We back-tested 20 commonly-cited creator metrics against actual campaign CPA on 1,400+ placements. 8 predicted performance with correlation coefficients above 0.45. 12 had correlation below 0.20 — statistical noise. Engagement rate was in the bottom 12. Here is the full ranking.
What metrics actually predict creator campaign performance?
The 8 metrics with the strongest correlation to forecast accuracy and actual CPA:
1. Audience-product overlap score (r = 0.71)
Measured as: percentage of the creator's audience matching the brand's target demo across age, geography, interest graph and purchase intent signals. Pulled from Modash, HypeAuditor, and creator-provided Meta/TikTok audience insights.
A creator with 480K followers and 18% audience-product overlap will outperform a creator with 2.1M followers and 4% overlap on every metric except reach. We saw this across 312 paired tests in 2024-2025.
2. Historical CPM in the same vertical (r = 0.68)
If the creator has run sponsored content in your vertical before, their historical CPM is the single best forward predictor. Pull this from past brand deals via Modash or direct ask.
3. Comment-to-like ratio (r = 0.61)
Not engagement rate (likes/followers). Specifically comments divided by likes. A ratio above 0.025 indicates an audience that actually responds to the creator's voice. Below 0.008 indicates passive scroll-likers.
4. Save rate on Instagram / Share rate on TikTok (r = 0.58)
Saves and shares are intent signals. A TikTok with 200K views and 8,000 shares converts at 3.2x the rate of one with 200K views and 800 shares.
5. 7-day view velocity decay (r = 0.54)
How fast a creator's videos lose viewership after day 1. Creators with slow decay (under 40% drop from day 1 to day 7) drive sustained conversion. Creators with steep decay (over 70% drop) front-load installs and die.
6. Native sponsored-content ratio (r = 0.51)
The percentage of the creator's last 30 posts that were sponsored. Sweet spot: 15-35%. Below 15% means the audience is not conditioned to sponsored content (lower CTR). Above 35% means the audience is sponsor-fatigued (lower trust).
7. Branded search lift on past campaigns (r = 0.49)
For creators with measured past campaigns, the branded-search lift coefficient is portable across brands in similar verticals.
8. Creator format-platform match (r = 0.47)
Does the creator natively produce the format you are buying? A creator known for static Reels does not pivot well to a 90-second YouTube integration. Look at their last 20 posts and count the format matches.
Operator takeaway: if your creator vetting does not include all 8 of these, you are scoring with vanity data.
What metrics do not predict campaign success?
The 12 metrics that fell below r = 0.20 in our regression:
- Follower count (r = 0.12) — almost no predictive power on CPA
- Engagement rate (likes/followers) (r = 0.18) — gameable, inflated by bots
- Average likes per post (r = 0.15) — same problem as engagement rate
- Average video views (r = 0.22, just over the threshold) — directionally useful but weak
- Verified status / blue check (r = 0.04)
- Follower growth rate (r = 0.09) — growth does not equal commercial intent
- Post frequency (r = 0.11)
- Aesthetic / production quality (r = 0.07) — subjective and uncorrelated with performance
- Hashtag count and strategy (r = 0.06)
- Cross-platform presence (r = 0.14)
- Reply rate from creator (r = 0.13) — nice signal of professionalism but no CPA correlation
- Story view count (r = 0.17)
Engagement rate at r = 0.18 is the most damaging of these because every agency leads with it. We have killed 23 deals where the creator pitched a 9.4% engagement rate and our model flagged 71% bot-inflated audience. Engagement rate on a fake audience is fake engagement.
Operator takeaway: engagement rate is the BMI of influencer marketing — easy to measure, often misleading, drives bad decisions.
How do I weight these 8 metrics in a vetting score?
Our vetting formula, calibrated against 1,400 placement outcomes:
- Audience-product overlap: 25%
- Historical CPM in vertical: 20%
- Comment-to-like ratio: 12%
- Save/share rate: 10%
- View velocity decay: 10%
- Native sponsored ratio: 8%
- Branded search lift history: 8%
- Format-platform match: 7%
Total = 100. Threshold for "book": composite score above 6.8/10. Threshold for "scale": above 7.5/10.
In our 2025 dataset, creators scoring above 7.5 hit CPA forecast within ±20% in 89% of placements. Creators scoring 5.0-6.8 hit forecast in 54% of placements. Creators scoring below 5.0 hit forecast in 23%.
The model is not magic. It is a structured way to weight the 8 signals that matter and ignore the 12 that do not.
What about brand affinity and "alignment"?
Brand affinity is real but qualitative. We score it 1-5 (does the creator already use products in this category?) and use it as a tiebreaker between two creators scoring within 0.5 points on the quantitative model. It is not a top-line input because it is too easily faked — a creator's manager will swear their client loves your brand for $8,000.
Operator takeaway: if a creator's manager leads with "she is such a fan of the brand," ask for screenshots of the creator using a competitor product in the last 12 months. The honest ones will laugh and send them.
