How AI Recommendations Work
How AI Recommendations Work
Attribution.ai uses AI for analysis, explanation, and decision support. It does not replace the underlying measurement system, and it should not be treated as proof of causality unless completed experiment evidence supports that claim.
What Every Recommendation Should Show
Every recommendation should include:
- Evidence basis: the source context used, such as MMM output, forecast pace, stored insight data, survey recovery, commerce scope, or measurement readiness.
- Confidence or readiness: whether the recommendation is ready to review, directional only, or blocked by setup.
- Why we recommend this: a plain-language explanation of the observed signal.
- Missing data: the specific data that could change the answer, such as paid media spend, survey recovery, order webhooks, commerce surfaces, forecast targets, or a green readiness gate.
Readiness Labels
| Label | Meaning | How to use it |
|---|---|---|
| Ready to review | The available evidence is strong enough for operator review. | Review source rows and scenarios before making changes. |
| Review first | Some evidence exists, but one or more checks need attention. | Treat as planning support until the listed gaps are closed. |
| Directional only | The system has a signal, but not enough evidence for action. | Use it to investigate, not to move budget directly. |
| Setup needed | Required data or integrations are missing. | Finish setup before relying on the recommendation. |
| Needs evidence | A stored insight lacks source context or a concrete next step. | Inspect the related dashboard view before acting. |
If AI Is Unavailable
When AI generation is unavailable or incomplete, Ask AI should degrade into a deterministic account read where possible. That fallback may use loaded workspace data, readiness checks, or recovered tool output, but it must say that the answer is bounded and list missing data.
If a fallback only shows setup or source rows, do not treat it as a generated recommendation.
Safe Weekly Review
- Start with Home and Performance to confirm business health and freshness.
- Open Budget or Recommendations only after checking the readiness label.
- Read the evidence basis and missing-data chips before changing platform spend.
- Use Ask AI to explain a recommendation, but verify the source rows and scenario before action.
- Defer spend changes when the recommendation says setup is needed, directional only, or missing evidence includes the readiness gate, paid spend, order webhooks, or commerce scope.
Escalate to Support
Contact support@attribution.ai if a recommendation appears without evidence context, if the stated source rows are missing from the dashboard, or if the missing-data list conflicts with integration status.