Social media scheduling attribution methods compared on one evidence grid

Social media scheduling attribution methods answer different questions. Some assign credit to observed outcomes. Others compare groups or periods to estimate what the activity changed. No method wins for every decision.

This comparison uses one hypothetical England business question: did a defined set of scheduled posts contribute to eligible enquiries during a fixed period? Values remain blank. Every method uses the same outcome event, population, currency where relevant, Europe/London dates and full-cost boundary.

What to take away

Attribution methods assign credit or estimate effects, but no single method suits every decision.
Rule-based attribution reconciles finance but cannot show what would have happened without the posts.
Observational comparisons yield adjusted associations unless design and assumptions justify causal claims.
Impact evaluation needs a credible counterfactual; outcome monitoring alone cannot establish causality.
Choose the least complex method that answers the decision honestly and preserve an unattributed category.

Common evidence grid

Output and unit

Last eligible touch: Count or GBP contribution credited to the last included touch before the outcome
Fixed multi-touch rule: Fractional count or GBP contribution allocated across included touches
Observational cohort comparison: Difference in outcome rate between defined exposed and comparison populations
Controlled or quasi-experimental evaluation: Estimated incremental outcome, with uncertainty

Minimum data

Last eligible touch: Dated touch records, eligible outcome and fixed lookback rule
Fixed multi-touch rule: Ordered path plus published weights
Observational cohort comparison: Comparable populations, complete outcomes and confounder plan
Controlled or quasi-experimental evaluation: Pre-specified intervention, credible counterfactual and proportionate analysis

Main assumption

Last eligible touch: The chosen final touch deserves all assigned credit
Fixed multi-touch rule: Editorial weights reasonably represent contribution
Observational cohort comparison: Measured adjustment deals adequately with selection differences
Controlled or quasi-experimental evaluation: Design assumptions hold and contamination is understood

Limitation

Last eligible touch: Earlier influences and unobserved routes receive none
Fixed multi-touch rule: Different weights change the answer without new outcomes
Observational cohort comparison: Unmeasured differences may explain the association
Controlled or quasi-experimental evaluation: Cost, feasibility, power and implementation quality constrain inference

Suitable wording

Last eligible touch: credited under last-touch rule
Fixed multi-touch rule: allocated under named weights
Observational cohort comparison: associated difference
Controlled or quasi-experimental evaluation: estimated causal effect

For every row, the outcome card states the event, eligible population, numerator, denominator, unit, date range, timezone, exclusions, data source and query version. A count has no denominator; a rate does. Missing observations are reported rather than silently discarded.

Attribution methods evidence grid

Last eligible touch

Output and unit: Count or GBP credited
Minimum data: Dated touches, outcome, lookback
Main assumption: Final touch deserves credit
Limitation: Earlier influences get none
Suitable wording: credited under last-touch rule

Fixed multi-touch rule

Output and unit: Fractional count or GBP
Minimum data: Ordered path plus weights
Main assumption: Weights represent contribution
Limitation: Weights change answer
Suitable wording: allocated under named weights

Observational cohort

Output and unit: Outcome-rate difference
Minimum data: Comparable groups, confounder plan
Main assumption: Adjustment handles selection
Limitation: Unmeasured differences remain
Suitable wording: associated difference

Controlled evaluation

Output and unit: Incremental outcome, uncertainty
Minimum data: Intervention, counterfactual, analysis
Main assumption: Design assumptions hold
Limitation: Cost, feasibility, power constrain
Suitable wording: estimated causal effect

Rule-based attribution

Last-touch and fixed multi-touch models can make finance reconciliation consistent. They do not reveal what would have happened without the posts. Save the attribution window, included channels, duplicate rule, identity method, outcome timestamp and version.

Google's Analytics attribution overview, checked on 6 September 2026, defines attribution as assigning credit and documents current product models and exclusions. Those are supplier-specific statements. A Google model name must not be used as evidence that a social campaign caused an outcome.

Observational comparison

A cohort design may compare eligible people or sessions with different recorded exposure. The analyst should pre-specify the exposure event, matching variables, outcome window and exclusions. Platform delivery, customer intent, seasonality or another campaign can still affect both exposure and outcome.

The output is an adjusted association unless the design and assumptions justify more. Present the two population counts, both outcome-rate numerators and denominators, the adjusted difference, uncertainty and missingness.

Impact evaluation

HM Treasury's 2026 Magenta Book explains experimental and quasi-experimental approaches around a counterfactual. Its QPIE guidance says outcome monitoring alone cannot establish causality.

Random allocation may be inappropriate, impractical or underpowered. A qualified evaluator should assess feasibility, ethics, contamination, sample size and analysis. If a credible design is unavailable, report the gap rather than relabel attribution as impact.

Privacy and decision rule

Joining identities across platforms and destination systems can involve personal data and device information. The ICO's current storage and access technologies guidance keeps PECR and UK GDPR questions visible. Minimise collection and have the qualified practitioner review the real design.

Choose the least complex method that answers the decision honestly. Use rules for credit allocation, observational work for bounded association and impact evaluation for causal questions. Preserve a manual unattributed category. A method that cannot represent missing evidence is not ready for investment claims.

Before you act

Define the outcome event, population, currency, dates and cost boundary.
Record the attribution window, included channels, duplicate rule and identity method.
Pre-specify exposure, matching variables, outcome window and exclusions for cohort designs.
Assess feasibility, ethics, contamination, sample size and analysis for impact evaluation.
Minimise collection of personal data and device information; have a qualified practitioner review.
Report missing observations rather than silently discarding them.

Common questions

What is the difference between assigned credit and a causal estimate?

Assigned credit methods, such as last-touch or fixed multi-touch rules, allocate observed outcomes to touches. Causal estimates come from controlled or quasi-experimental evaluations that compare outcomes against a credible counterfactual. The article states that rule-based models do not reveal what would have happened without the posts, while impact evaluation estimates incremental outcomes with uncertainty.

When can an observational cohort comparison support a causal claim?

An observational cohort comparison yields an adjusted association unless the design and assumptions justify more. The article says the output is an adjusted association, and unmeasured differences may explain the association. To support causality, the design must adequately deal with selection differences and meet other assumptions, which is rare without a credible counterfactual.

What should be done if a credible impact evaluation design is unavailable?

If a credible design is unavailable, report the gap rather than relabel attribution as impact. The article advises choosing the least complex method that answers the decision honestly, using rules for credit allocation and observational work for bounded association. Preserve a manual unattributed category and avoid investment claims that cannot represent missing evidence.