MUD\WTR · Internal
press enter
Creative Analytics · Northbeam · 12 months

Which signals actually
predict a winning ad?

We took every ad signal you've asked about — CTR, cost-per-engagement, cost-per-3-sec-view, thumbstop, % new visits, and new-customer % — and tested each one against the only outcome that matters: does the ad acquire customers at a low CAC? Across ads and 12 months, so it's signal, not 14-day noise.

Window Jul 2025 – Jun 2026 (12mo) Ads ≥ $1k spend Attribution Northbeam · Clicks-only · per-ad CAC
The Verdict

The signals that find gems are about new audience, not creative flash.

The two metrics that best separate a low-CAC ad from a high-CAC one are % new visits and new-customer % — i.e. whether the ad actually brings in new people. The pure attention metrics are weak or worse: thumbstop moves CAC only ~15%, cost-per-engagement and cost-per-3-sec-view show no reliable link at all, and CTR is actively misleading on its own. If you want to spot a winner early, watch who the ad brings in — not how many people it makes pause.

Ranked
Signal scorecard

Each signal's best-third vs worst-third spend-weighted CAC, the gap between them, and the per-ad rank correlation. Ranked by how reliably the signal flags a low-CAC ad.

SignalBest-third CACWorst-third CACCAC gapRank ρpVerdict
Signal by Signal
Does each one separate winners from losers?

Ads split into thirds by the signal (best / mid / worst), then the spend-weighted CAC of each group. If the signal works, the green (best) bar sits well below the red. Watch CTR break this rule.

The Trap
Why CTR lies to you

Per individual ad, higher CTR looks like it tracks lower CAC (rank ρ = −0.34). But weight by spend and it flips: the highest-CTR third of ads carries a $101 CAC, while the lowest-CTR third sits at $74. Why? Your lowest-CTR spend is concentrated in high-intent placements (branded search, retargeting) that convert cheaply, while high-CTR broad video buys cheap clicks that convert worse.

Takeaway: CTR's meaning flips depending on ad type, so it's useless as a standalone winner-signal. Cheap clicks ≠ cheap customers.

The Evidence
Top 20 ads by spend

Every signal and the CAC outcome for the 20 biggest-spend ads. Scan the % new visits and new-customer % columns against CAC — the relationship is there; the attention columns are noise.

AdTypeSpendCTRCost/EngCost/3sThumb%NewNewCust%CAC
So What
How to use this

01 Judge new ads on new-audience signals.

% new visits is the cleanest early tell of a low-CAC ad and isn't gameable by a flashy hook. Put it on the new-ad scorecard above thumbstop and CTR.

02 Demote the attention metrics.

Cost-per-3-sec-view and cost-per-engagement carry no reliable CAC signal; thumbstop is weak. Use them to diagnose a creative, never to decide scale or kill.

03 Never scale on CTR alone.

It flips direction by ad type. A high-CTR broad-prospecting ad is not a buy signal unless its new-customer conversion holds up.

04 Treat new-customer % as a vital sign, with eyes open.

It separates CAC best of all (−33%), but it shares "new customers" with the CAC formula, so it's partly definitional. Real, but not an independent crystal ball.

Read This Before Quoting It
Caveats & method
  • Per-ad CAC = Northbeam clicks-only, cash, spend ÷ first-time transactions for that ad. Absolute levels are directional; the relationships are the finding.
  • Rates are recomputed from counts (clicks = CTR×impressions, etc.) summed across the 12 months per ad — not averaged — so the aggregates are correct.
  • Populations differ: thumbstop and cost-per-3-sec-view only exist for video ads (n≈542); CTR, cost-per-engagement, % new visits and new-customer % cover all ads (n≈1,000). Spearman ρ uses ranks, robust to the skew.
  • New-customer % is partly mechanical — new customers appears in both it and CAC, so some of its −33% gap is definitional, not predictive.
  • Association, not proof of cause. A high signal doesn't guarantee a winner; it shifts the odds.