Affiliate platforms strategy: six steps in order

Most partner programs are launched in the wrong order. The platform is bought, partners are recruited, rules are written when the first argument happens, and measurement is added when somebody asks whether any of it worked. Reversing that order costs nothing and changes the outcome more than the choice of product does. An affiliate platforms strategy is mostly that: the order in which the decisions get made. What follows is a sequence, with the reason each step comes where it does.

What to take away

Step one: name the purpose, in one sentence

Programs get built for three different jobs and the jobs are usually confused.

Reach into audiences you cannot buy directly. Here the program is a sourcing exercise, and the platform matters less than your recruitment, which looks a lot like the work described in the guide to creator marketplaces.

Payment for performance from partners you already have. Here the program is a payment rail with reporting attached, priced on a cost per action basis, and the tracking chain is nearly the whole product.

Incremental revenue from a channel you do not yet have. Here the honest requirement is a measurement plan, because otherwise you will not be able to tell whether it worked.

Write the sentence and keep it. It is the pass condition for every later decision, and it is what stops the program becoming whatever the platform is best at.

Step two: settle the rules before the first partner

The first enforcement sets the standard for the year, which means the rules have to predate the partners.

Written as rules rather than intentions, the five are: attribution (which click counts, over which window, across which devices), deduplication (which channel is credited when two report the same sale), brand bidding (which branded terms partners may and may not bid on), coupon and extension attribution (who gets credit when a code, a cashback site or a browser extension closes the sale), and disclosure (the Federal Trade Commission's Endorsement Guides, 16 CFR Part 255, require partners to disclose the paid relationship).

Two of these need somebody outside the program to agree, which is why they take a week and get skipped. The brand-bidding rule needs whoever owns paid search to accept which branded terms partners may bid on and which stay off-limits. The deduplication rule needs the owner of your other channels to agree which channel is credited when two of them report the same sale. Doing them first is the whole of the strategy. The alternative is to discover in month three that two teams have been reporting the same sale, which is not a technical failure and never gets resolved technically.

Step three: instrument every path, then break it on purpose

Place a real order through every route a customer can use, including the ones added since the last time anyone looked. On a network like CJ Affiliate, Awin, Rakuten Advertising or ShareASale that means a publisher's deep link, a coupon page, a cashback click and a browser extension; on a management platform like Impact, Partnerize, Refersion, Tapfiliate or PartnerStack it means every tracking template you handed a partner. Then try to break the chain: buy across a device change, buy after a long gap, buy through a route where the identifier is unlikely to survive. What you are looking for is not perfection. It is knowing exactly where the silence is, so that a quiet partner type can be diagnosed rather than dropped.

Record the result as a dated list. In eighteen months, when a report looks wrong, you will want that list. It is the same artifact recommended in the guide to what a partner program is made of.

Step four: the first quarter is a calibration exercise

The fourth row prevents more damage than the rest combined. Nobody objects to a routing figure described as a routing figure, and the trouble starts when a number has been read as growth for two quarters. What a model can and cannot establish here is the same caution set out in the guide to measurement tools, and it applies before the first report rather than after it.

Step five: add partner types deliberately

Turn on one type at a time and give each a month of its own. Content partners, review sites, cashback, coupon aggregators, extensions, email partners and creator partners all behave differently and interact with each other in ways that are impossible to untangle if they arrive together.

The order is yours to choose, and the choice should follow from the sentence in step one. A program built for reach starts with content and creator partners, and the sourcing looks much more like commissioning than like a network integration. A program built as a payment rail starts with the partners you already have and adds nobody until the payments are boring.

Creator partners deserve their own month, because this is where an affiliate program meets influencer work. A creator who earns a commission through an affiliate link is also making an endorsement, and the FTC's guidance on its Endorsement Guides says a creator in that position should disclose the affiliate relationship both in the video and in the description near the links. An agency that runs the client's influencer campaigns and its affiliate program should hold both on one creator record: the flat fee, the commission link and the disclosure check, so the same creator is not briefed twice under two different rules.

Step six: build the review before you need it

An annual review that has to be invented under pressure produces a defense of the status quo. Decide now what would make you change something.

Name the condition that would make you leave the platform: its reports cannot be reconciled with your own order data, or it changes the attribution rule mid-term without telling you. Name the condition that would make you close the program: a holdout shows the channel is not incremental over the partners you already had. Name the partner behavior that would end a relationship: bidding on your branded terms against the written rule, or ignoring the disclosure requirement. Then write down the one number you would look at to know whether any of those had happened, and where it comes from.

A decision recorded as a claim that could turn out wrong is reviewable a year later by somebody who was not in the room, which is the point made at more length in the guide to comparing suppliers.

What this sequence deliberately leaves out

There is no recruitment volume target here, and no schedule for scaling. Both depend on the category, the margin and the partners available, and any figure printed on a page like this would be copied into a plan by somebody who has not tested it against their own economics.

There is also no advice on choosing between products, because at this stage it barely matters. A program with the rules written, the paths tested and the reporting labeled honestly will work on a modest platform. A program without those will disappoint on any of them, and the second year will be spent arguing about the software.

Common questions

How long before a program is worth judging?

At least one full validation cycle after the partner mix stopped changing. Judging it while you are still adding types tells you about the additions rather than the program.

Should we start with a network or go direct?

Direct is cheaper and slower and works when you have few, large partners. A network such as CJ Affiliate, Awin, Rakuten Advertising or ShareASale buys reach and adds an override on top of the commission you pay, plus somebody else's standard terms: the network's publisher agreement, its tracking and attribution standard, and its payment cycle. The sentence from step one usually answers this.

Who should own the program internally?

Someone who can enforce a rule, which usually means someone who can afford to lose a partner. When an agency runs the program on a client's behalf, the client signs the rules and the exit conditions, because an agency that cannot say no produces rules that exist only on paper.

Can we skip the holdout if the client is happy?

You can, and you are then carrying an unexamined claim on their behalf. At minimum, write down that no test has been run, so the next person knows what the figure is and is not.