Affiliate programs, attribution and incrementality

Every brand eventually asks the same question about its creator program: what did it actually cause? Tracking links and discount codes give a partial answer, counting sales that can be traced directly to a creator. But customers often see a creator post, think about it for a week, search for the brand and buy on another device, leaving no trace. Some tracked sales would have happened anyway. This lesson covers affiliate programs and tracking, attribution models that share credit across touchpoints, and incrementality methods such as holdout tests, lift studies and marketing mix models that estimate what the campaign truly added. It ends with the privacy rules that govern tracking.

Affiliate programs

An affiliate program pays creators a commission on sales or sign-ups they generate, tracked through unique links or codes. Affiliate programs can run on their own, as an always-on channel open to many creators, or as part of hybrid deals with selected creators. They are attractive because the brand pays only for tracked results, and they reward creators who genuinely drive sales.

Key design choices include the commission rate, often a percentage of sale value or a fixed amount per sign-up; whether commission applies to new customers only or to all purchases; the cookie window, the period after a click during which a purchase is credited, often 7 to 30 days; how returns and cancellations are handled; and how quickly commissions are paid. Brands should also set rules on where creators may promote links. Allowing creators to bid on the brand's own name in search ads, for example, lets affiliates claim credit for sales that would have happened anyway.

Discount codes are popular because they work across devices and in podcasts and videos where links are awkward. Their weakness is leakage: codes get posted on coupon websites and used by people who never saw the creator, so the creator is credited and the brand gives an unnecessary discount. Single-use or creator-specific codes with short expiry dates reduce leakage, and reviewing which codes show unusual spikes helps catch it.

Tracking links and UTMs

Links used in creator campaigns should carry tracking parameters so that the brand's web analytics can identify the source. The most common system uses UTM parameters added to the end of a URL, such as utm_source for the platform, utm_medium for the channel type, utm_campaign for the campaign name and utm_content to identify the individual creator or post. Consistent naming conventions matter: if one team writes "TikTok", another "tiktok" and a third "tt", reports will split the same traffic three ways. Short links make UTM-tagged URLs easier for creators to share and audiences to type.

Many brands also use affiliate or influencer platforms that generate unique links per creator and handle commission tracking, and some platforms offer native shopping features where purchases happen inside the app and are attributed automatically. Each system counts slightly differently, so expect totals from the platform, the affiliate tool and web analytics to disagree. Choose one as the source of truth for each KPI and explain the differences in reports.

Attribution models

Customers usually encounter a brand several times before buying: perhaps a creator video, then a search ad, then an email. Attribution models decide how to share credit for the sale among those touchpoints. Each model embodies a different assumption about what matters.

Attribution models only reshuffle credit among touchpoints that were recorded. They miss impressions that left no click, they cannot tell whether a sale would have happened without any marketing, and they are increasingly limited by privacy restrictions on cross-site tracking. For creator marketing, where much of the effect comes from people who watched but did not click, attribution tends to undervalue the work. That is why serious programs add incrementality testing.

Incrementality and holdout tests

Incrementality asks what happened because of the campaign that would not have happened otherwise. The gold standard is a controlled experiment: split comparable people or places into a test group exposed to the campaign and a control group that is not, then compare outcomes. The difference, scaled to the test group's size, estimates the incremental effect.

For creator campaigns, individual-level control groups are hard because brands cannot stop specific people seeing organic posts. Practical options include geographic tests, where creator content is amplified or creators are recruited in some regions but not in matched others; platform lift studies, where the platform randomly withholds paid creator ads from a control group and surveys or tracks both groups; and time-based tests, which compare periods with and without activity but are weaker because other factors change over time.

How big does a test need to be?

Small tests produce noisy results. A rough rule of thumb for comparing two conversion rates is that each group needs about n people, where p is the baseline conversion rate and d is the smallest absolute difference worth detecting. This approximation corresponds to about 80 % power at a 5 % significance level; statisticians use more precise formulas, but the rule shows why small lifts need large samples.

Brand lift and search lift

For awareness and consideration objectives, brand-lift studies survey people who were exposed to a campaign and a control group who were not, asking about awareness, ad recall, consideration and purchase intent. The difference between the groups estimates the lift. Several platforms offer brand-lift studies for paid campaigns, including paid creator content, often with minimum spend requirements. Search lift compares branded search volumes between exposed and unexposed groups or regions. Both measures capture effects that clicks miss.

Marketing mix modelling

Marketing mix modelling (MMM) uses statistical analysis of historical data, typically two or more years of weekly sales, spending by channel and outside factors such as seasonality, prices and competitor activity, to estimate how much each channel contributes to sales. MMM has regained popularity because it does not rely on tracking individuals and so is less affected by privacy restrictions. Open-source MMM tools have made it more accessible, but it still needs enough data and variation in spending to separate each channel's effect. For creator programs, MMM works best when creator spending is large enough and varies enough over time to be visible in the data. Many brands combine MMM for the big picture with experiments to calibrate it.

Privacy and tracking

Tracking customers across websites and apps involves personal information, and privacy laws limit what brands may do. Browser and device changes have also reduced third-party cookies and cross-app tracking, pushing brands towards first-party data, platform-reported results and aggregate methods such as MMM and experiments.

In Canada, PIPEDA requires meaningful consent for collecting, using and disclosing personal information, and the Office of the Privacy Commissioner has said that online behavioural advertising generally requires at least opt-out consent with clear notice, and express consent for sensitive information. Quebec's Law 25 goes further, requiring that technologies that identify, locate or profile people be turned off by default unless the person activates them, and requiring privacy impact assessments in some cases. Federal privacy reform has been proposed several times but, as of late 2026, PIPEDA remains in force; check for changes.

Practical steps include using consent tools on websites, sharing only aggregated or hashed data with platforms under proper agreements, avoiding sensitive categories such as health conditions in targeting without the required consent, and telling customers in privacy notices how campaign tracking works.

Choosing a measurement mix

No single method answers every question, so mature programs combine several. Tracked links and codes give fast, creator-level feedback for day-to-day optimization. Attribution models show how creators interact with other channels. Holdout and lift tests, run a few times a year, measure what the program truly adds and calibrate the faster metrics. Marketing mix models put creator spending in context with all other marketing over longer periods. The practical skill is knowing which tool answers which question and not asking a tracking link to prove incrementality or an MMM to judge an individual post.