How to Evaluate Your Shop's Affiliate Programme

There are two ways to evaluate an affiliate programme. You can look into the network once a month, check how much you paid out in commissions and, if the number does not hurt, let the programme run. Or you can ask the less comfortable question: how many of those orders would have arrived without the publishers, which publishers actually bring you new people, and what one affiliate order really costs once you add the network fee. I run programmes for my own online shops and promote other people's programmes as a publisher, so I see both sides of the table. This article is about the second approach.

Metrics that actually tell you something

The network interface shows clicks, conversions, revenue and commission. Those are inputs, not an evaluation. From them you need to assemble a handful of numbers that say something about the programme.

Active publishers

The number of approved publishers in a programme means almost nothing. In most programmes I have seen from the inside, roughly a fifth of partners record at least one click a month and only a few individuals account for more than 80% of revenue. Define an active publisher yourself, say at least 20 clicks or one order in 30 days, and watch how that number develops month by month. If the number of approved publishers grows but the active count stays flat, recruitment is not working and you need to find out why publishers never deploy anything after signing up. Usually what is missing is creatives, a link to the product feed, or simply a reply to their first message.

Clicks and conversion rate

Affiliate conversion rates for content sites and comparison engines usually sit somewhere between 1 and 4%; coupon sites can easily reach 10% or more, because the customer already has a full basket and is only looking for a discount. That is why comparing conversion rate across all publishers with a single number makes no sense. Split them by type and look for anyone who stands out within their type. A review site converting at 15% is either exceptional or something is happening there that you want to look at more closely.

Average order value in affiliate versus other channels

Compare the average order value from affiliate with the average across the whole shop. If affiliate is noticeably lower, it usually means publishers are sending customers to cheap entry-level products or that most orders arrive through discount codes. If it is noticeably higher, the cause is often a single comparison engine sending people to a specific, more expensive category.

Share of new customers

In my view this is the metric that decides whether the programme makes money. Match affiliate orders against customer history in your own system (by e-mail or account) and count how many were first purchases. With a good content publisher the share of new customers tends to be above 60%; with cashback and coupons it can drop below 20%. Once you know this, you stop measuring every publisher with the same yardstick.

Approval and rejection rate

The network shows how many conversions you rejected. Track the rejected share separately for each publisher. For cancellations and returns a normal figure is around 5 to 15%, depending on your range. A publisher with 40% of orders rejected is either sending undecided people (typically from PPC on generic keywords) or producing orders that should never have gone through. Both are worth a conversation.

Cost per order including the network fee

The publisher's commission is not the whole cost. Networks usually charge 20 to 30% on top of the commission paid out, plus a monthly fee and, with some networks, a setup fee. If you pay out a 10% commission and the network takes 25% on top, you are paying 12.5% of revenue. Work out the real cost of one approved order and compare it with what you pay per order in PPC and in shopping comparison engines. Only that number tells you whether raising the commission makes sense, or whether you should be looking for a cheaper route to publishers.

To give you an idea, here is what such a monthly table might look like for a smaller fashion shop:

Attribution and why the network and your analytics disagree

Affiliate networks work almost exclusively with a last-click model inside a single channel. If a customer clicks a publisher's link and buys within thirty days, the network credits the conversion regardless of whether the customer came back via Google, your newsletter or by typing the address in by hand. Your analytics, on the other hand, sees every channel and may assign that same order to e-mail, because the last click before purchase came from the newsletter. Both numbers are correct under their own rules; they just answer different questions.

Overlap with PPC and e-mail

The most common overlap is with brand campaigns in PPC and with your own e-mailing. A customer reads a review on a publisher's site, clicks, looks at the product and leaves. In the evening they receive an abandoned-basket reminder, or they type the shop's name into Google and click your ad. The network assigns the order to the publisher, Google Ads claims it too and so does the e-mail tool. Add up the conversions from all the tools and the total can be 30 to 50% higher than the actual number of orders. If you do not know this, every channel looks like it performs brilliantly while you are paying three times for one order.

Deduplication

The fix is called deduplication and most networks support it, provided you set it up. The principle: you pass information about the source of the last visit into the conversion code, and the network credits the conversion only when that last source was affiliate. It requires a change on your site (storing the source in a cookie or session on arrival) and an agreement with publishers that no commission is due when there is an overlap with paid search or e-mail. Not every publisher will jump at it, but the serious ones know the practice and expect it. The opposite extreme, deduplicating against organic search and direct visits as well, hurts publishers needlessly. Someone who read about the product on their site and later remembered your address is still their customer.

Differences that remain even after deduplication

Some reasons never go away. Ad blockers and browser privacy protection knock out part of the tracking scripts, typically 10 to 25% of visits, and a different part for each tool. Analytics usually counts conversions by session, the network by cookie, so a purchase three days later on another device is never attributed to affiliate in analytics at all. A gap in the region of 10 to 20% is normal. If it is substantially higher, look for an implementation error or fraud.

Cookies in practice

Your programme terms say a 30-day cookie. In practice it survives that long only in some cases. Safari with ITP limits the lifetime of cookies set via JavaScript to 7 days, and with some redirects through a tracking domain to just 24 hours. Firefox and Brave block known tracking domains outright. On mobile, people often click inside apps (Instagram, Facebook, an e-mail client) that open the link in a built-in browser, where the cookie stays trapped and never reaches the regular browser. If a customer reads a review on their phone and buys on a laptop that evening, the publisher gets no commission whether the cookie is set to 30 or 90 days.

For your evaluation this means that some conversions simply cannot be attributed to publishers and nobody can do anything about it. When comparing two networks or two tracking systems, ask whether they can track server-side and whether they offer a first-party cookie on your domain. A longer cookie looks nice in the offer, but the difference between 30 and 60 days is smaller than it seems, because the vast majority of attributed conversions happen within 24 hours of the click.

Coupon sites and cashback

This is where affiliate managers argue the most. Coupon and cashback sites tend to have the highest revenue in a programme and at the same time the lowest share of new customers. The most common scenario: the customer has a full basket, sees the "discount code" field, opens a new window and types the shop's name plus the word discount. The coupon site is first in the results, the customer clicks, copies the code and completes the purchase. You pay a commission on an order that would have happened anyway, and you have given a discount on top.

At the same time it would be a mistake to write off coupons and cashback because of this. Cashback portals have a strong base of users who first check where they can get cashback and only then choose a shop. To those people you are invisible without a cashback listing. You cannot tell the difference between collecting commission and genuinely bringing in a customer from the publisher type; you can only tell from the data.

Signs of commission collecting

A few things you can check in a single afternoon:

• Time between click and order. If a publisher's median is under 5 minutes, the vast majority of people arrived with a finished basket.
• Share of new customers from your own data. Under 20% means you are mostly paying for your own regular customers.
• Landing page. If the overwhelming majority of clicks go to the homepage or straight to the basket rather than to a product, nobody is recommending a product.
• Search for your shop's name together with the words discount, coupon, code. You will see who is targeting your brand and whether they are offering codes that do not even exist.
• The code used. If a coupon site is using your generic newsletter code, the publisher has just repeated what you sent the customer yourself.

Rules for coupons and cashback

You do not have to ban coupons; you just need to give them rules. What has worked for me: coupon sites may publish only the codes you assign to them and nothing else, not even a "try code SAVE10" with no guarantee. Publishing generic newsletter codes is grounds for rejecting the commission. Set a lower rate for coupons and cashback than for content publishers, for example 4% instead of 8%, or pay commission on new customers only. Write the rules into the programme terms so the publisher sees them before signing up. If they are also in your listing on AffilPlanet, a publisher who found you in the catalogue reads them too, and you save yourself the arguments later.

Fraud and risks

Most publishers play fair. The few who do not can make a programme expensive within a couple of months and sour the relationship with everyone else. Four things I look at regularly.

Brand bidding

A publisher buys search ads on your shop's name, or on misspellings of it, and the ad leads through an affiliate link straight to you. A customer who was searching for you would have arrived anyway, but now you pay a commission and you have also pushed up the cost per click in your own brand campaign, because someone is competing against you in the auction. You spot it by regularly searching for your shop's name, ideally from a different IP address and in a private window, and reviewing the ads. In Google Ads the auction insights report shows who is appearing on your terms. In network data brand bidding looks like a publisher converting above 10%, with a short time to purchase and no content on their site. Ban it explicitly in your terms, including misspellings and variants with spaces.

Cookie stuffing

The publisher sets an affiliate cookie on visitors without them clicking anything. It is done with hidden iframes, redirects through an image, or a browser extension that quietly rewrites links. In the statistics you see a publisher with an enormous number of clicks and a zero or very low conversion rate, because the "clicks" were just page loads. Or the reverse: a publisher with no traffic who nevertheless converts steadily across your whole range, with no seasonal swings.

Fake orders

Orders to non-existent addresses, uncollected cash-on-delivery parcels, card payments that come back as disputed a week later. If you approve commissions before the goods actually ship and the customer pays for them, you are paying commission for thin air. Signs: a cluster of orders from one publisher in a short time, similar e-mail addresses, the same phone prefix, orders just above the threshold for a higher commission tier. The fix is simple: approve only after the returns period has passed and payment has cleared.

Self-purchases through the affiliate link

A publisher, or one of your own employees, buys through their own affiliate link. In a typical shop that amounts to a few percent of extra discount, which does not hurt you as long as it goes no further. It gets worse when a publisher sets up a group where they tell people how to buy through their link in exchange for a cut of the commission. That is no longer affiliate marketing but a hidden discount you never approved. State in your terms whether self-purchases are allowed, and keep an eye on publishers where the e-mail on the order matches the e-mail on their registration.

Approving commissions

Approval is the most sensitive point of the relationship for publishers. From my own experience as a publisher: a programme that approves within 30 days and explains its rejections gets more space on my sites than a programme with a higher commission that approves after three months with no comment.

The period should match your returns and payment terms, usually 14 to 45 days. You may need longer for goods with long delivery times or services with a trial period, and in that case explain it in the terms. Approve in a regular rhythm, for example every week whatever has passed the waiting period, rather than one big batch at the end of the quarter. Publishers see that something is happening in the programme, and you do not pay out a large sum all at once.

Communicate rejections. Networks let you state a reason for every rejected conversion, and you really should: customer cancellation, uncollected parcel, return, duplicate order, breach of terms. A publisher who sees "cancellation" on 8% of orders understands. A publisher who sees 30% rejected with no reason stops promoting you and tells other publishers about you. And if you reject for a breach of terms, write to the publisher directly and give them a chance to fix it. Sometimes it is a misunderstanding, such as an old code left on the site after a promotion ended.

Late approval has a direct effect on programme performance. Publishers plan their content around what earns them money, and if your commissions sit in "pending" for two months, they spend that time working for a competitor who approves in three weeks.

Regular reporting

The monthly review should take an hour, not a day. I recommend building one spreadsheet into which you paste the network export and the order export from your shop each month, with everything else calculated automatically. What belongs in it:

1. Overall figures for the month and the trend against the previous month and the same month last year: clicks, orders, revenue, commission, cost including the network fee, share of new customers.
2. Top publishers by approved revenue, with type, conversion rate, share of new customers and rejection rate for each. Five to ten names usually cover most of the revenue.
3. Newly active publishers, meaning whoever showed up with a click or an order for the first time this month. Write to them, say thank you and offer creatives or a dedicated code.
4. Publishers who stopped. Whoever had revenue last year and nothing this year. You will often find a banner dropped out, a product URL changed, or the publisher simply moved to a competitor with a better commission.
5. Anomalies: a sudden jump in clicks with no conversions, a publisher converting above 10%, clusters of rejected orders.

Inactive publishers are a chapter of their own. In most programmes they are more than 70% of the total. There is no need to throw them out, but once every six months it is worth writing to them: a short e-mail with the state of the programme, a new code or a bonus for the first order. A fraction reply, but those tend to be the cheapest performance gain you can get, because they are already in your programme. Anyone who has not replied after a year can be removed.

Changing the commission and the terms

The commission is worth changing when the economics change. Margins have fallen, the cost per affiliate order is higher than in PPC, or the programme is stagnating while competing shops offer more. Before you touch the rate across the board, try differentiation first: a lower rate for coupons and cashback, a higher one for content sites and new customers, a bonus on the first ten orders for a new publisher.

When a blanket change does come, announce it in advance. A commission cut at least 30 days ahead; an increase can take effect immediately. Spell out exactly what changes, from when and why, and what stays the same. Publishers can take a commission cut; what they cannot take is discovering it in their payout statement. The same applies to changes in terms, such as banning coupons or a new deduplication method.

On AffilPlanet this is handled by programme updates. When you edit the commission, terms, codes or offer in your listing, every publisher following the programme receives a notification and an e-mail automatically, without you having to mail a list by hand. You can also add an update manually for a seasonal promotion or a temporary bonus. For you that means a single place to record the change; for publishers it means certainty that they will hear about it before it shows up in their payout. What a listing looks like, and what it takes to claim a programme that is already in the catalogue, is described on the page for online shops.

A routine for the first quarter

If today you evaluate the programme only by total commission, do not try to introduce everything at once. A sensible order: first work out the real cost per order including the network fee and the share of new customers, because those two things decide whether the programme makes money at all. Then set up deduplication against paid search and e-mail so you stop paying three times for the same thing. In the third month go through publishers one by one, by type; you will uncover coupon sites that merely collect, and any brand bidding. And when the numbers tell you which kind of partner is missing from your programme, look through the publisher catalogue, where you can filter by type and topic. Usually the answer turns out to be content sites and reviews, not yet another coupon portal.