Understanding results

Begin with the verdict

SaliencyLab uses three action states:

• Rebuild — the evidence points to a foundational creative problem.
• Sharpen — the idea is legible, but a focused edit should improve the decision.
• Scale — the creative is comparatively strong enough to justify the next real-world step.

“Scale” is not a performance guarantee. It means the modeled creative evidence is comparatively strong; media performance still depends on targeting, auction conditions, offer, landing experience, and measurement.

The verdict is a decision aid, not a label of creative quality in every context. A brand film and a short direct-response placement may require different tradeoffs even when they use the same product and audience.

Scores are diagnostic dimensions

Read each KPI with its evidence. A high score without clear brand attribution may still be a poor brand decision. A lower proposition score may be expected for an upper-funnel execution. Context determines which gap matters.

Read the result from claim to evidence

Use this order whenever a finding appears important:

1. Identify the claim made by the verdict or KPI.
2. Locate the visual, audio, copy, or timing evidence supporting it.
3. Check whether the project and creative metadata are correct.
4. Review the comparison benchmark and confidence.
5. Decide whether the evidence is strong enough to change the creative.

This sequence reduces the risk of treating a prominent score as self-explanatory.

Benchmarks and confidence

Use the narrowest relevant slice supported by enough evidence. Broader fallback slices can be directionally useful, but should not be described as precise category norms.

Look for:

• Platform and placement match
• Market and category match
• Sample size
• Observed versus modeled labeling
• Confidence or fallback language

Exact and fallback benchmarks

An exact comparison uses the most relevant available combination of context. If that slice lacks enough support, the system may fall back to a broader comparison. A fallback can still be useful, but it answers a less specific question. Keep the fallback label when sharing the result.

What confidence means

Confidence describes support for the modeled interpretation given the available evidence and context. It does not mean “chance of campaign success,” and it should not be converted into a forecast of CTR, sales, or return on ad spend.

Turn evidence into an edit

A useful recommendation names what to change, where it appears in the creative, and which KPI gap it addresses. Prefer one testable edit over a list of cosmetic suggestions.

Share results without losing context

When presenting a result to a client or stakeholder, include:

• The creative version and intended placement
• The action verdict and its explanation
• The relevant evidence from the asset
• The benchmark context and any fallback
• The proposed edit or next test
• A clear distinction between modeled and observed evidence

Avoid presenting a cropped score, removing confidence language, or describing a modeled result as measured consumer response.

Common interpretation mistakes

• Maximizing every KPI: creative strategy often requires deliberate emphasis and tradeoffs.
• Comparing unlike formats: a short feed video and a longer pre-roll solve different attention problems.
• Ignoring attribution: attention is not automatically valuable if viewers cannot connect it to the brand.
• Treating a small change as certainty: direction and evidence usually matter more than a narrow numerical gap.
• Using BuyerLens as validation: synthetic perspectives explain hypotheses; they are not an independent outcome measure.

For evidence definitions, continue to Methodology and confidence.