Measurement / framework / Free to use

The Creative Delivery Evidence Review Framework

Separate uploaded assets, eligible ads and actual exposure before deciding what creative to retire or test next. Includes a reusable evidence worksheet and a worked review.

Growthcraft Editorial · 2026-10-08. AI-assisted research and implementation. Examples are synthetic; Akshay's personal review is not claimed.

Copyable template

Select and copy the complete template below. With JavaScript enabled, you can edit, copy and download it in the interactive workspace.

CREATIVE DELIVERY REVIEW — original Growthcraft synthesis
Decision / reviewer / review date:
1. DEFINE: campaign, reporting window [start,end), timezone, export timestamp:
Reporting grain (ad, asset, combination); why impressions are additive:
Eligible creative roster, ID/version map, approval and active dates:
Business objective and mature outcome evidence (separate from exposure):
2. RECONCILE: one unique row per creative, zero-delivery roster included:
Source total / summed rows / excluded rows with reasons:
Currency, audience, placement, bid and eligibility differences:
3. DESCRIBE: total / listed / delivered / zero-delivery / largest share:
Squared-share concentration / inverse concentration (effective count):
Comparison window and any roster or eligibility changes:
4. INVESTIGATE: export defect / ineligibility / adaptive allocation / unknown:
Evidence that supports and contradicts each explanation:
5. DECIDE: HOLD_FOR_EVIDENCE / READY_FOR_HUMAN_REVIEW:
Proposed follow-up, owner, cost constraint, expiry and rollback:
If a causal question remains: experiment unit, randomization, primary outcome,
minimum detectable effect, power plan, guardrails and stopping rule:
No automatic creative retirement, bidding or budget changes.

Use exposure evidence before judging a creative

This review suits ecommerce, app, subscription and B2B startup teams comparing creative delivery. Use it when a large asset library has produced little learning, or someone wants to stop an ad because it received few impressions. Do not use it to rank creative quality, certify an A/B winner or diagnose fatigue from a single snapshot.

A media operator owns the roster and export; an analyst owns grain and arithmetic; a creative lead owns the message hypothesis; a business owner approves follow-up. These can be the same person in a small team, but the checks remain distinct.

First gate: establish a comparable roster

Freeze a settled time window, timezone, campaign boundary and reporting grain. Map stable creative IDs to actual versions. Join the export onto the eligible roster so creatives with zero impressions are not silently dropped. Aggregate repeated daily rows only after checking that they are disjoint; duplicate IDs in the tool are rejected rather than guessed away.

Do not add asset-level impression counts if several assets can receive credit for the same ad impression. Use a mutually exclusive ad or combination grain instead, or retain the analysis as an explicitly different metric. Document rejected, paused and late-added ads separately. A zero is observed non-delivery, not proof the creative was given an equal chance.

The review sequence

Reconcile: compare the sum with the same-scope source total and explain omissions. Stop if the denominator cannot be reconciled.
Describe: report total volume, roster size, delivered size, zero count, maximum share and effective count together.
Investigate: inspect approval dates, targeting, placement eligibility, bidding and adaptive serving before offering a creative explanation.
Decide: fix a data defect, gather missing evidence or design a controlled follow-up. Do not force equal live delivery simply to improve the diagnostic.

Synthetic review: five listed, four delivered

Hook A receives 7,000 impressions; B, C and D receive 1,000 each; E receives zero. Total is 10,000 and the largest share is 70%. Squared shares sum to 0.52; the reciprocal is about 1.9231. This distribution is as concentrated as approximately 1.92 equally delivered categories, despite five listed IDs.

The analyst records descriptive concentration, not failure. The operator then finds E was pending approval for most of the window. E cannot support a quality verdict. A's high exposure may be consistent with adaptive serving, but this snapshot cannot establish why it happened. The creative lead retains the untested hypothesis and proposes an eligibility-matched follow-up for human review.

Decision criteria and limits

Hold the review when the reporting grain overlaps, IDs changed without version mapping, totals fail reconciliation or the roster is incomplete. Proceed to human review when those contracts are explicit, even if concentration is high. There is no universal pass mark for effective creative count. A delivery objective may legitimately concentrate impressions; a learning objective may need a separately designed experiment.

Effective count is not effective statistical sample size. Impressions can repeat across people; equal shares do not prove randomization or adequate power. A change across windows can reflect placement mix or eligibility rather than fatigue. Record unresolved explanations in the worksheet instead of manufacturing a score.

Sources checked October 8, 2026: Google creative-testing guidance, October 1; Google ad rotation; scikit-bio inverse Simpson definition. The review workflow and marketing interpretation are original synthesis, not vendor endorsement or a statistical test.