Smart+ Creative-Level Reporting: How to Read Asset ID Data and Run A/B Variable Tests

Since the Smart+ upgrade on September 30, 2026, you can finally see performance data for a single creative asset — not just the ad group, not just the campaign, but the individual material ID. And most teams opened that new report and immediately made the same mistake: they read it like a leaderboard. “Asset #4 wins, asset #9 loses — make more like #4.”

That is how winning structures get lost. Asset-level data is not there to crown individual clips. It is there to tell you which structure wins — the skeleton of hook, pacing and shots underneath the pixels. Read it that way and the report becomes a production brief. Read it like a leaderboard and you will keep cloning one lucky clip until it fatigues.

This guide covers what the upgrade actually gives you, how to design a same-structure variable test, and the four most common ways people misread the numbers.

What the September 30 Upgrade Actually Gives You

Three changes matter for creative testing:

1. Automation in three tiers — full automation, partial, and manual — across four modules: audience targeting, budget allocation, placements, and creative. You can hand the auction the boring parts and keep control of the creative, which is exactly where tests are won or ruined.
2. Creative-level reporting by material ID, covering video, image, catalog and text assets. For the first time, the platform’s reporting granularity matches the granularity where decisions are actually made.
3. Up to 30 asset groups per ad group, which turns the ad group into a real testing container instead of a junk drawer.

Why does this matter? The industry consensus for years has been that more than half of auction outcomes trace back to the creative itself. Until September, your reporting unit was the ad group — a container that averages away the only signal you cared about. Now the instrument matches the physics.

Build the Test Matrix: One Structure, Many Controlled Variables

The efficient way to use 30 asset groups is the replica method: pick a video that already performed for someone else’s product, keep its hook logic, pacing and shot sequence, and map your product into the same skeleton. Then vary exactly one axis at a time.

A workable matrix for a single structure family:

3 hooks × 2 scenes × 2 angles = 12 assets, all sharing one skeleton. That last part is the point: because the structure is constant, differences in the report can be attributed to the variable you changed — not to five things changing at once.

Tooling compresses the production side. Viral Video Remake takes a reference clip plus your product photos and returns a new video with the same structure and your product in it, so a 12-asset matrix is an afternoon, not a sprint. Collect references with Media Download, and if you want to study exactly why a reference works, pull its script and subtitles with Speech to Text and mark the beat changes at 0–3s, 3–8s, 8–15s.

One discipline rule: label every asset before launch — structure ID, hook type, scene, angle. If your file names don’t encode the test design, the new report gives you answers to questions you can’t ask.

Read the Report by Label, Not by ID

When the first week of data lands, resist the leaderboard instinct. Instead:

1. Aggregate by variable. Average CPA and hook-rate across all “curiosity gap” hooks vs. all “result first” hooks. A hook that wins across three different scenes and two angles is a finding. A hook that wins in one scene is a hypothesis.
2. Require siblings. Treat a structure as validated only when at least 2–3 assets sharing that skeleton land in the top half of the report. One asset is an anecdote; a pattern across variants is a structure.
3. Scale the structure, not the clip. Once a skeleton validates, the move is not to raise budget on that one asset — it is to generate the next batch of variants on that skeleton (new hooks, new scenes, new angles) and let the top asset ride while its siblings compound the finding.
4. Feed it back into production. The output of the report should be a written brief: “structure X with direct pain-callout hooks and domestic scenes wins” is worth more than any single winning ID.

Four Ways to Misread Asset-Level Data

1. One lucky spike is not a structure win. Early distributions are noisy, and a novelty effect can carry an asset for 48 hours. Before declaring a winner, check whether its structural siblings show the same direction.

2. Low spend ≠ losing. Auction systems concentrate delivery on early performers; an asset with thin spend is untested, not disproven. Judge assets only above a minimum spend or impression floor, and start tests with even budgets so the comparison is honest.

3. Confounded variables. If your new variant changed the hook and the scene and the offer framing, the report cannot tell you what won — and neither can any tool. Change one axis per group. This is the entire reason to run same-structure matrices instead of “creative sprints.”

4. Fatigue masquerading as failure. When a winner decays, that does not invalidate the structure — the audience may just be done with that specific execution. The professional move is a single-variable refresh: same skeleton, new first three seconds. The AI Video Editor is useful here for swapping hooks and cuts without rebuilding the whole asset.

Conclusion: Turn the Report Into a Production Brief

The Smart+ upgrade handed you an instrument that reads at the same resolution where creative decisions are made. Use it well and the loop is short: label assets before launch, aggregate by variable instead of staring at IDs, demand sibling evidence before declaring a structure winner, and refresh winners with single-variable changes instead of panic rewrites.

Do this and “which asset won” stops being the question. The question becomes “which structure wins, and what is the next variant to prove it” — which is the only question that scales into Q4. If you want to run the matrix-building part faster, start at the VEONIB dashboard and ship your first same-structure variant set this week.