The Cost Of Perfectionism: Why Agencies Filming 5 Ads A Month Lose In The Andromeda Era (Hidden Costs Beyond The Retainer In 2026)

There is a specific conversation happening right now in roughly the same form in hundreds of business owners' inboxes. A CMO or founder asks their agency why CPLs are up 40% over the last quarter. The agency replies with some combination of "rising CPMs," "iOS signal degradation," and "we're testing new creative concepts." A new 60-second hero ad lands two weeks later, on brand, well-lit. CPL ticks down for nine days, then climbs back. Repeat next month.

What's actually happening: the agency is producing 4–8 polished ads per month against an algorithm — Meta's Andromeda retrieval engine — that is built to evaluate thousands of candidates per impression. The agency hasn't done anything wrong by 2019 standards. They've simply been producing for a different platform than the one their client now advertises on, and the math has quietly tipped over.

This post is for three audiences: agency owners who feel the retention curve buckling, in-house marketing leads who need to defend or replace their agency to a CFO, and clients on a $5k–$15k retainer about to start a "should we switch" conversation. The numbers below come from real account audits in home services, real estate, coaching, and DTC in the last 12 months.

TL;DR

A typical $5k–$15k/month creative agency retainer produces 4–8 finished ads per month — call it ~$625–$3,750 per finished ad.
Meta's Andromeda retrieval engine (live and default since 2024–2026) needs 50–300+ creative variants per month at most service-business spend levels to perform well. The gap is structural, not stylistic.
At $625–$3,750 per ad, hitting Andromeda's creative volume floor would cost $30k–$1.1M/month in production alone. That math is not "tightenable." It's a different operating model.
The 30–60 day window after Advantage+ became the default is where most agency-client relationships break. Results crash, the agency blames the algorithm, the client starts shopping.
The agencies surviving are the ones that have rebuilt around volume — usually by adopting some flavor of batch video ads production and a real diversity-score discipline.
Polish still wins in TV, OOH, brand films, organic flagship content. Polish actively loses in paid social under Andromeda, where the algorithm prefers variety over varnish.

The Retainer Math, Without The Spin

Let's start with a clean look at what a typical creative agency actually delivers per dollar.

A mid-tier creative agency retainer for a service business or DTC brand in 2026 lands somewhere in the $5,000–$15,000 per month range. Inside that retainer, the deliverable list usually includes strategy, account management, some level of creative direction, scripts, shoot days, editing, and a defined output of finished assets. The output number is what matters.

The honest distribution we see when we audit accounts:

Monthly Retainer	Typical Finished Ad Output	Implied Cost Per Finished Ad
$5,000/mo	4–6 ads	$833–$1,250
$7,500/mo	5–8 ads	$937–$1,500
$10,000/mo	6–10 ads	$1,000–$1,666
$15,000/mo	8–12 ads	$1,250–$1,875
$20k+/mo (boutique)	4–8 hero ads	$2,500–$5,000+

The cost-per-ad number is the part most clients have never seen written down. Agencies don't sell on cost-per-ad because the number is unflattering. They sell on "concept development," "production quality," "brand consistency," and "platform strategy." All four of those are real and have value — for a media environment that no longer exists at the scale these agencies are pricing for.

Caveat: some of that retainer is genuine strategy, not pure production cost. Fair. But strategy is what Andromeda has progressively absorbed — audience definition, placement selection, budget pacing. That line item is shrinking in real value while creative volume explodes in importance. The argument here is about paid social under Andromeda; premium hero work still has its place, covered below.

Andromeda's Appetite, In Ads Per Month

We covered the underlying mechanics in What The Meta Andromeda Update Actually Means For Small Business Ads. The relevant facts for this post:

Andromeda is a deep-learning retrieval engine running on NVIDIA GH200 superchips, evaluating roughly 10,000x more ad candidates per impression than the system it replaced.
Detailed targeting is now mostly cosmetic. Advantage+ Audience is the default.
The algorithm needs creative variety to find audience pockets, because it can no longer rely on you to define those pockets.

The practical floor for serious lead-gen accounts, as a function of spend:

Monthly Ad Spend	Andromeda-Era Active Creative Floor	Refresh Cadence
$1k–$3k/mo	15–30 active ads	Monthly partial refresh
$3k–$10k/mo	30–75 active ads	Bi-weekly refresh
$10k–$20k/mo	75–150 active ads	Weekly refresh
$20k+/mo	150–500+ active ads	Continuous batch loops

Now overlay that on the retainer math.

A $10,000/month spender on a $7,500 creative retainer is getting 5–8 ads per month and needs 75–150 active creatives plus weekly refresh. They're operating at ~5–10% of the algorithm's working volume. A $15k/month creative agency might produce 8–12 ads against the same need. Still under 15% of the volume floor.

To hit 100 ads per month at the agency's per-ad cost of $1,250, you'd need a $125,000/month creative retainer. That number is not a typo. It's what perfectionism costs at Andromeda volume.

Nobody pays $125k/month for creative on a $10k/month media spend. So the gap doesn't get closed. The account runs at the volume the retainer allows, which is roughly an order of magnitude below what the algorithm wants. The CPL number reflects the gap.

The Unit-Economics Break

Let's make the break visible with a single account scenario.

Account profile: Regional HVAC operator. $12,000/month Meta spend. $8,000/month creative agency retainer. Target CPL $55. 5 ads produced per month, gradually accumulated to ~20 active over a rolling quarter.

Before Andromeda became binding (Q3 2024): 20 active creatives kept delivery diverse, detailed targeting still filtered audience meaningfully, CPL hovered $52–$61.

After Advantage+ became default + Andromeda fully in production (Q3 2025–Q1 2026): the same 20 active creatives now compete in an auction where competitors run 80–300 active creatives. Frequency on top 3 ads climbs to 5–7 within 10 days of each refresh. CPL drifts to $84 over six months. Agency response: new hero concept, more polished production, refreshed quarterly brand book.

Unit-economics math on that drift: $12,000 spend × CPL drift from $55 → $84 = ~89 fewer leads/month. At 18% close rate × $1,200 average ticket = ~$19,200/month in lost revenue. Annualized: ~$230k in lost top-line on a single account from creative volume drift.

That $230k of opportunity cost is paid against a $96k/year creative retainer. Math the client doesn't need a CFO to interpret. The agency is producing a perfectly polished negative-ROI line item.

This is the unit-economics break. Per-ad cost between $625 and $3,750 was viable when the algorithm needed 10–20 active creatives. It is suicidal at 75–150. The numerator (cost per ad) hasn't changed, but the denominator (what counts as "enough ads") moved by 5–15x.

Why Polish Started Hurting Specifically Under Andromeda

This is the part that hurts to internalize for anyone who came up in production. Polish doesn't merely "matter less" under Andromeda. In paid social specifically, polish actively underperforms — and there are two structural reasons.

1. Polished ads look like ads. Native content looks like content.

The Meta and Instagram feed in 2026 is overwhelmingly UGC-adjacent. Talking heads, casual phone-vertical shots, raw screen recordings. A perfectly color-graded 60-second narrative ad with B-roll, music bed, and brand sting is visually flagged by users as "ad" within the first half-second. Scroll velocity goes up. 3-second view rate drops. The algorithm reads the drop as "this creative is weaker" and reduces delivery accordingly.

Internal data on roughly 200 audited accounts shows that "agency-polished" hero ads underperform "founder/team raw vertical" ads on 3-second view rate by 25–55% in the lead-gen verticals (home services, real estate, mortgage, coaching). That's before you get to cost per result.

2. Polish requires concentration of creative volume.

When each ad costs $1,000+ to produce, you can't make many. When you can't make many, you can't feed the diversity score. Polish and volume are in direct economic tension under any normal retainer. Choosing polish chooses against volume, every time.

The volume-first operations get to skip this tradeoff. A batch-produced ad library at $10–$30/ad means you can have 200 ads and a handful of premium hero pieces. You don't have to choose. The agencies that are still choosing polish are choosing it because their production model can't make the other choice work financially.

The Client Retention Crisis Hitting Agencies Right Now

Andromeda's rollout was gradual enough that early effects looked like normal account variance. The compounding effect across 2025–2026 has not been quiet.

The pattern we see in agency-owner conversations:

Q4 2024: Advantage+ becomes default for most lead objectives. Some accounts notice a brief dip, most don't.
Q1–Q2 2025: CPLs start drifting. Agencies refresh creative, blame iOS, blame seasonality, blame "the platform."
Q3 2025: 30–60 days after accounts fully transition to Advantage+, clients hit a noticeable performance crash. CPLs up 30–60%, lead quality complaints up, sales pipeline thinner.
Q4 2025 – Q1 2026: Clients start churning. The "we love the creative but the numbers don't work" call.
Q2 2026 (now): Agency owners who haven't rebuilt their production model are watching MRR drift down 15–35% YoY.

The brutal part: the clients who churn don't go to a competitor with the same model. They go to a volume-first operation, a batch production provider, or in-house with cheaper tooling. The replacement is a structurally different operating model that costs less per ad and produces 10–50x the volume. If you run an agency, the strategic question is not "how do we save this account," it's "how do we rebuild production to match what the algorithm now requires."

5 Questions To Ask Your Agency This Quarter

If you're an in-house marketing lead or founder paying an agency retainer right now, these are the five questions that surface whether your provider has actually adapted to the post-Andromeda environment, or whether they're running a 2019 playbook with newer slides.

1. "How many distinct hooks did you ship last month — and what does 'distinct' mean to you?"

You want a specific number, and you want their definition of distinct to include hook line, visual entry, talent, format, and angle — not just caption variants. If the answer is "we tested 4 hook variations of our hero ad," they're not operating at Andromeda scale. A real answer in 2026 looks like "we shipped 38 distinct hooks across 6 angle categories, with 11 talent permutations and 4 format variants per hook category."

2. "What's your creative diversity score discipline?"

There is no single number Meta publishes called "creative diversity score" — but the agency should know what dimensions contribute to it (visual layout, hook, length, format, voice/talent, angle, aspect ratio) and how they track diversity across their active creative set. If they look at you blankly, they're not measuring it. If they're not measuring it, they're not improving it.

3. "What's your per-ad cost at the volume we actually need, and how does it scale?"

If the answer is some flavor of "we can do more ads but it costs more linearly," they don't have a batch production pipeline. They have a custom-build shop. A pipeline operation can quote you a fundamentally different per-ad cost at 100 ads/month than at 10 ads/month — because the templating and asset-library work amortizes. A custom shop quotes linearly, because every ad is rebuilt from zero. Those are two different businesses pretending to be the same one.

4. "Show me your refresh cadence and how you kill underperforming creative."

You want to see automated rules, a kill criterion (e.g., CPA > 2x target after $50 spend, 3-second view rate < 25%), a documented refresh cadence (weekly is the floor for accounts above $5k/month), and a method for deciding when an ad has earned more spend. If creative changes are still discussed monthly in a creative review meeting, the cadence is wrong by an order of magnitude.

5. "What does our active creative set look like right now — by angle, format, and talent?"

Open Ads Manager together. If 60–80% of impressions go to 1–3 ads, you don't have a creative library; you have 1–3 ads with a tail of unused inventory. If the angles are dominated by one (usually "social proof" or "offer"), you're missing the other 5–7 angle categories Andromeda is trying to retrieve from. The active set is the audit.

If your agency can answer all five with specifics, they've done the work. If they can't answer three, you're paying retainer rates for a 2019 service in a 2026 market.

The Producer-Vs-System Mindset Shift

The single biggest internal change for agencies that have made the transition is mental, not technical: the team stops thinking of itself as a producer of finished creative and starts thinking as a system operator.

A producer's job is to ship one polished asset; the metric is craft and approval; the cycle is brief → concept → review → revision → ship. A system operator's job is to keep a creative pipeline running; the metric is throughput, diversity score, and CPL performance; the cycle is hooks library → script generation → batch capture → batch assembly → launch → automated kill rules → learning → next batch.

The distinction matters because the entire skill stack is different. In a producer-first agency, the bottleneck is shoot days, edits, and revision cycles. In a system-first operation, the bottleneck is hook ideation, angle library breadth, and how fast the pipeline absorbs learnings from the last batch. Most agencies haven't made this transition because the producer-first model was their entire value proposition. The ones that don't switch tend to become luxury hero-ad shops for brands with TV budgets — fine business, just not the right partner for a $10k/month Meta lead-gen account.

When Polish Still Earns Its Keep

To be fully honest, polish hasn't died. It moved. The places where high-craft, low-volume production still earns clear ROI:

TV and connected TV (CTV). Linear and CTV reward production value. A 30-second polished spot has a different job than a 15-second Reels hook.
Out-of-home (OOH). Billboards, transit, place-based media. One asset, many impressions. Craft is the whole thing.
Brand films and explainer content. Pinned-to-homepage videos, sales deck content, investor-facing narrative work. Made once, used for years.
Top-of-funnel YouTube / pre-roll at scale. YouTube fatigues creative more slowly than Meta or TikTok and rewards a level of production quality that batch-vertical doesn't typically match.
PR / earned media assets. Anything that's going to be picked up and redistributed by a third party.
Brand-defining moments. Launch films, anniversary content, founder origin films. The handful of pieces a brand needs to feel like a brand.

Notice what's not on the list: paid social lead-gen. That entire channel has shifted to volume-first, and the production model needs to match.

The agencies winning in 2026 have figured out how to do both at the same time — a polished hero studio for the assets that need it, plus a batch video ads pipeline for the paid social engine. They're two different production stacks under one roof, priced and staffed differently, and clients buy them as separate line items rather than a single retainer.

What An Andromeda-Era Agency Proposal Actually Looks Like

If you're evaluating agencies right now — either as a client looking to switch, or as an agency rebuilding your offer — here's what a proposal that's actually built for Andromeda includes. The differences from a 2019 proposal are not cosmetic.

Volume guarantees, not concept approvals.

The proposal commits to a creative volume number per month (e.g., 80 net-new ads/month, 30+ active at any time, 4 refresh waves). It doesn't promise "concepts approved" or "campaigns launched." Volume is the deliverable. Approvals are baked into the angle-category level, not the individual-ad level.

Diversity metrics in the SLA.

The agency commits to ranges on visual format mix, hook category mix, talent permutations per category, and length distribution. If only 1 in 6 ads should be over 30 seconds, that's in the document. If 3 angle categories minimum need to be active each month, that's in the document.

A per-ad cost that drops with volume.

The cost model amortizes. At 30 ads/month it might be $40/ad. At 100/month, $22/ad. At 300/month, $14/ad. This is the signal that the production pipeline is real and the economics scale. Linear or near-linear pricing is the tell that you're paying for custom builds, not pipeline output.

Performance windows tied to volume hits.

Instead of "we promise CPL improvement," the proposal might say "after 90 days at full volume cadence, accounts in your vertical see 25–45% CPL reduction; if we don't deliver volume we eat the gap." Volume is the controllable input the agency promises. Outcomes are reported with context, not promised in isolation.

An explicit handling of the organic loop.

Andromeda rewards engaged audiences, and engaged audiences come partly from organic presence. A modern proposal links paid creative production to an organic distribution rhythm — using winning ad creative as organic posts, surfacing organic content into paid testing, and feeding the warm audience continuously. This is where done-for-you social media shows up alongside paid production in a proper proposal, not as a separate add-on.

Tooling and reporting transparency.

You should see the ad library, the diversity score dashboard, the kill rules, the refresh schedule, and the angle taxonomy. None of this is proprietary in 2026. Agencies hiding the operating system are usually hiding the fact that they don't have one.

If you read a proposal in 2026 that talks about "creative concepts," "campaign strategy," "platform expertise," and a fixed monthly retainer with no volume commitment — that's a 2019 proposal printed on 2026 letterhead. Politely pass.

The Internal Path For Agency Owners Who Want To Adapt

For agency owners not ready to lose another quarter to retention drift, here's the rebuild path. None of these steps require firing your existing creative team — they require building a parallel system next to it.

Pick one client to pilot. A $8k–$15k/month Meta spender with a willing operator on the client side. Get permission to run a 60-day volume-first sprint.
Build the hooks library. 50 distinct hooks across 6–8 angle categories for that client's offer. Most agencies have never written a 50-hook document.
Set up batch capture. One half-day with the founder or talent, captured vertical-native, producing enough raw material for 100+ ads.
Adopt a batch assembly stack. Build internally (CapCut, Descript, AI voice, batch render plugins) or partner with a batch video ads provider that handles assembly while you keep the strategic relationship.
Launch with kill rules. Automated pauses on weak ads after $50 spend, automated scaling on winners, daily review of alerts rather than weekly creative reviews.
Track diversity score discipline. Review the active set by angle, format, talent, and length each Monday. If one bucket is empty, the next batch fills it.
Price the new offer as a separate line item. Don't absorb volume production into the legacy retainer — it cannibalizes margin. Sell it as a new SKU at a per-ad-at-volume rate.

Most agency owners who've done this report that at 90 days, the pilot client is the highest-margin and lowest-churn account in their book.

For cross-context: Creative Is The New Targeting covers why this is happening at a platform level, and Why Filming One Perfect Ad Is Dead covers the production side from the creator's perspective.

Frequently Asked Questions

My agency just gave us our best month ever. Doesn't that contradict everything here?

More often it means you caught a winner still in its early scaling window. The question is what happens 30–60 days from now when that creative fatigues. If the agency can produce 20 candidates inside two weeks, you're fine. If the next refresh is a single new hero ad in six weeks, you're about to see the same regression. Best months are lagging indicators; production cadence is the leading one.

What if we love our agency for strategy and want to keep them?

The increasingly normal structure is to keep the strategic agency on a smaller retainer focused on strategy, brand, and high-craft assets, and bolt on a separate batch production provider for paid social volume. The two work together, with the strategy agency briefing into the batch pipeline.

Isn't volume creative going to look spammy?

It would if every viewer saw all 100 ads. They don't — each individual sees one or two at a time, matched to their pocket. From any single viewer's perspective, your brand looks like it's running thoughtful, relevant ads. The variety lives across the audience, not stacked on one person.

How do we know the agency's batch production is "real" and not AI slop?

Ask to see 30 sample ads from a recent batch. Look for angle diversity, hook variety, talent permutations, format variation, and whether the ads feel native to the platform. If they all look like the same script with different music, that's not batch production — that's a single ad with wrappers.

Will Meta penalize us for uploading 100+ ads at once?

No. Advantage+ Sales/Leads campaigns currently support up to ~150 ads per campaign, and Meta's guidance has been pushing toward creative variety since 2023. The penalty is for thin creative supply, not for volume.

Is this advice the same for TikTok and YouTube Shorts?

Directionally yes. TikTok fatigues creative even faster than Meta, so volume needs are higher relative to spend. YouTube Shorts has longer half-lives but still favors variety.

The Bottom Line

The cost-of-perfectionism argument isn't aesthetic. It's unit-economics. Agencies producing 4–8 polished ads per month on a $5k–$15k retainer are running negative ROI for their clients under Andromeda — not because they're bad at their craft, but because the platform has changed structurally. Per-ad costs of $625–$3,750 cannot scale to the 50–300 ads per month the algorithm now requires. The math doesn't permit it.

The agencies winning in 2026 have either rebuilt around volume themselves or partnered with providers that have. The clients winning have either pushed their agency to make that shift or moved on. If you're ready to look at what a volume-first operation actually delivers, the production model behind batch video ads and the parallel organic engine in done-for-you social media are the same operating model the better in-house teams and forward-leaning agencies have rebuilt around. Whether you buy it from us or build it inside your agency, the rebuild is non-optional. The retainer math beneath the old model has already broken.