Peak season on Meta: why constant edits cost more than the auction, and what Andromeda changed
On 21 October 2026 Meta published peak-season advice: money is lost not in the auction but through constant setting changes. Here's the 50-conversions-a-week rule, the formula for how many ad sets your budget supports and what the Andromeda update changed.
What does Meta advise for peak season?
Meta's main point: at peak, budget is lost to frequent changes in optimisation, bidding and attribution, because every major edit restarts learning. The benchmark is roughly 50 conversions a week per ad set; below that, Meta says, the system is guessing. In Meta's analysis such ad sets ran about 28% more expensive per purchase, though Meta doesn't disclose the sample or period. Hence the formula: weekly budget ÷ (cost per result × 50) is the most ad sets your budget can support.
What is Andromeda and what changed?
Andromeda is the first stage of Meta's ad recommendation system: from all eligible ads it picks candidates in milliseconds, which are then ranked in detail and sent to auction. On 28 October 2026 Meta described the new version: over a billion ads per second on a single GPU server, about 100 times more than in 2024, with about 300 times less compute per evaluation. Meta attributes the rebuild to generative tools helping advertisers create hundreds of millions of new ads. According to Meta, Andromeda has delivered +2.4% conversions on Facebook since 2025, with no baseline or method given.
What does it mean for advertisers?
Andromeda adds no new settings or reports, so there's nothing to switch on. The logic matters, though: an ad that isn't selected as a candidate never reaches the auction. The peak-season advice comes down to stability: fewer ad sets with enough conversions, and fewer, well-considered edits.
What not to do at peak
Don't add new ad sets at peak if your budget can't give each about 50 conversions a week. Don't change optimisation, bidding and attribution one after another: each edit means a new learning phase. Don't judge changes against last month — cost per result rises at peak anyway.
What to do now
- Check your biggest campaign: is it optimising for what actually makes the business money? If not, change it once and leave it.
- Use the formula to see how many ad sets your budget supports, and merge the extras.
- Write a three-line brief for each campaign: what to optimise for, how hard to bid, how to count results.
- Bundle all edits into one, so you go through one learning phase instead of several.
- Measure with an A/B test or Conversion Lift, not by comparing with last month.
Metriverra shows spend and results for every Meta campaign daily, so you can get the cost per result for the formula and see how a campaign behaves after an edit.
In short
- At peak, money is lost to constant edits, not the auction.
- Benchmark: ~50 conversions a week per ad set.
- Formula: weekly budget ÷ (cost per result × 50) = maximum ad sets.
- Andromeda picks candidates before the auction; no new settings.
- Meta's figures (28%, +2.4%) come without a published method.
Sources: Meta for Business — Peak Season 2026: You Don't Beat The Auction. You brief it. (21.10.2026) · Engineering at Meta — Meta Andromeda (28.10.2026)
