Creative analytics: Video ad metrics, reports, and decisions

Creative analytics is reading ad results creative by creative, on metrics defined in writing, to decide which ads to scale, iterate, retire, retest, or hold and what the next brief tests.

Teams improve creative performance with analytics by running a weekly loop on performance data: define the metrics, compare creatives fairly, set each creative's status and decision, and brief the next test.

Video ad metrics from attention to cost

Read video results top-down and act on the first funnel layer below the account's median for that platform and placement.

• Attention: Meta reports 2-second continuous video plays and 3-second video plays. TikTok reports 2-second video views per impression session, excluding replays.
• Hold: ThruPlays on Meta are plays to completion or for at least 15 unique seconds. Meta's ThruPlay goal runs at 6 or 15 seconds, and a video shorter than the goal counts when watched to 97% completion. TikTok's 6-second video views count plays of at least 6 seconds in an impression session, full plays of videos under 6 seconds, and plays with an engagement in the first 6 seconds.
• Click: name the click type. Google's Demand Gen asset report does not count clicks that pause a video or open its watch page.
• Conversion and cost: installs, sign-ups, or purchases and cost per result, from one attribution source and window across platforms.

When attention and hold are healthy and cost per result is weak, check the offer, click, and landing page before re-editing the ad.

Label hook rate and hold rate as derived ratios, with the formula in the column header; Starti's hook rate guide gives each platform's formulas.

Shared metric definitions for Meta, TikTok, and YouTube

On a 45-second video, the watch thresholds for a Meta ThruPlay, a TikTok 6-second view, and a skippable YouTube in-stream view are 15, 6, and 30 seconds. YouTube also counts a click on an element of a skippable in-stream ad as a view, and its played-to percentages are shares of viewers who started the player. On Shorts, a view takes 10 seconds or the end of the video, whichever comes first, or a click on the call to action, Like, Comments, or Share.

Meta's played-to counts include plays that skipped ahead to the mark and add a 95% point. Meta and TikTok report each milestone as a count, which the report divides by impressions; YouTube view rate and played-to percentages take separate columns.

Rank creatives across platforms on cost per result alone; attention and hold explain a result within one platform and placement.

How to compare creatives fairly

Meta's A/B testing best practices call for one variable, a test audience kept out of other running campaigns, and 7 to 30 days. Run longer when customers take more than a week to convert. TikTok's split test best practices ask for at least 7 days, a budget that gives a power value of at least 80%, and test group settings that differ significantly.

Google Ads video experiments split traffic evenly, show directional results at 70% and 80% confidence, and call results conclusive at 95%. TikTok split tests name a winning ad group at a 90% confidence level, and Meta's A/B test results show the cost per result for each version. Scale on a named winner or a conclusive result, and retest on a directional one.

Hold creative, audience, and budget steady while a comparison runs. Outside a test, spend decides how much evidence each creative has.

• Same context: one objective, placement, and date range per comparison.
• Identical visuals: Meta groups ads that share key visual and thematic attributes as variations of one creative, with shared learnings and delivery. Variants that differ only in text share one row.
• Minimum evidence: a team-set count of impressions and outcomes per creative, below which a row stays grey. Replay past creatives with a known final decision week by week; the minimum is the outcome count most of them had reached when their weekly status stopped changing.

How to lay out a weekly creative report

Cut by placement first: a creative that wins in Reels and loses in Feed holds two findings that a blended row hides. Starti's guide to creative intelligence covers the tag vocabulary.

Give each creative the first status that applies, in this order, with its paired decision. Launch cohorts group creatives launched in the same week.

• Learning limited: its ad set carries Meta's learning limited flag, set when the ad set is unlikely to get about 50 optimization events in the week after its last significant edit. Decision: hold.
• No delivery: little or no spend while other ads in the ad set took the budget. Decision: retest in a controlled test, or retire and replace it. Google's Demand Gen asset report says to consider replacing an asset that gets zero impressions for several weeks.
• Below minimum evidence: impressions or outcomes under the team's minimum. Decision: hold for another week.
• Fatiguing: beat its launch cohort's median cost per result in its first two weeks, and now has a higher cost per result and lower click-through rate than in those weeks. Decision: iterate with a new version of the proven idea. Starti's guide to ad fatigue lists each platform's fatigue signals.
• Weak from launch: behind its launch cohort's median cost per result since its first full week. Decision: retire it and brief a new concept, offer, or audience.
• Winning: ahead of its launch cohort's median cost per result. Decision: scale.

Example: A weekly creative analytics report template

• Header: the last 7 complete days against the previous 7, platforms and accounts, metric definitions with each view threshold, the attribution source and window, and the minimum evidence per creative.
• Summary: spend share by concept, creatives launched, scaled, and retired, and one line per finding with its decision.
• Rows: one per creative under its concept, with creative ID, opening tag, format, length, placement, and launch date.
• Delivery columns: spend, impressions, and frequency.
• Funnel columns: 2-second plays and plays at 50% per impression, click-through rate with click type, conversion rate, cost per result, and YouTube view rate and played-to percentages.
• Trend column: click-through rate, cost per result, and frequency against the creative's first two weeks.
• Status and decision columns: the status and its paired decision.
• Cuts: the same metrics by concept, opening, format, and placement, with creative count and spend.
• Next brief: finding, supporting creatives, counterexamples, element to change, elements to keep, success metric, and owner.

Keep creative reporting in one file: media buyers act on status and decision, the creative team on the cuts and next brief.

From finding to next brief

A finding is ready for a brief when it names one element to change, the elements held fixed, and the metric that decides the next test. Write one brief per finding, and keep findings from different placements in separate briefs.

Retest any pattern carried by one standout ad before scaling it. Starti's guide to creative testing covers holding a brief steady while one element moves.

Creative analytics in Smart Insight

Smart Insight, part of Starti AI Studio, follows one creative across platforms and campaigns, with its spend, engagement, and video completion in the same record. Its Insights view uses complete-day windows and excludes the current partial day.

Creatives can be grouped by any tagged attribute and compared on the outcome that matters against the rest of the cohort. Each finding shows its supporting creatives, counterexamples, and sample size, with timestamped scenes to inspect. Findings become briefs with the supporting evidence, what to change, and what to test, for the existing creative team or production in AI Studio.

See how Smart Insight turns findings into briefs.

Frequently asked questions

What is creative analytics?

Creative analytics is the analysis of ad results by creative and by element, such as the opening, format, or placement, to decide what to scale and what to produce next.

How do marketing teams use analytics to improve creative performance?

They read every running creative each week within one objective, placement, and date range. Each creative gets a status that sets its decision, and the clearest finding becomes the next test's brief.

How is creative analytics different from creative intelligence?

Creative intelligence starts with tagging each ad's elements, such as opening type or brand entry, and compares tag groups across the ad library. Creative analytics is the weekly report that uses those tags as cuts beside each creative's results, status, and decision.

How does creative analytics relate to creative testing?

A creative test isolates one variable between split groups for a set period. Creative analytics reads every running ad each week, where spend share and placement vary, and chooses the variable the next test isolates.

How often should a creative analytics report run?

Run it weekly. A week matches the 7-day minimum Meta and TikTok recommend for tests. Assign statuses once a week and leave daily numbers to pacing checks.