Why your CAC went up. And why your agency won't tell you the truth

Telegram, Tuesday, 11:47

"ERA, need an outside read. CAC over the last 90 days is up 42%. The agency shows me a report — CTR is fine, impressions are up, 'traffic quality has improved.' But MER dropped from 4.1× to 2.7×, and I can physically see less cash in the account. They're telling me it's seasonality plus iOS. I don't buy it. What do I check?"

This is the most valuable kind of inbound. Not "help me come up with creatives," not "should we try TikTok" — the founder already sees the symptom and doesn't believe the interpretation he's being handed. This is the highest-leverage moment for diagnosis, because he hasn't yet spent six months fighting the wrong fire.

What follows is what I wrote back, expanded into a self-audit any founder of a €1–5M business can run in an hour before their next agency call. Without this audit, the conversation stays in CTR and impressions when it needs to be in contribution margin and incremental MER. Different conversations entirely.

What the agency shows you — and where it's a shell game

The standard quarterly report looks like this:

CTR stable, **1.8–2.4%** on cold audience — "within benchmark"

Impressions up 28% — "increased reach"

CPM up 19% — "seasonal auction inflation"

ROAS in Meta Ads Manager **5.1×** — "channel efficiency holds"

Quality score and engagement rate — "positive trajectory"

What's true here: CPM really is up, and not only because of seasonality (more on that below). The ROAS Meta reports is also technically a real number — it comes straight out of Ads Manager.

What's a shell game: none of these metrics describes the economics of the business. They're channel metrics, optimized for presentation. Real CAC is total marketing spend / new customers, not modeled ROAS inside the platform. Real efficiency is MER, not ROAS. The real funnel is site conversion plus repeat rate, not CTR.

The agency isn't lying on purpose. It's showing you the metrics it knows how to produce — the ones Ads Manager generates. P&L-level metrics — contribution margin, incremental MER, attribution leakage — don't show up there, because producing them requires simultaneous access to Shopify, Klaviyo, and the P&L, plus the competence to stitch it into one number. That's CMO work, not agency account manager work.

So the first step isn't "hire a different agency." It's moving the conversation from channel level to business level. Below are the four lenses I run through any rise in CAC.

Channel Saturation Curve — Mental Model #1

This one comes from Andrew Chen and goes by the Law of Shitty Clickthroughs: any paid channel decays over time. CPC rises, CTR falls, audiences burn out. What delivered 2.5× ROAS at launch delivers 1.4× twelve to eighteen months later on the same creative mechanics — not because the creative is bad, but because the fresh audience inside the channel's targeting parameters has run out.

Two things need to be separated here. Seasonal CPM inflation is real but predictable and reversible. Per MHI Growth Engine and AdAmigo for 2026: global Meta CPM holds around **$6.59**, the US around **$23**, premium MENA audiences **$10–14**, CEE (Romania, Moldova) **$3–6**. Q1 is roughly **15% below** annual average; Q4 is peak. If a founder is comparing Q4 to Q3, part of the CPM jump is auction calendar, not his problem.

Channel saturation is a different animal. It's structural exhaustion of TAM inside a channel under fixed targeting. Tells: CPC up **>2×** over twelve months with stable targeting, CTR holding or falling more slowly than CPC rises (the audience still responds, it just gets more expensive), frequency climbing — the same people seeing the ad eight to twelve times a week — and conversion rate on that audience eroding.

Diagnostic question: show me CPC and CTR over the last 18 months, monthly, not quarterly average. Find the inflection point. If CPC is up 2.5× and on-site conversion rate hasn't moved, that's an audience ceiling, not bad creative. The thing that helps least is "new creatives." What helps is audience expansion (geo, age, interest), a new channel in the portfolio, or a parallel brand layer that reduces dependence on cold acquisition.

Wrong way founders use it: they swap agencies, then swap again. They explain the drop as "creative fatigue," hire a motion designer, ship 40 new videos. Three months later CPC is higher still, because creative volume doesn't cure TAM saturation — it just prolongs the agony.

Specific 2026 signal: Advantage+ Shopping Campaigns (ASC) on Meta average **4.52× ROAS** against **3.70× for manual**, a 22% gap. But this only holds at ≥**50 conversion events per week per ad set** and ≥**15–50 creatives in the system**. Founders running five ASC simultaneously with three creatives each fragment the algorithm and underperform manual.

MER vs ROAS Hierarchy — why ROAS lies when CAC is rising

This is the single most important metric hierarchy for a €500K–5M business, and most founders don't hold it. ROAS is revenue from ads / ad spend — a channel metric Meta or Google reports. MER (Marketing Efficiency Ratio) is total revenue / total ad spend — a business metric you can only see when you reconcile Shopify revenue against total marketing spend.

ROAS lies through attribution. After iOS 18 and the stabilization of ATT opt-in around **30–35%** (Adjust Q2 2025), the Meta pixel misses more than half of real conversions in many e-com accounts. Modeled ROAS in Meta currently runs **20–40% below** reality — this is the case where the platform underrates its own contribution, and a founder watching only Ads Manager doesn't see part of the incremental revenue.

But the inverse problem is worse. If MER falls from **4.1× to 2.7×** while platform ROAS holds steady, that means platform attribution is over-counting — claiming purchases that would have happened anyway (organic traffic, repeat buyers, branded search). On paper Meta ROAS is 5×; in the bank account MER is 2.7×. The delta is attribution overlap, and neither Meta nor Google volunteers it.

Healthy MER for e-com is **3×+** with gross margin >55%. For subscription/services, **4×+**. MER 3× at 70% gross margin is a healthy business. MER 3× at 40% margin after shipping, returns, and payment fees is a slow bleed. The math runs after all variable costs, not against gross margin.

What matters for this founder: MER 2.7× in beauty at €68 AOV is the edge. Control calculation: contribution margin per order = AOV × gross margin − shipping − payment fees − average return cost − fulfillment. If you land at €18–22 per unit and CAC has climbed to €25–28, every new sale loses money, and the business is growing in revenue while cash drains.

What to do: measure incremental MER, not attributed. Cleanest method is geo-lift — turn the channel off in 5–10 matched cities for 14 days, measure delta vs forecast. For Moldova and other single-geo markets, geo-lift rarely works because there aren't enough geo-units — there you use time-series holdout: shut the channel for 14 days, compare total revenue against forecast. Per eMarketer 2026, **36.2%** of marketers are increasing investment specifically in incrementality testing. This is now baseline, not nice-to-have.

Site conversion vs channel — the background leak the agency blames on the channel

Third common cause of rising CAC that Meta can't show you from the inside: site conversion dropped, and the agency writes it off as "traffic quality." Math is simple. If paid CAC is up 42% and site conversion fell from **2.1% to 1.5%** (still within the e-com median of 1.5–2.5%), almost all of the CAC rise is explained by the site, not the channel. The channel delivered the same clicks for the same money, but they convert to purchase less often — formally CPC didn't move, in reality CAC did.

Possible causes of site conversion drop: PSP swap with worse UX, new fields in checkout, slow page-load after a Shopify theme update, a new banner or popup overloading the page, a mobile checkout bug nobody tested. Industry math: every **100ms of page-load degradation = −1.11% conversion**. For a €2.3M business that's **−€25K per year** per 100ms.

What to check: the funnel impression → click → landing view → product page → cart → checkout → purchase, conversion rate at each step over the last 90 days vs the prior 90. Where's the largest relative drop? E-com median: cart→checkout drop **70%**, strong brands hold **50%**. If your current period shows 85% cart abandonment, that's **+30% revenue potential** from fixing checkout, with zero touch to paid.

Specific diagnostic — Klaviyo flows. If browse abandonment flow isn't live or opens at **<35%**, that's potentially **€2–4K/month** of additional revenue on a list the size of this founder's. Per Klaviyo 2026 benchmarks: browse abandonment averages **$1.07 RPR**, **P90 $7.21**. If a founder is paying **+42%** for traffic acquisition while abandoned browse isn't running, he's subsidizing the leak. Flow vs campaign asymmetry: flows generate **41%** of email revenue from **5.3%** of sends, with **18× higher** revenue per recipient. It's the cheapest lever, and it's always the last one pulled.

Wrong way founders use it: they accept "low-quality traffic" in the agency report as fact. The actual check takes 30 minutes — open GA4 and Shopify, look at conversion rate month over month, do the subtraction. If site CR didn't drop, the problem is in the channel. If it did, the problem is the site, and any budget increase to paid deepens the loss.

Attribution noise — iOS 18, ATT, CAPI, GA4

Fourth lens, no conspiracy thinking: after iOS 18 and the stabilization of ATT opt-in at **30–35%**, attribution is structurally broken. It doesn't mean Meta is "lying" or Google is "stealing data." It means user-level deterministic tracking is impossible by OS design, and the platforms moved to modeling.

What that means mathematically: the Meta pixel alone (no CAPI) misses **>50% of conversions** in many e-com accounts. Without enhanced match (email/phone hash) and deduplication between pixel and CAPI, a business overpays CPM by **15–25%** on blind bids. Google Ads without enhanced conversions loses **25–35%** of attribution through consent mode and cookie blocking. GA4 can disagree with Shopify revenue by **15–30%** — that's not a bug, that's the sum of attribution gaps.

What needs to be connected in 2026 for any e-com business above €1M:

**CAPI** server-side via Stape, Aimerce, Cometly, or server-side GTM

**Enhanced match** on hashed email and phone

**Deduplication** of events between pixel and CAPI (shared event_id)

**Enhanced conversions** in Google Ads

**Unified attribution layer** — Triple Whale or Northbeam on top of Shopify for a sane view of blended CAC and incremental MER

Triple Whale in 2026 gives you fast insight via the Compass layer, which fuses MTA + MMM + incrementality. Northbeam is the pick when your CFO wants methodological rigor: at the end of 2025 they partnered with Meta, TikTok, Snap, Pinterest, and MNTN on a Clicks + Deterministic Views model, which delivers platform-verified view events rather than purely modeled.

At €2–5M revenue, it's worth looking at MMM too. The open standard now is Google Meridian (January 2025), which displaced Robyn in active development. Meridian handles reach/frequency for video, diminishing returns, and shipped Scenario Planner in open beta in February 2026. An implementation that used to take six months on a consulting team now lands in **4–8 weeks** with weekly-cadence data. For a €2.3M business, that's inside the envelope of reasonable.

The signal I lean on: **46.9%** of US marketers plan to increase MMM investment in 2026, and **27.6%** name MMM the single most credible measurement methodology — first place in the eMarketer survey, above MTA and platform attribution. Attribution moved out of tracking tools and into measurement methodology, and the founder who doesn't track this is reading an old map.

Protocol — the self-audit before your next agency call

Seven questions. For each, a specific number or "I don't know." "I don't know" is a valid answer and instantly surfaces the first leverage point.

**1. What's your MER over the last 90 days vs the prior 90?** Total revenue / total marketing spend, no attribution magic. Down >25%, the problem is systemic, not seasonal. Stable, the problem is in the platform, not the business.

**2. What's your contribution margin per order after ALL variable costs?** COGS + payment fees (2.5–3.5%) + shipping + return cost × return rate + support cost. If it's below paid CAC, the business is scaling at a loss regardless of ROAS.

**3. What's your site conversion rate over 90 days vs the prior 90?** Down >15% means it's not a paid channel problem, it's a site/UX/checkout problem. E-com benchmark: median **1.5–2.5%**, strong **3.5%+**.

**4. What's your 6-month repeat purchase rate by cohort?** Not lifetime average across all customers — a specific Q1 2025 cohort. E-com median **25–30%**, strong **45%+**. Below median, rising CAC deepens the retention leak, and focus belongs on activation, not acquisition.

**5. Which three channels have you seriously tested in the last year — budget ≥€3K, period ≥30 days, ≥3 creatives?** If the answer is "one, Meta," the business has no channel diversification and is structurally exposed to saturation. If "we tried everything at €500," that's noise, not tests.

**6. CAPI and enhanced match connected? Deduplication on event_id set up?** If not, you're overpaying CPM by **15–25%** and Meta attribution is structurally understated. This isn't a quarterly project, it's a two-week build and mandatory infrastructure for any business above €1M.

**7. If tomorrow I doubled your CAC efficiency — would that fix the business problem?** If you have to think for more than five seconds, the real constraint isn't CAC. It's downstream: retention, capacity, product, pricing. The premise audit is the most painful and the cheapest step.

If you can't answer five of seven, your next agency call shouldn't be about new creatives. It should be about management reporting and the attribution stack. That's not a CMO conversation — it's the base layer without which the CMO conversation is meaningless.

Close

CAC doesn't rise in a vacuum. It always has a specific cause in one of four lenses: channel saturation, MER vs ROAS attribution gap, site conversion drop, attribution noise. The agency shows you what it knows how to produce — channel metrics. P&L-level diagnosis is different work, and it saves **3–6 months** of wrong-direction spend.

If you want to walk through these seven questions live with my reads on your specific numbers, DM me on Telegram **@mrkt_era**. First 30-minute diagnostic is free, and by the end of the call you'll know what to fix first, what to fix second, and what to leave alone until next quarter.