Affiliate tracking should answer a decision, not construct a dossier on every reader. You may need to know whether a link works, which placement attracted clicks, whether a network credited a purchase, and whether an approved commission was actually paid. None of those numbers, on its own, proves that a reader was satisfied or that your recommendation caused a sale.

The practical approach is to collect the minimum information needed, respect applicable consent choices, and compare reports only when their definitions and time windows align. A network’s attribution rules determine credited actions; your own analytics describes activity it can observe on your site. Neither is a complete picture of all buyers.

Start here: Pick one decision and one placement. Record the relevant program terms, test the link and consent behavior, and report clicks, attributed actions and payment statuses separately. If a figure or term is unavailable, write not disclosed or unknown, rather than filling the gap with an assumption.

Every number in the examples below is hypothetical. No example represents a real campaign, a typical conversion rate, a promised earning, or a legal conclusion.

Map the journey before choosing a metric

A typical path runs from a reader seeing a recommendation, to selecting a link, to a network or merchant recording an action, to a possible commission. These stages are not interchangeable:

  1. Placement impression: A defined opportunity to see the recommendation, such as a tracked view of the section containing it. A page view is not automatically proof that the link was visible.
  2. Outbound-click event: An action recorded by your site or measurement tool when the affiliate link is activated. Repeat clicks can be separate events; blocked or unmeasured events may be missing.
  3. Network-recorded click: A click the network records under its own rules. Find out whether the number is raw, filtered or deduplicated. A network’s “unique” definition does not necessarily mean a distinct human being.
  4. Attributed action: A lead, trial, purchase or other event credited under the program’s rules. A signup is not the same event as a paid purchase.
  5. Commission status: A possible earning can be pending, approved, reversed, scheduled or paid. Labels and transitions differ by network; approval is not the same as money received.

Document the source and definition of every number. For example, PartnerStack describes separate partnership-performance, link-performance and commission reports, with different grouping options and fields. Its link report follows downstream activity associated with clicks in a selected timeframe; the partnership report can group events by their measure date. Do not assume every dashboard uses either approach. 1

A click can occur without a purchase, and a purchase can occur without being credited to you. Tracking can be limited by consent choices, device changes, browser settings, link errors, attribution rules, filtering, and reporting delays. These are possible explanations for a discrepancy, not proof of any particular cause. Do not attempt to defeat a reader’s privacy settings to close a reporting gap.

Choose the smallest measurement that can answer the question

Before adding a tag, ask what decision the data will change. Then choose the least intrusive reasonable approach and check the rules for your actual implementation.

  • Is the link working? Manually test the published destination and approved parameters. A test does not need a reader profile. Do not submit real or synthetic purchases unless the program explicitly permits the test method.
  • Is a placement attracting attention? If an appropriately configured analytics setup is permitted, compare an aggregate placement impression and outbound-click event. Define visibility and repeat events first.
  • Is the program crediting actions? Use the network’s aggregate link or campaign report if available. Check the applicable attribution rules before comparing it with onsite activity.
  • Has the commission been paid? Reconcile the network’s status and payment record with your own accounting records. A click counter cannot answer this.
  • Would people buy without the recommendation? Ordinary affiliate dashboards do not answer this counterfactual. A credited sale is not proof of incremental impact.

If an aggregate report resolves the question, do not export customer names or create an individual-level tracking scheme for convenience. If the answer requires data you cannot obtain responsibly, record the limitation and choose a decision that can be supported. A less complete but respectful report is better than one built by ignoring user choices.

A disclosure about commissions and permission to use tracking technology are different issues. A clear affiliate disclosure explains the commercial relationship; it does not itself obtain consent for cookies, pixels or other measurement. Conversely, a consent banner is not a substitute for an appropriate affiliate disclosure.

Requirements vary with the technology, purposes, parties and jurisdictions involved. Get advice for your situation; this article does not determine your legal obligations. For a concrete UK example, the Information Commissioner’s Office (ICO) says the Privacy and Electronic Communications Regulations can apply when technologies store information on or access information from a device. It illustrates an affiliate transaction pixel that reads stored referral information. Its guidance says online advertising uses covered by those rules, including ad affiliation and ad measurement, require consent and cannot use the listed exceptions. 2 3

The same ICO guidance describes a narrow statistical purposes exception for qualifying service-improvement analytics. It includes conditions on purpose, aggregation, notice and an objection mechanism; it is not a blanket exemption for affiliate ad measurement. Calling a tag “analytics,” “first-party,” or “server-side” does not, by itself, settle the question. A static URL label, a cookie, a pixel and a server request can involve different operations, so assess the actual data flow rather than treating them as identical. 2 3

Turn the policy into a testable flow

  1. Inventory the journey. List your analytics tags, outbound redirect, affiliate network, merchant destination and any server-side forwarding. For each, record what is collected, its purpose, who receives it, and whether it stores or accesses information on the device. Ask the network or merchant when their behavior is not publicly disclosed.
  2. Determine responsibilities. Identify which controls you operate and which are controlled by the network or merchant. Review applicable law, platform rules and contracts with qualified advice where needed; do not assume a merchant’s banner covers your site or vice versa.
  3. Define available choices. Where consent is required, configure your tools so the relevant tracking does not run until the applicable choice permits it. Make rejection and changes of choice usable; test that your actual implementation behaves as described. Do not treat silence as an affirmative choice.
  4. Plan the decline path. If a tracking-dependent affiliate journey cannot operate without the required consent, consider a clearly identified, ordinary non-affiliate destination without your tracking parameters, if the destination and program permit it. Verify what happens after the click: the destination may have its own technologies and responsibilities. Never imply that this alternative makes the whole journey tracking-free.
  5. Test changes and withdrawal. In a clean browser session, check the initial state, accept, reject and change-of-choice paths. Inspect requests, cookies and local storage for your own tools. Repeat after a tag or consent-platform change. Record date, setup, outcome and any open issue without storing a reader’s browsing history.

Google says its Consent Mode communicates consent choices to Google tags but does not provide a consent banner. In Google’s basic implementation, tags are blocked until the user grants consent; in its advanced implementation, tags can send cookieless pings while consent is denied. Those are different data flows, and the mode name does not establish that either setup satisfies a particular law or policy. Check your configuration and the current documentation. 4

For commission disclosures, the FTC says the relationship should be clear and conspicuous and that the closer the disclosure is to the recommendation, the better. A plain example, when accurate, is “We may earn a commission if you buy through this link.” The FTC cautions that “affiliate link” alone may not explain the relationship to consumers. Adapt the wording and placement to the medium; a footer alone may not be seen alongside a recommendation. 5

Separate incoming UTMs from affiliate sub-IDs

Campaign labels solve different problems depending on which site receives them. Mixing them is a common source of false attribution.

Incoming UTMs are parameters on a link that sends traffic to a site whose analytics you can inspect. For example, an email you control could link to your own guide:

https://example.com/guide/?utm_source=newsletter&utm_medium=email&utm_campaign=offer_guide&utm_content=body_link

This is an illustrative URL, not a working campaign. It labels the newsletter source, email medium, broad campaign and link placement. Google recommends consistently using utm_source, utm_medium and utm_campaign, and notes that values are case-sensitive. In its reporting, Newsletter and newsletter are different values. Use a consistent naming convention and check the receiving property’s documentation. 6

Affiliate sub-IDs or tracking IDs are network-specific fields, when a program supports them, that may identify a placement in affiliate reports. A permitted value such as guide_body_a can be useful, but the field name, allowed characters, length and reporting behavior must come from the network or program. Do not assume that a parameter called subid works everywhere, and do not append arbitrary UTMs to a merchant link unless the program allows it.

The distinction matters: a UTM on your incoming email link can help you understand traffic reaching your guide. It does not automatically appear in the affiliate network’s sales report. Likewise, a network sub-ID may identify an affiliate placement without telling your analytics property which newsletter brought the reader there. Some integrations can pass additional data, but verify the actual connection and consent rules instead of assuming a bridge exists.

Set up a simple naming register

Create a small shared register before publishing links. For each label record its owner, channel, campaign purpose, placement, destination, creation date, status and permitted platform. Use stable lowercase names such as newsletter, email, offer_guide and body_link. Keep a short approved vocabulary so two editors do not use email, Email and e-mail for the same channel. Retire old labels rather than quietly reusing them for a new meaning.

Never encode a person’s name, email, phone number, account ID, purchase details or sensitive search term in a campaign label or URL. Avoid individual-level codes, even if they look random or hashed: a code can still be linkable to a person. Google explicitly warns against sending personally identifiable information in campaign parameters, URLs, page titles and other analytics fields. Its email redaction is described as best-effort, so prevent the leak at the source rather than relying on cleanup. 7

Before release, check the full URL in a private test session. Preserve any existing query string and fragment correctly, confirm that the expected destination loads, and ensure the network still recognizes the approved link. Do not copy a customer’s URL into a public issue or screenshot. If the link breaks, remove the unverified parameter and ask the program for a supported format.

Make the reporting window match the question

A calendar report and a click-cohort report are not the same thing. A calendar report can include purchases this month from people who clicked earlier. If you divide those purchases by this month’s clicks, the result is a mixed-period ratio, not the conversion rate of this month’s click cohort.

A click cohort starts with clicks recorded in a defined interval, then follows actions attributed to those same clicks. If the network provides that view, record its exact report type and filters. If it does not, you can still report calendar totals, but label the mismatch and do not call the ratio a cohort conversion rate. Aggregate cohort reporting does not require you to build a separate person-level database.

For each report, record:

  • The placement, offer, link or approved sub-ID, source and currency.
  • The click period, time zone, report date and whether dates refer to clicks, transactions, approvals or payouts.
  • Whether clicks are raw, network-filtered or unique under a stated definition; whether actions mean leads, purchases or renewals.
  • The applicable attribution rule and window, plus the program’s review, reversal and payment terms. Mark any undisclosed detail unknown.
  • Whether numbers are pending, approved or paid, whether adjustments are included, and what remains provisional.

A cohort can mature only after its last eligible click has had the full applicable attribution window, relevant transactions have passed the program’s review process, and any later payment or adjustment has been checked. If a program does not disclose a review period, do not invent a maturity date: set a review reminder and keep the report provisional. For recurring commissions, specify a separate observation end date; future renewals and reversals cannot be known today.

Attribution is a contractual reporting rule, not a universal journey map. Awin describes a hierarchy in which standard and incentive clicks ordinarily take precedence over soft clicks and post-view contacts; details depend on the implementation and program. PartnerStack’s public overview describes its own referral-cookie behavior and a same-browser-and-device limitation. These are platform-specific descriptions, not rules you can apply to every offer. Read the applicable offer terms and ask the network where the rule is unclear. 8 9

Worked example: reconcile a hypothetical cohort

Suppose a single placement has a defined click cohort. For this example only, assume the network report can connect its recorded clicks and subsequent actions to that same cohort, and all relevant attribution, review and payment periods have passed:

  • Your site recorded 100 outbound-click events.
  • The network recorded 80 clicks from the same placement and cohort. Assume these correspond to 80 of those outbound events for this illustration; real dashboards may not permit that matching.
  • The network attributed 4 initial purchases to those clicks, each initially earning $30. Assume the purchases came from four different recorded clicks.
  • One $30 commission was reversed; the remaining three were approved and paid. Assume no other adjustments.

The tracked purchase-to-network-click ratio is 4 ÷ 80 = 5%. The initial pending amount is 4 × $30 = $120, while paid commission is 3 × $30 = $90. The paid commission per network-recorded click is $90 ÷ 80 = $1.125, or about $1.13 rounded to cents. Label it “paid commission per network-recorded click for this cohort,” not a guaranteed EPC or profit.

You could also calculate 4 ÷ 100 = 4% using the site’s outbound events, but it answers a different question because the denominator comes from a different system. It does not measure the purchase rate among unique people. Neither percentage includes purchases the network did not attribute to the affiliate.

The difference between 100 and 80 is an investigation prompt, not evidence that the network “lost 20 sales” or even 20 eligible commissions. Review definitions, test results, filters and reporting latency first. A privacy choice or blocked request may also affect observation; do not try to infer an individual’s choice from aggregate totals.

Now assume, hypothetically, that attributable promotional spending was $50 and a directly related fee was $5. Paid commission minus those stated costs is $90 − $50 − $5 = $35. The $35 excludes other possible costs, taxes, overhead and the value of your time; it is not a complete profit figure. If either expense is unknown, show it as unknown instead of treating it as zero.

EPC is only useful with its ingredients

Earnings per click divides a stated earnings amount by a stated click count. A program’s published EPC may use a different period, population, status or click definition from yours. It may also combine traffic sources you cannot replicate. Ask whether the numerator includes pending commissions, refunds, adjustments or renewals, and whether the denominator is raw, unique or filtered. If those ingredients are not disclosed, mark comparability unknown. Do not present a program-reported EPC as an expected result for your audience.

A repeatable reconciliation workflow

A weekly or monthly review can stay small. The goal is to identify an actionable problem, not to force every dashboard to match.

  1. Freeze the scope. Choose one offer and placement. Save the report date, click period, time zone, attribution terms and source. Note any campaign, destination, consent or program changes during the period.
  2. Test the path. Check the current link and permitted sub-ID without generating prohibited traffic or purchases. Verify the consent paths you control. Note a failed redirect or an unexpected parameter before interpreting the metrics.
  3. Collect aggregate figures. Pull the onsite event count and the network’s corresponding click, action and commission totals. Apply matching filters where possible. Export only fields needed for the question.
  4. Separate status and timing. List pending, approved, reversed and paid amounts; compare cash received with the payment record. A network’s “pending” label may include more than one intermediate state, so use its definition.
  5. Investigate material differences. Check raw versus filtered clicks, time zones, repeat events, changed destinations, reporting delays and cohort availability. Check renewal inclusion and currencies before comparing totals. Do not assume the privacy explanation, technical failure or fraud without evidence.
  6. Ask a narrow support question. If a discrepancy remains, send the network the report name, date range, time zone, link ID and definition of the mismatch through its approved channel. Share transaction details only when necessary and permitted. Ask which rule or status explains the difference; avoid sending customer data in ordinary email.
  7. Record the decision and next review. Examples include fixing a broken link, leaving a report provisional, or pausing an offer whose terms cannot be verified. Keep the unanswered question visible when you revisit it.

A compact reporting note might read: “Placement guide_body_a; click cohort [start–end], [time zone]; network-recorded clicks [count]; attributed initial purchases [count]; paid commission [amount and currency]; review status [provisional/final for stated horizon]; rule or definition [known/unknown]; next action [task].” Brackets are fields to complete, not suggested results. Avoid customer-level columns unless there is a documented need, an appropriate basis, controlled access and a retention plan.

Test placements without overstating results

A useful experiment starts with a decision: for example, whether readers can find a relevant recommendation in the body of a guide. Write the hypothesis and success measure before changing the page. Keep the offer, destination, substantive claims and clear disclosure consistent. If possible, assign eligible impressions to versions using an appropriate test method and document how assignment works. Decide when to review; do not repeatedly stop at the first favorable fluctuation.

Hypothetical mini case: two placements. In a defined test, placement A receives 200 measured placement impressions and 20 recorded outbound clicks. Placement B receives 100 measured placement impressions and 15 outbound clicks. Their observed click-through ratios are 20 ÷ 200 = 10% and 15 ÷ 100 = 15%, respectively. Assume both use the same impression and event definitions and the same consent-aware measurement setup.

B’s observed rate is higher, but those figures alone do not prove the design caused the difference. Assignment may have been uneven, the samples are small, and unmeasured visitors may differ from measured visitors. If A appeared near an introduction and B appeared after a detailed explanation, position and reader intent also differ. Record the caveat, inspect usability and continue gathering appropriate evidence if a decision needs more confidence.

Clicks alone do not establish a better buyer outcome. Suppose the network later attributes two purchases to A’s 16 recorded network clicks and one purchase to B’s 12 recorded network clicks, within the same defined click cohorts. The observed ratios are 2 ÷ 16 = 12.5% and 1 ÷ 12 ≈ 8.3%. These tiny, hypothetical samples are not a reliable ranking or a reason to hide unfavorable data. Use matching cohorts and report the counts alongside percentages. Do not splice these figures into a causal claim about user satisfaction or profit.

If you cannot assign traffic fairly or identify comparable impressions, call the exercise an observation rather than an experiment. A broken link or inaccurate price should be corrected immediately; it does not need to wait for a test to finish. A test should never use deceptive urgency, suppress required disclosures or bypass consent in pursuit of a higher count.

Protect the data you decide to keep

An aggregate count can still become sensitive when combined with other records, especially in a small segment. “Anonymous” is not a label to apply automatically. Consider whether a person could be singled out or linked back through a URL, timestamp, sub-ID or external dataset.

The FTC recommends minimizing sensitive information, limiting access, setting retention rules and disposing of information when it is no longer needed. Apply that principle to both analytics exports and network reports. Some network commission reports can contain customer names, emails and customer keys, so select aggregate fields when possible. 1 10

Give access only to people who need it. Protect accounts with strong authentication, keep exports in an appropriately secured location rather than a public spreadsheet, and remove access when a role changes. Decide how long each report is needed and how it will be deleted, accounting for any applicable recordkeeping obligations. Do not promise that a particular deletion schedule is legally sufficient for every business.

If an identifier appears in a URL or export by mistake, stop the source of collection, restrict access to the affected material, document the incident and follow your incident-response and notification obligations where applicable. Do not paste the identifier into a public support ticket. Ask the relevant provider how to remove or correct affected data; masking a screenshot does not erase the underlying record.

Common mistakes that make a report misleading

  • Calling attributed purchases “all sales.” The network can report only what its system observes and credits under its rules.
  • Treating every click as a person. Repeated actions, filters and different definitions make event counts and people counts different.
  • Dividing unlike periods. Calendar-month purchases over calendar-month clicks may mix unrelated cohorts. Label a mixed-period ratio or omit it.
  • Comparing pending money with paid money. Show status, reversals and the reporting date. Do not equate approved commission with cash received.
  • Assuming “first-party” means exempt or invisible. Check the actual storage, access, recipients and purpose, not the architecture label. 2
  • Using names or individual identifiers as sub-IDs. A campaign label should describe the placement, not the reader. Even an aggregate-looking code can become identifying when joined with another file.
  • Treating the consent banner as a test result. Verify what each tag and redirect actually does when a choice is made or changed. Consent Mode alone is not a banner. 4
  • Optimizing the easiest number at the reader’s expense. More clicks obtained through an unclear claim or hidden disclosure are not evidence of a better recommendation.

Questions to ask before trusting a dashboard

Why are site clicks and network clicks different?

They may count different events or apply different filters. Timing, consent choices, technical failures and redirects can also matter. Compare definitions and time zones, test the path, and ask the network about a specific unresolved gap. The difference alone does not identify its cause.

Can I track sales without identifying readers myself?

A network may offer aggregate placement or link reports that avoid your maintaining a customer-level dataset. Availability and permitted use vary by program, and aggregate reporting does not automatically remove consent or privacy obligations. Check the data flow and terms.

Does a UTM make an affiliate sale trackable?

Not by itself. A UTM can label an incoming campaign for a receiving analytics property; an affiliate program may instead require its own referral link or approved sub-ID. Check both systems, and do not assume they share data.

When is a cohort final?

Only relative to a stated observation horizon. Allow the relevant click attribution window and review process to run, then check payments and adjustments. Recurring commissions may continue or reverse later. If timing is not disclosed, keep the result provisional.

What if a visitor declines tracking?

Honor the choice in the tools you control and follow the applicable rules for the full journey. If a tracked affiliate link cannot be used without the required consent, consider an ordinary destination link where permitted, while recognizing that the destination may use its own technologies. Do not work around the choice with fingerprinting or an undisclosed identifier.

Your next review

Choose one live placement. Verify its disclosure and destination, inventory the tracking and consent behavior you control, and create one entry in your naming register. Then write a report with a defined click period, network click definition, attributed action type, commission status and explicit unknowns. Schedule the next check around the program’s actual terms, not an assumed deadline.

Use the resulting evidence alongside audience fit and funnel inspection. Tracking can help you detect friction or reconcile a payment; it cannot replace an honest judgment about whether an offer deserves the recommendation.

Sources