Creative Analytics: What to Track and What to Ignore
A digital marketing guide to creative analytics: the four video metrics that change decisions, the vanity numbers to drop, and how to tag creative attributes.
A media buyer I know keeps a dashboard with 31 metrics on it. Asked which ones had changed a decision in the last quarter, she named three. The other 28 exist because somebody once asked for them, or because the platform surfaced them by default, and now they take up the top of the screen and the first ten minutes of every review call.
Creative analytics has a specific job: tell you what to make next. Not whether the business is healthy — that's finance. Not which channel deserves budget — that's attribution. Just: given what we've published, what should the next batch look like? Almost every metric that doesn't answer that question is noise dressed as rigor.
This post is a subtractive one. It covers the small number of creative metrics that reliably change production decisions, the ones that don't, and the tagging discipline that turns a pile of published videos into something you can actually query. It's written for digital marketing teams producing weekly video, not for analysts building models.
The four metrics that change what you make
Hook rate — the share of impressions that reach roughly three seconds. This is the single most actionable creative number, because it isolates the opening from everything else. If hook rate is low, nothing downstream matters; the rest of the video was never seen. It's also the fastest metric to read: you can judge it in a day, on organic or paid.
Hold rate — of the people who got past three seconds, how many reach the halfway mark. This separates "bad hook" from "bad middle," which are completely different production problems. A video with a high hook rate and a collapsing hold rate has a promise it doesn't keep — usually a hook that's disconnected from the content.
Completion rate, banded by length. Raw completion is meaningless across lengths; a 12-second video will always beat a 45-second one. Band your videos by duration and compare within bands. What you're looking for is the length at which your category's completion falls off a cliff.
Cost per meaningful outcome — per click, per lead, per purchase, whichever is closest to money and has enough volume. This is the only one that ties creative to the business, and it's also the noisiest, so it should be read over weeks and not days.
That's it. Four numbers, each with a distinct decision attached.
What to stop looking at
| Metric | Why teams track it | Why it doesn't help |
|---|---|---|
| Raw view count | Feels like reach | Definitions differ per platform; a "view" can be 1 second |
| Follower growth | Feels like progress | Lags creative changes by weeks; confounded by everything |
| Average watch time (unbanded) | Sounds rigorous | Dominated by video length, not video quality |
| Likes | Visible, immediate | Weak relationship to any commercial outcome |
| Shares (in isolation) | "Virality" | Useful as a tiebreaker, misleading as a target |
| Impressions | Big number | An input you bought, not a result you earned |
| Engagement rate (blended) | One tidy number | Blends metrics that move in opposite directions |
None of these are lies. They're just not levers. If a number goes up and you can't name the production change it implies, it belongs in a monthly summary, not a weekly creative review.
One nuance on shares: share rate is genuinely predictive in some categories, particularly where the content is useful or funny enough to send to a specific person. Track it, but as a secondary signal next to hook rate rather than a headline.
Tag your creative, or you're just guessing
Metrics tell you which video won. Tags tell you why. Without attribute tagging, a year of analytics produces a folder of winners and no rules.
Tag every published video with a small, fixed vocabulary — five to eight fields, each with a closed list of values:
- Hook type: problem, result, question, pattern-break, direct-address
- Format: talking head, demo, b-roll montage, screen capture, slideshow
- Subject: person, product, text card
- Voiceover: yes / no
- Length band: 0–15s, 16–30s, 31–60s, 60s+
- Offer: which promotion or none
- Production route: which generation approach produced it
Closed lists matter more than the specific fields. Free-text tags rot within a month. Once these exist, the question changes from "which video did well?" to "do problem hooks beat result hooks for us?" — which is a question with a reusable answer. The hook side of that vocabulary is worth reading alongside brand video hooks: the first 3 seconds.
Reading the numbers without lying to yourself
A few corrections that catch most bad conclusions:
- Normalize by placement. A Reels placement and an in-feed placement produce different hook rates for identical creative. Compare like with like or you're measuring placement.
- Watch the age of the creative. Performance decay over a creative's life is normal and predictable; a video that's been running six weeks isn't worse than a new one, it's tired. That's a scheduling issue, covered in creative fatigue and refresh cycles.
- Beware the small-denominator winner. The video with the best conversion rate often has 40 impressions. Set a minimum volume threshold before anything enters the leaderboard.
- Separate audience effects. Retargeting audiences flatter every metric. Rank creative inside audiences, never across them.
- Don't average across platforms. A blended engagement rate across five platforms describes nothing that exists. Compare platform to platform only for trend direction, and keep each platform's baseline separate.
A weekly review that takes 20 minutes
Structure the review around decisions, not around the dashboard's layout.
- Two minutes: total published, total spend or reach. Context only.
- Six minutes: hook rate leaderboard, filtered to the last 14 days and a minimum impression floor. Name the top two and the bottom two.
- Five minutes: hold rate for anything with a strong hook. Identify videos whose middle is losing people.
- Four minutes: attribute rollup — which hook type and format won this period.
- Three minutes: decide next batch. Three videos using the winning attribute, one control that deliberately breaks it.
That last step is the entire point. If a review ends without a production instruction, it was a status meeting.
The control video matters more than it looks. Doubling down on the winning attribute forever narrows your creative until it stops working and you don't know why. One deliberate outlier per batch keeps a live comparison running and occasionally finds the next thing.
Where analytics stops and testing starts
Creative analytics is observational. It tells you what happened to videos you happened to publish, under conditions you didn't control. It's excellent for generating hypotheses and terrible for confirming them, because everything is confounded — the videos you made with more effort were also the ones you promoted harder.
So the loop is: analytics generates a hypothesis, a controlled test confirms it, the confirmed rule becomes a production default. Skipping the middle step is how teams end up with a folder of superstitions. If you're not sure how to build that middle step, A/B testing video creative properly covers the mechanics.
FAQ
What's a good hook rate?
It varies enormously by platform, placement, and category, so external benchmarks are close to useless. Build your own baseline from 20–30 published videos, then treat anything more than a fifth above that baseline as a signal worth copying.
Should I track different metrics for organic and paid?
Mostly the same top-of-funnel metrics — hook rate, hold rate, completion by band — because they describe the creative. The difference is downstream: paid gets cost per outcome, organic gets share rate and profile visits. Don't compare the two sets to each other.
How many videos do I need before analytics is meaningful?
Roughly 20–30 published pieces with consistent tagging before attribute-level patterns are readable. Below that you can still review individual videos, but don't draw category rules from five data points.
Is watch time a vanity metric?
Unbanded average watch time is, because it mostly reflects video length. Watch time expressed as a percentage of duration, within a length band, is genuinely useful. The fix is the banding, not dropping the metric.
Who should own creative analytics?
Whoever decides what gets produced next. If analytics lives with an analyst who doesn't brief creative, the insights arrive too late and too abstract to change anything. A creative lead reading four metrics beats an analyst reading thirty-one.
If you want the review to run itself, wire your production into saved workflows so attribute tags are set at generation time rather than reconstructed later — and see how per-post metrics surface in brand video analytics inside Versely.