We predicted the winning ad
before spending a single rupee.
4 creatives. A synthetic panel simulation on Aug 20. The same 4 ads live on Meta two days later. Splitroom's champion matched Meta's top performer — and the ranking held.
| Creative | Duels Won | Avg Preference | Result |
|---|---|---|---|
| B | 2 / 3 |
64%
|
Champion |
| C | 2 / 3 |
58%
|
Runner-up |
| D | 1 / 3 |
48%
|
— |
| A | 0 / 3 |
26%
|
— |
| Creative | Impressions | CPM (₹) | Spend (₹) | Link CTR |
|---|---|---|---|---|
| B | 2,193 | 403 | 884 | 2.42% |
| A | 3,819 | 235 | 896 | 1.75% |
| C | 9,197 | 97 | 891 | 0.83% |
| D | 10,205 | 87 | 888 | 0.79% |
Meta's auction served C and D to cheaper 18–24 audiences (CPM ₹87–97), while B and A were pushed into harder, more competitive audiences (CPM ₹403 and ₹235). B fought the most expensive audience of the four — and still won with the highest CTR. At equal impressions, B's lead over C and D only widens.
If all four creatives had received equal impressions, here's how many clicks each would have generated — using their actual CTRs from Meta.
| Creative | Link CTR | Clicks @ 10k Impressions |
|---|---|---|
| B | 2.42% | 242 |
| A | 1.75% | 175 |
| C | 0.83% | 83 |
| D | 0.79% | 79 |
B generates 3× the clicks of C and D at equal impressions. The delivery gap doesn't change who wins.
Round 1: Splitroom Agrees with Meta.
Creative B won the synthetic panel on Aug 20 — and won on Meta two days later with a 2.42% CTR, nearly 3× the performance of C and D. The top-of-bracket prediction was correct before a single rupee was spent.
One asterisk: Splitroom ranked A last (0/3 duel wins). Meta ranked A second. The panel penalised A's comparison-chart format in head-to-head duels, but real scroll rewarded its instant scannability. B's win was clean — A's placement is something we're watching across future rounds.
Panel calibration matters
Round 1 used a broad panel — nearly all age groups and income levels. The synthetic audience skewed young and budget-heavy, which may not match ANTAR's actual buyer. ICP-matched panels should improve accuracy in future rounds.
Audience composition is a confound
Meta's auction delivered C and D to cheaper audiences and A and B to pricier ones. Equal impressions don't mean equal audiences. Despite this, B still won — a sign the signal was strong enough to survive delivery noise.
Duels vs. the feed are different tests
A ranked last in Splitroom duels but second on Meta. Head-to-head duel scoring penalises creatives that are solid but not exciting. In the feed, scannability matters more — a chart communicates in 0.5 seconds without needing to be read.
The confidence flag was accurate
Splitroom flagged: "if your buyers are Brand Devotees or Careful Researchers, this verdict isn't your verdict." A was the top pick for both segments. On Meta — where people encounter an unknown brand cold — they behave exactly like Careful Researchers.