cannot distinguish rejection from non-delivery and spends a fortnight chasing people who were never going to answer.

**One shared link.** Everybody promoted, revenue arrived, and nobody can be attributed. The consequence appears in the second campaign, where the same effort goes to the partner who produced four customers as to the one who produced two hundred.

**The kit that is a brief.** Partners accept enthusiastically and most of them never post. It reads as a partner quality problem and it is a work-transfer problem, diagnosable by the fact that the ones who did post are the ones with an in-house designer.

**Strangers before advocates.** The programme opens with deal sites and coupon aggregators. The blended conversion rate lands somewhere uninterpretable, brand-bidding disputes start in week one, and the programme is judged on a number that never described your real audience.

**No written clawback rule.** The first refund becomes a negotiation, whatever you concede becomes the precedent, and the second partner has heard about the first.

**Commission on gross including tax.** On cross-border invoices you pay a percentage of tax you never received. It surfaces at the year-end reconciliation as a number nobody can explain, and correcting it retrospectively means telling partners their historic earnings were wrong.

**The rate that cannot be lowered.** Forty per cent published in month one, unaffordable by month fourteen. The reduction reads as a pay cut delivered to everyone simultaneously, the announcement costs more goodwill than the rate ever bought, and the best partners are the ones with somewhere else to go.

**Brand bidding.** A partner buys advertisements on your own name. Your paid search costs rise, you pay commission on customers who typed your brand into a search box, and it is invisible unless somebody looks at paid results for your own name from a browser that is not signed in.

**The expired code in an evergreen post.** A partner's article ranks for years carrying a code that stopped working in month three. Support tickets arrive indefinitely from people who feel misled, and there is no version number on the asset to point at.

**Silent forfeiture of small balances.** Dozens of partners hold balances below the payout minimum and never reach it. It becomes a public complaint, and the amount involved is always smaller than the reputational cost of having kept it.

## What this skill does not do

- It does not decide the legal or tax position. Disclosure and endorsement rules, contract enforceability, withholding and reporting on payouts, and consumer-protection rules on incentivised recommendation vary by jurisdiction and change. This names the questions and a qualified adviser answers them, before terms are published rather than after.
- It does not implement tracking, attribution or payouts. An affiliate platform handles per-partner links, server-side attribution, refund reversal, fraud screening, tax documentation and payouts in several currencies, and rebuilding those costs more than the programme earns in its first year.
- It cannot supply your refund rate, churn rate or gross margin. Every number in part seven comes from your own data, and a contribution figure built from estimates is a guess with a decimal point on it.
- It cannot judge a partner's audience. One placement with a tracked link answers that question in a week, and no analysis answers it at all.
- It does not detect fraud. Cookie stuffing, brand bidding, self-referral rings and unauthorised coupon distribution need continuous monitoring and a monthly human review, and a clause only tells you what to do once you have found one.
- It does not decide whether affiliates are the right channel. A programme applied to a product with poor retention pays partners to fill a leaking bucket faster, and the commission is spent whether or not the customer stays.