Creative Testing Cadence in the Andromeda Era: Signals, Not a Calendar
TL;DR
Why fixed creative refresh calendars no longer work post-Andromeda, and the fatigue-signal framework (frequency, CTR, CPA) that should drive cadence instead.
Meta's Andromeda retrieval engine changed how creative gets matched to audiences at a deeper level than most advertisers have fully adjusted for. The old mental model - find a winning ad, run it until it fatigues, then find the next one - assumed a slower delivery system doing coarse audience matching. Andromeda's retrieval-based approach evaluates far more creative-audience combinations, far faster, which means the practical question isn't just "what should our next ad be" but "what testing cadence actually keeps pace with how fast the system can now separate winners from losers."
Here's the one-paragraph version: post-Andromeda, creative refresh cadence should be driven by fatigue signals (rising frequency, softening CTR, climbing CPA) rather than a fixed calendar, and testing volume needs to be genuinely continuous rather than campaign-based, because the delivery system itself is evaluating creative-audience fit continuously.
What changed with Andromeda, in practical terms
Andromeda is a retrieval engine that dramatically expanded the number of ad-to-user matching evaluations Meta's system can make, moving from a smaller candidate pool per auction to a much larger one. In practice, this means the system can identify which specific creative resonates with which specific micro-segment of your audience faster and more precisely than before - which is good news if you're feeding it a genuinely diverse creative pool, and bad news if you're still running one or two ads and expecting broad reach from narrow input.
Why fixed refresh schedules no longer make sense
A calendar-based refresh cadence (say, "new creative every two weeks") assumes fatigue happens on a predictable schedule. It doesn't - a strong creative-audience match can run productively for a month before showing real fatigue, while a weaker match might fatigue within days. Refreshing on a fixed calendar means either replacing content that's still working (wasting production effort) or leaving fatigued content running too long (wasting ad spend). The fix is watching the actual signals instead.
The fatigue signals worth watching
- Frequency - the average number of times a user has seen the ad. Rising frequency on a previously stable CPA is an early warning sign, not yet a crisis.
- Click-through rate trend - a gradual CTR decline on an otherwise-unchanged ad usually means the audience has started to tune it out.
- Cost-per-result trend - the trailing signal; by the time CPA climbs meaningfully, frequency and CTR have usually already been signaling it for a while.
| Signal | What it means | Action |
|---|---|---|
| Frequency rising, CPA stable | Early fatigue warning | Prep fresh variations, don't panic yet |
| CTR declining, frequency rising | Active fatigue in progress | Introduce fresh creative now |
| CPA rising meaningfully | Fatigue has already cost you performance | Replace immediately, review what's next in the pipeline |
What "continuous testing" actually requires operationally
Reacting to fatigue signals only works if there's already a pipeline of fresh creative ready to go - which means creative production has to run continuously in the background, not get triggered reactively once an ad has already started declining. In practice this means maintaining a rolling brief and production calendar (for both AI-generated test variations and genuine UGC, per the hybrid approach we run for testing new angles) so a replacement is ready before the old creative has meaningfully fatigued, not two weeks after.
How many active creative variations is enough?
There's no universal number, but a useful heuristic: enough distinct angles running that no single ad is carrying the majority of an ad set's spend and impressions. If one creative is responsible for 70%+ of delivery within an ad set, the system has too narrow a pool to route around it once it starts fatiguing - widen the pool before that becomes a problem, not after.
A worked example: catching fatigue before it costs money
Take an account running a winning ad at a stable ₹450 cost-per-purchase for three weeks. In week four, frequency climbs from 2.1 to 3.6 while CTR softens slightly - CPA hasn't moved yet, sitting at ₹460. On a calendar-based refresh schedule (say, a fixed four-week cycle), this ad might already be up for its scheduled replacement regardless, which happens to work out fine here - but only by coincidence, not because the schedule was actually reading the signal.
Now take the same scenario but the winning ad happens to be only two weeks old when frequency starts climbing. On a fixed four-week schedule, it has two more weeks to run "as scheduled" - and by the time the scheduled refresh arrives, CPA has climbed to ₹650, well past the point where fresh creative should have been introduced. Watching frequency and CTR directly, rather than trusting the calendar, would have triggered a fresh variation two weeks earlier, before the ₹200 per-purchase cost increase actually happened. The calendar wasn't wrong every time - it was just disconnected from what was actually happening in the account, and the gap only becomes visible when the timing doesn't line up by luck.
How Balistro builds this into account management
Every Meta Ads account we run tracks frequency and CTR trend at the ad level as a standing part of account management, not a special project - so refresh decisions happen off real signal, with fresh creative already queued through our creative-strategy pipeline before fatigue becomes a visible cost problem.
FAQ
What frequency level should trigger concern?
There's no single universal threshold since it varies by audience size and campaign objective, but a frequency climbing past 3-4 within a stable audience, especially alongside softening CTR, is worth treating as an active signal rather than ignoring.
How is Andromeda different from Meta's previous ad delivery systems?
Andromeda is a retrieval-based system that evaluates a much larger pool of ad-to-user matches per auction than earlier systems, allowing more precise creative-to-audience matching - which rewards advertisers who supply a genuinely diverse creative pool and penalizes those relying on one or two ads for broad reach.
Should small-budget accounts also refresh creative this frequently?
The same fatigue-signal logic applies, but smaller budgets naturally generate signal more slowly, so the practical refresh cadence tends to be less frequent simply because it takes longer to see a clean fatigue signal at lower spend levels.
Is more creative volume always better?
Only up to the point where each variation can still get a meaningful amount of delivery to produce a real signal - spreading a limited budget across too many variations at once can mean no single one gets enough data to evaluate properly.
Creative cadence in the Andromeda era is a signal-response discipline, not a calendar exercise. If you want help building a genuinely continuous creative-testing system into your account, talk to Balistro. - Naman Khetawat, Balistro
