SkillVaultskills Browse all 500 skills

Marketing · Version 1.4.0 · Reviewed 2026-08-02

Marketing Attribution Analyst

Make model selection and incrementality with evidence, explicit trade-offs, and a verification plan.

4 method steps 5 documented failure modes 5 diagnostic checks 7 quality gates

Analyzes channel attribution with honest treatment of multi-touch, incrementality, and overlap.

₹99 one-time

Get this skill archive

What this skill helps you do

  • Model selection
  • Incrementality
  • Overlap handling

How Marketing Attribution Analyst works

You provide

Customer evidence, segment data, and the decision at stake

It inspects

Sample quality and selection bias behind model selection

It decides

A incrementality recommendation with confidence stated

You verify

A leading indicator defined before the lagging metric moves

What it checks first

Marketing Attribution Analyst analyzes channel attribution with honest treatment of multi-touch, incrementality, and overlap. Use it when the work involves Model selection, Incrementality, Overlap handling.

  1. Whether a claim is supported by an observed customer behavior or by an internal opinion restated as fact.
  2. Sample size and selection bias behind any pattern, since three loud accounts are not a segment.
  3. Whether the stated problem is the customer's framing or the vendor's framing of the customer.
  4. The decision the analysis must support, and what evidence would change that decision.
  5. Whether a metric measures activity or outcome, because activity metrics reliably improve without results improving.

Failure modes it recognizes

  • Confusing correlation with mechanism, so a segment that converts well is credited to messaging rather than to pre-existing intent.
  • Survivorship bias from analyzing only closed-won accounts, which hides the reason others left.
  • Objection handling that addresses the stated objection rather than the underlying risk the buyer perceives.
  • A positioning claim that competitors can make identically, which produces no differentiation.
  • Optimizing a funnel stage in isolation and shifting the bottleneck downstream without net gain.

Answers it will reject

  • Presenting a conclusion without stating the evidence quality, which prevents anyone from disagreeing usefully.
  • Using aggregate averages across mixed segments, which hides that one segment is failing badly.
  • Treating a feature list as a value proposition, which forces the buyer to do the translation.
  • Declaring intent from a single interaction rather than a behavioral pattern.

Decision rules it applies

  • State the confidence and the sample behind every claim; an unqualified claim is not usable for a decision.
  • Separate what the evidence shows, what it suggests, and what is assumed.
  • Prefer a smaller, well-evidenced recommendation over a broad, weakly supported strategy.
  • Define the measurement and the review date before the recommendation is accepted.

Evidence it asks for

  • Quote the source evidence — call excerpt, ticket, usage metric — beside each finding.
  • Report counts and denominators, not only percentages.
  • Define the leading indicator that will show whether the recommendation is working before the lagging metric moves.

The method inside

  1. Start from audience evidence and the behavior to change
  2. Map each message to a problem, outcome, and proof
  3. Remove claims that are vague, duplicated, or unsupported
  4. Define the measurement and approval criteria before publication

Deliverables

  • Model selection evidence map
  • Incrementality revision brief
  • Overlap handling measurement checklist

Evidence requirements

  • Approved claims and supporting research
  • Specific audience, buying situation, and desired behavior
  • Customer language, campaign brief, or content source material

Quality gates

  • Every material claim traces to supplied evidence or is labeled as a hypothesis.
  • The response follows the declared deliverable contract.
  • No execution, access, measurement, or verification is invented.
  • Secrets and personal data are redacted rather than repeated.
  • The user receives a concrete independent verification step.
  • The relevant failure modes in this domain were considered rather than only the reported symptom.
  • No listed anti-pattern was recommended as a solution.

Example task

Input

Last-touch attribution says paid search drives everything, so we want to shift budget into it.

Expected output

Last-touch credits the channel closest to conversion, which is why branded search always looks dominant: those users were already coming. Test incrementality with a holdout before reallocating, because the measured lift is usually far below the attributed value...

Boundaries and compatibility

Ideal for

  • Model selection: produce a decision or artifact grounded in supplied evidence.
  • Incrementality: produce a decision or artifact grounded in supplied evidence.
  • Overlap handling: produce a decision or artifact grounded in supplied evidence.

Out of scope

  • Fabricating testimonials, statistics, or competitive claims
  • Presenting correlation as causal proof

Agent compatibility

  • GitHub Copilot custom agents
  • Claude Agent Skills / SKILL.md
  • Any instruction-following chat model

Tool policy: Advisory by default. No tools are assumed. If the host provides tools, use read-only evidence gathering unless the user explicitly approves a scoped write or execution action.