Account-Level Learning vs Ad Set Learning Explained

Quick Answer

Meta runs two parallel learning systems: account-level priors that persist across all your campaigns and update slowly, and ad set-level posterior models that focus on a specific ad set fingerprint and update fast. New ad sets inherit account priors as their starting point and refine them with their own data. Account-level learning is what makes mature accounts converge faster on new campaigns.

The Mechanism Explained

The two systems work at different time scales:

Account-level priors track:

• Historical conversion rates by event type


• Average eAR by creative type and vertical


• Audience response patterns across all campaigns


• Page reputation and trust signals


• Long-term complaint and feedback rates

These update over weeks/months and persist indefinitely (with slow decay). They're what makes a 5-year-old ad account behave differently from a 5-day-old one with the same creative.

Ad set-level posteriors track:

• This ad set's specific eAR profile


• Bandit state for ads within this ad set


• Audience response specific to this targeting + creative combination


• Learning phase exit confidence

These update over days and reset on significant ad set edits.

When you launch a new ad set, the system queries account-level priors as a starting point. A mature account hands the new ad set rich starting beliefs ("conversion rate is probably around 2.1% based on similar ad sets you've run"). A new account hands it generic vertical defaults ("conversion rate is probably around 1.4% based on the category average").

This is why two identical ad sets in two different accounts behave differently. The account context shapes the priors, which shapes the early-stage delivery, which shapes the rate of learning convergence.

Practical Implication

Diversify the campaigns running in your account. Even if you don't need them strategically, having 3-5 campaigns active helps the account-level priors stay rich and recent. A single-campaign account has thinner priors than a multi-campaign account at the same total spend.

Also: don't rotate accounts. Each new account starts at zero priors. Stay in one account for the long-term compounding benefit.

Real Numbers

• Account-level prior decay rate: half-life ~6-9 months
• New ad sets in established accounts exit learning 2-3 days faster than equivalent in new accounts
• Multi-campaign accounts have 15-25% lower average CPA than single-campaign accounts at matched spend

FAQs

Q: Can I see account-level priors?
Not directly — they're internal model state.

Q: Does deleting old campaigns hurt account priors?
No — historical data is preserved for prior calculation.

Q: Will running unrelated campaigns dilute my main campaign's priors?
Slightly — but the diversity benefit usually outweighs the dilution.

Q: Does ad account pause affect priors?
No — paused accounts retain priors.

Q: How long until a new account "matures"?
Roughly 60-90 days of consistent activity.

Pix-Vu

Account priors compound over time, but only if your delivery stays clean. Pix-Vu helps you keep creative quality consistent so your account doesn't pick up negative priors that slow down future campaigns — at https://pix-vu.com.