Attribution Model Translator

Variables

Replace each placeholder before you run the prompt.

• {{last_click_data}}Per-channel: conversions and revenue under last-click attribution.

  Last 90 days, last-click: Paid Search 200 conv / $1.2M, Organic Search 80 conv / $400K, Direct 60 conv / $800K, Email 40 conv / $200K.

• {{context}}B2B context — sales cycle length, multi-touch reality, branded-search prevalence.

  B2B SaaS, 90-day sales cycle, ~7 touches per closed deal, brand search dominant in last-click.

Translate the last-click attribution data below into an estimate of what data-driven attribution would credit instead. Be explicit about your assumptions.

LAST-CLICK DATA:
{{last_click_data}}

CONTEXT:
{{context}}

OUTPUT:

1. Estimated data-driven attribution per channel. For each: conversions and revenue under DDA, and the % delta from last-click.
2. Reasoning per channel: why DDA would credit this channel more / less than last-click. Tie reasoning to the cited context (sales cycle, branded prevalence, etc.).
3. The "branded-search ceiling": what fraction of last-click branded-search revenue is genuinely incremental vs would have closed anyway?
4. Confidence rating per channel (high / medium / low) — where is the model most likely wrong?
5. A "if we could only run one DDA experiment" recommendation.

Do NOT claim DDA gives "the truth" — both models are estimates. Be explicit that this is one estimate among several.
Do NOT inflate paid search credit beyond what the cited context can support.

Expected output shape

Per-channel DDA estimate with reasoning, branded-search ceiling discussion, confidence ratings, and one recommended experiment.

Why we wrote it

Most teams have last-click data and a vague sense that DDA "would change things." This prompt makes the change explicit and explainable to a CFO.

How to use

1. Open Claude or ChatGPT. The recommended model for this prompt is claude-opus-4-7 — opus when the prompt requires deep reasoning, sonnet for the rest.
2. Replace every {{variable}} with content specific to your account. The examples above are starting points, not templates to ship as-is.
3. Paste the prompt and run.
4. Read the output against the expected shape above. If the model produced a structurally different response, re-prompt rather than accept the drift.