Measurement
Part of Influencer agencies measurement: populations and an evidence ladder
Influencer attribution methods: how to compare them on one campaign
Compare influencer attribution methods on one campaign: evidence grids, platform claims, counterfactual tests, descriptive review, and no universal winner.
Influencer agencies attribution methods answer different questions. Compare them on one England campaign, one commercial event, one period and the same full-cost boundary.
This comparison was researched on 6 September 2026. Google and TikTok documentation below describes global platform products, not independent validation or England-specific evidence. The London-based Advertising Standards Authority (ASA) applies UK advertising rules to England campaigns, but it does not validate attribution results.
Attribution methods compared
Evidence and unit
- Campaign code or link rule
- Accepted internal events matched to a documented code or link; count and contribution value
- Platform-reported attribution
- Platform conversion field, model and window; credited events and stated value
- Internal multi-touch rule
- Deduplicated event path across permissioned sources; fractional or single credit
- Counterfactual evaluation
- Outcome difference between comparable exposed and unexposed units; incremental outcome and uncertainty
- Descriptive contribution review
- Operational, attributed and qualitative evidence reported separately
What it can support
- Campaign code or link rule
- Traceable rule-based credit where identifiers survive
- Platform-reported attribution
- Activity credited inside that configured system
- Internal multi-touch rule
- Consistent allocation under a buyer-owned rule
- Counterfactual evaluation
- Causal inference when design and data are adequate
- Descriptive contribution review
- Bounded learning where causal study is infeasible
Main failure mode
- Campaign code or link rule
- Untracked paths, sharing, leakage and selection into code use
- Platform-reported attribution
- Duplicate credit, modelled records and incompatible windows
- Internal multi-touch rule
- Identity gaps and a rule mistaken for causal effect
- Counterfactual evaluation
- Contamination, weak power or non-comparable groups
- Descriptive contribution review
- Narrative certainty that exceeds the evidence
What to take away
- Each method answers a different question under the same campaign and cost boundary.
- Counterfactual evaluation is the only row that supports causal language when design and data are adequate.
- Where evaluation is infeasible, report delivery, attributed records, costs and rival explanations separately.
- Privacy approval under PECR and UK GDPR is independent of the statistical quality of the method.
Apply one evidence grid
For every row, record the event, eligible population, numerator and denominator or unit, period, Europe/London timezone, source, configuration version, exclusions, missing and late data, owner and buyer threshold. Name the spend or volume level the buyer treats as material. Keep creator payments, agency fee, gifting, media, production, software, internal labour, assurance and corrections inside one cost boundary.
Evidence grid fields to record
- Event and eligible population
- Numerator and denominator or unit
- Period and Europe/London timezone
- Source and configuration version
- Exclusions, missing and late data
- Owner and buyer threshold
Define which accepted orders qualify for a code method, including cut-off, cancellations, returns, multiple codes and unattributed events. The result is rule-attributed contribution, not incrementality.
Worked example, illustrative and anonymised: assume an England homeware brand runs one creator campaign in March 2026 with a £18,000 full-cost boundary. Platform reporting credits 240 orders. The link rule matches 150, internal multi-touch rules deduplicate to 130 and a holdout suggests 45 to 70 incremental orders with wide uncertainty.
Descriptive review records delivery, cost and rival explanations. These figures are illustrative only, not a result to copy. The grid forces each method to answer on the same event, period and cost boundary.
What to ask an influencer agency
Before you compare methods, put the same questions to every influencer agency on your shortlist. Ask which attribution model and conversion window sit behind any platform total, whether credited events are deduplicated or added across platforms, how view-through credit is treated, and whether creator payments, agency fee, gifting, media, production, software and labour sit inside one cost boundary.
Ask who owns the raw records, what privacy review applies to the tracking, and whether the agency can support a holdout or matched comparison. The answers show which row of the grid each agency's numbers actually belong to.
Read platform totals as configured claims
Google explains that an attribution model assigns conversion credit across eligible interactions. Its conversion-window documentation states that the selected period determines which post-interaction conversions are recorded. Capture both settings with the export.
Google's Analytics Help records that the first click, linear, time decay and position-based attribution models are no longer available as of November 2023, which leaves last-click and data-driven models as the practical choices in Google Analytics 4. Before you compare, ask whether the platform total came from Google Ads or from Google Analytics 4, because each product applies its own model and conversion window.
TikTok's first-party page, updated in November 2025, lists click-through and view-through choices within ad-group attribution settings in TikTok Ads Manager. This is a supplier description. It does not show that a credited event was caused by a creator asset or that the same order was not credited elsewhere.
Do not add platform conversion totals. Reconcile identifiers to internal records where permitted, report unmatched and modelled records, and keep view-through credit distinct from a buyer's link rule.
Reserve causal language for a suitable design
Counterfactual evaluation compares observed outcomes with a credible estimate of what would have happened without the campaign. HM Treasury and Cabinet Office QPIE guidance says monitoring a change alone does not establish causation and explains why matched or randomised comparison groups can strengthen attribution.
Predefine assignment, outcome, analysis, contamination checks and uncertainty. A holdout may fail if audiences cross groups or the test lacks power. A named evaluator should review the design.
Where strong evaluation is not feasible, use the descriptive contribution row. Present delivery evidence, audience observations, attributed commercial records, costs and rival explanations without converting them into an incremental percentage.
Decide without overclaiming
Choose the method that answers the buyer's question with data that can lawfully and securely be obtained. The ICO's 2026 storage and access technologies update confirms that tracking technologies sit within a PECR and, where relevant, UK GDPR boundary.
Hold the report if the internal event is undefined, settings are absent, duplicated credit cannot be explained, cost is incomplete or causal language lacks a credible counterfactual. Qualified evaluation, privacy and PECR, advertising, consumer and finance reviewers must assess the configured case. This is a research framework, not legal or financial advice.
Before you act
- Record the event, eligible population, unit and period for every row.
- Record source, configuration version, exclusions, owner and buyer threshold.
- Keep creator payments, fees, gifting, media, production and labour inside one cost boundary.
- Capture the attribution model and conversion window settings with the platform export.
- Do not add platform conversion totals; reconcile identifiers and report unmatched records.
- Predefine assignment, outcome, analysis, contamination checks and uncertainty before any holdout.
Common questions
How should a buyer compare influencer attribution methods on one campaign?
Each method answers a different question, so the comparison runs on one England campaign, one commercial event, one period and the same full-cost boundary. Campaign codes, platform attribution, multi-touch rules, counterfactual evaluation and descriptive review each carry distinct failure modes.
What does a platform-reported conversion total actually prove?
It shows activity credited inside that configured system, using the platform's conversion field, model and window. Google and TikTok documentation describes global products, not independent validation or England-specific evidence.
When can a report use causal language?
Only when a suitable counterfactual design supports it. QPIE guidance says monitoring a change alone does not establish causation, and matched or randomised comparison groups can strengthen attribution. Predefine assignment, outcome, analysis, contamination checks and uncertainty, and have a named evaluator review the design.
