Your reporting knows that ad 23847_v3 spent ₹41,000 and produced nine leads. It does not know that the ad was a testimonial, that the proof was a chat screenshot rather than a customer story, that the offer was a hesitation-reducer, or that the same structure has been running in your category for eleven months.
That gap is what creative analytics closes. Not "which ad won" — which attributes of that ad won, so the next brief can be written from evidence instead of taste.
This guide covers the unit problem, the attributes worth tagging, the metrics that actually mean something, the tool categories, and how to start without buying anything.
Media analytics vs creative analytics
| Media analytics answers | Creative analytics answers |
|---|---|
| Which campaign spent most | Which hook type earns the click |
| Which ad set converted | Which offer role converts — hook, reward or hesitation-reducer |
| When CPA rose | Which attribute stopped working, and in what order |
| What happened | What to make next |
Both matter. Only one of them produces a brief.
The unit problem
An ad ID is a useless unit of analysis. Two ads with the same ID history can share nothing, and two ads that look unrelated can share the thing that made them work.
So the first job is to stop treating the ad as the unit and start treating its attributes as the unit. Three live ads, decoded the way an analytics layer should decode them — and watch what the decoding surfaces that the spend column never would.
CarDekho leads with the saving: up to ₹40,000 off new car insurance. Decoded, it's a price-focused hook where the offer is the hook, no proof element, aimed at someone who has just bought a car. It ran 6 days — a short, honest life for a number-led static, because the moment the number stops being surprising the ad is finished.
Master AI opens somewhere else entirely: on waking up tired despite the alarm. A problem-solution hook, in Hindi, with the offer — a ₹1 trial — arriving only after the frustration has been named. Decoded, it selects a completely different person: salaried, lower-middle-income, looking for a way out rather than a discount. It ran 4 days.
Two Brothers doesn't open on a problem or a price. It opens with a command and a compliment — moms, your voice is worth hearing — and asks mothers about their children's food, with a free gift attached to taking part. It ran 2 days, the shortest of the three, which is what you'd expect from a lead-gen survey ad with a finite audience.
Three categories with nothing in common commercially — insurance, a self-improvement app, organic groceries — and the decode shows two of them making the same structural move: name the moment before you name the product. CarDekho doesn't. That's a finding you can brief from. "Ad 23847 won" isn't.
The attributes worth tagging
A workable scheme reads each ad along seven dimensions — hook type, offer type, ad type, emotion, social proof type, authenticity type and intended audience — and that's enough to make an account queryable. The ones that earn their place first:
| Attribute | Why it matters |
|---|---|
| Hook type | The single strongest predictor of whether the ad is scrolled past |
| Offer type and its job | Hook, reward or hesitation-reducer — the same discount behaves differently in each role |
| Ad type / format | Testimonial, comparison, listicle, POV, unboxing, offer banner |
| Emotion | What the first second is trying to produce |
| Social proof type | Rating, review quote, customer story, expert, volume claim |
| Authenticity type | Studio-polished, creator-shot, screen-recorded, native-feeling |
| Intended audience | Who the creative selects — often different from who you targeted |
Tag these consistently and your account stops being a list of ads and becomes a dataset you can ask questions of.
One attribute, three executions
Here's why "social proof" as a tag is too coarse on its own. All three of these are proof-led, and they are not doing the same job.
Astrotalk screenshots a chat in which a prediction about buying a home came true. The proof mechanism is the customer's own words, in their own interface, and what it removes is the objection is any of this real. Ran 2 days — and that brevity is characteristic: a specific testimonial is highly persuasive to the few people in exactly that situation and invisible to everyone else.
Plum offers a different kind of proof entirely: Japanese rice water, 99.99% pure niacinamide, acne marks fading in two weeks. No customer, no quote. A testable claim with an ingredient percentage and a deadline attached, removing the objection will it actually do anything. Ran 8 days.
Qubo's proof is a household anecdote — milk packets that kept vanishing from outside the door, until the doorbell. Still running — live 14 days. What it removes is the hardest objection of the three: do I even need this. Nobody searches for a doorbell to stop milk theft, so the ad has to supply the problem before it can sell the solution.
Same tag, three different objections handled, in three different Indian categories, with run lengths of 2, 8 and 14 days. An analytics layer that can't tell these apart will tell you "proof works" — which is not actionable, because everything is proof if you squint.
The metrics that actually mean something
Creative analytics has its own measures, separate from CPA and ROAS:
| Metric | What it tells you | Watch for |
|---|---|---|
| Longevity | How long an ad stayed live | The only public signal available for competitors |
| Copy count | How many near-identical versions run | How hard a brand is backing an idea |
| Fatigue composite | CTR decline weighted most, frequency least | Five levels: healthy, watch, moderate, high, critical |
| Share of spend per creative | Concentration risk | Above 60% on one ad is a single point of failure |
| Attribute win rate | Which hooks and formats convert across ads | Needs consistent tagging or it's noise |
| Refresh rate | New concepts shipped per month | Distinguishes production from strategy |
And the diagnosis layer on top: offer fatigue, burnout starting, high CTR with low sales, costs rising but still working, clicks too expensive, only existing buyers left, wrong audience, cold start failure. Eight named states, each with a different correct response — which is the whole point of naming them.
The attribute isn't destiny — and the tag you compare on decides the answer
A caution that keeps teams honest, and it's the most useful thing in this guide. These two ads look like siblings: Indian D2C, app-first purchase, a price in the headline, a discount in the copy.
Ran 72 days.
Ran 7 days. Ten times the shelf life, on what a coarse tag sheet would record as the same kind of ad.
Now look at what a finer decode says. ShopZilo's hook isn't the discount at all — it's no more waiting, a problem-solution hook about delivery time, with the 50% riding along underneath as a reward. Plix's hook is evidence: week 1 against week 4, a before-and-after. Two genuinely different hook types, which the "offer-led, price in the headline" tag flattened into one.
So the lesson is sharper than "tags don't predict winners". It's that the dimension you choose to compare on determines whether you see anything at all. Compare these two on offer type and they're twins with inexplicably different outcomes. Compare them on hook type and a hypothesis appears: in a convenience category, the delivery promise may outlive the discount. That hypothesis might be wrong — one pair of ads proves nothing — but it's a brief, and the first reading wasn't.
Use attributes to generate hypotheses, not to predict winners. A team that treats the tag sheet as a formula ends up producing the average ad in its category.
The tool categories
| Category | What it does | Buy it when |
|---|---|---|
| In-platform reporting | Meta and Google's own asset-level breakdowns | Always — it's free. Thin on the "why" |
| Creative analytics platforms | Tag your own ads, tie attributes to performance | You have enough volume for patterns to emerge |
| Ad libraries and swipe tools | Competitor creative, searchable and saved | Your gap is inspiration and category awareness |
| Full-stack: research → production → analytics | Decode, generate, publish, measure in one loop | You want the analysis to end in a brief |
| BI on top of exports | Your own warehouse, your own definitions | You have an analyst and unusual questions |
Most teams need the free layer plus one paid tool. The mistake is buying a tagging platform before you have the volume to make tags meaningful — under about twenty new creatives a quarter, you'll learn more by reading your own ads carefully than by charting them.
How to start without buying anything
- Export your last 90 days of ads. Just the creatives and their spend.
- Tag them by hand, once. Hook type, offer's job, format, proof type. An afternoon.
- Sort by longevity, not spend. Ads Manager sorts by money; survival is the more interesting column.
- Find the attribute your winners share. There's usually one, and it's usually not the one you'd guess.
- Re-sort on a different attribute when the first one shows nothing. As the ShopZilo and Plix pair above demonstrates, the wrong comparison dimension hides the finding completely.
- Write next month's briefs from it. Three concepts that share the winning attribute, executed differently.
- Then decide whether you need a tool — by which point you'll know exactly what you want it to do.
Step two is the one people skip and the one that teaches most. Tagging fifty of your own ads by hand is the cheapest creative education available.
Demo: get your own ads decoded, with the fix next to each flag
We'll read 180 days of your account, decode every creative along the seven dimensions, score fatigue across five levels, name the failure mode behind anything declining, and set it beside what your category is running — so you leave with the attribute your winners share and three briefs built on it.
Book a demo, or see how creative intelligence works. If you'd rather start with the numbers, the CPM calculator and ROAS calculator are open to use.
FAQs
What are the best tools for creative analytics?
It depends on which layer you're missing. Meta and Google's asset reporting is free and worth exhausting first. Beyond that: creative analytics platforms for tagging your own ads, ad libraries for competitor creative, and full-stack tools where decoding feeds production. Volume decides — below roughly twenty new creatives a quarter, manual tagging beats any of them.
What is creative analytics?
Analysing the content of ads rather than the delivery: hooks, offers, formats, proof types and audiences, tied to performance, so you can brief the next ad from evidence.
How is it different from normal ad reporting?
Normal reporting tells you which ad ID spent and converted. Creative analytics tells you which attributes those ads shared, which is the part that transfers to the next brief.
What should I tag my ads with?
Start with four: hook type, the offer's job, format and proof type. Add emotion, authenticity and intended audience once the first four are consistent. Consistency matters more than completeness.
Why do two similar ads perform completely differently?
Usually because they are not as similar as your tags say. The 72-day and 7-day pair in this post share their offer type and differ on hook type — compare them on the wrong dimension and the result looks random; compare them on the right one and a hypothesis appears.
How do I measure creative fatigue?
As a composite, not a single metric: CTR decline carries the most weight, frequency the least, read on a five-level scale so "watch" triggers a brief and only "high" triggers a replacement.
Can I see competitors' creative performance?
Not their spend or conversions — nobody can. You can see how long each ad has run and how many near-identical copies exist, which together are a reliable read on conviction.
How much creative volume do I need for this to work?
Enough for patterns to survive noise: roughly twenty new concepts a quarter before charts beat careful reading. Below that, tag by hand and trust your eyes.