Your CPA climbed 40%, 5 things to check before blaming the algorithm
Before you cut budgets, rebuild the campaign, or call your rep, run these five checks in this order. Most CPA blowouts we audit are four small leaks stacked on top of one honest market shift, not the algorithm turning against you.
The short answer
A 40% CPA jump is almost never a single cause. Across WordStream-tracked accounts in 2026, cross-industry median CPA is already up about 12% year over year, CPC up ~12%, conversion rate down ~9%, before anything specific to your account. Layer on iOS attribution decay (Meta reporting 40–60% worse than pre-ATT), creative fatigue past ~3.0 frequency on cold traffic, a slow mobile landing page, and PMax or broad-match query drift, and 40% is a normal stack. Check them in that order.
1. First, subtract the market
Before you diagnose anything, subtract what the whole channel is doing. If your CPA is up 40% and the benchmark drift is 12%, you own 28 points, not 40. That reframes everything below.
The 2026 numbers, honestly reported:
- Cross-industry median CPA up ~12.35% YoY, to roughly $23.74. (Foundry, 2026)
- Average Search CPC up ~12% YoY to $2.96 in Q1 2026. (Digital Applied, 2026)
- CTR is actually up ~7.5%, but conversion rate is down ~9% across 13 of 14 verticals, more clicks, fewer conversions per click. (Foundry, 2026)
- ROAS across the same panel is down ~10%. (Foundry, 2026)
- A separate Optmyzr panel of 21,000 accounts shows CPA up 4.4% YoY with CVR down ~1%, a milder version of the same story. (Optmyzr Q1 2026)
The mechanics: Smart Bidding and Performance Max now drive about 78% of Google Ads spend, that compresses the high-intent inventory faster and pushes clearing prices up. (LeapOut, 2026) At the same time, AI Overviews eat into organic clicks on many query types, and that displaced demand rolls straight into paid auctions. If you have not changed a bid, the price still went up.
So the diagnostic mindset is subtraction: the market took X points; my account is responsible for the rest. The rest is what the next four checks are for.
2. Check attribution before you change anything
Half of the "CPA blowouts" we open turn out to be measurement changes, not media changes. If Meta or Google can suddenly see fewer of your conversions, your reported CPA rises even though the real one didn't.
Meta iOS is the biggest offender. Attribution vendors converge on a 40–70% attribution gap on iOS-heavy accounts, with a realistic band of 40–60% under-reporting for most brands even after Conversions API is in place. (Ryze, 2026) A 2026 review states plainly: Meta attribution has deteriorated 40–60% since iOS 14.5, and January 2026's attribution-window changes accelerated the decline. (DojoAI, 2026) Even with a clean CAPI + AEM setup, accounts still lose 10–20% of true conversion signal compared to pre-ATT baselines. (SteerAds, 2026)
Google side, the equivalents are enhanced conversions coverage, GA4 event dedupe, and consent-mode v2 dropouts. If any of these degraded in the last 30–60 days, the CPA drift is measurement drift, not media drift.
Five-minute check:
- Pull backend orders (Shopify / CRM / Stripe) for the last 30 days versus the previous 30, iOS traffic specifically.
- Pull Meta's reported iOS purchases for the same windows.
- Compute the ratio for each window. If the gap widened, attribution moved and your real CPA is closer to flat.
Do this before touching bids. Cutting spend on a measurement problem is how you turn a reporting issue into a real revenue loss.
3. Look at frequency and CTR delta, not just spend
Creative fatigue does not announce itself. It shows up as a slow CTR decay while frequency creeps up and CPMs harden. There is no single universal number, Meta itself does not publish one, but the practitioner consensus for 2026 lands here:
| Audience type | Early warning | Likely fatigue zone | CTR decline trigger |
|---|---|---|---|
| Cold prospecting | ~2.5 frequency | 3.0–3.5+ | ~15–30% below baseline |
| Retargeting / warm | ~4.0–5.0 frequency | 5.0–7.0+ | ~15–30% below baseline |
| Broad / interest | No universal number | Trend against baseline | Sustained CTR decay + CPM rise |
The pattern you want to spot: frequency rising, CTR falling more than ~15% off the trailing 30-day baseline, CPMs rising in parallel. Any two of those three, together, means creative is the bottleneck, not bidding.
One useful sanity number: Meta creatives typically hold peak performance for 10–36 days before results start to dip, and that band varies by audience size and refresh cadence. (Adamigo, 2026) If your top ad is 45 days old and CPA is up, fatigue is a strong candidate.
4. Check the landing page, especially on mobile
CVR is half of CPA. If page speed or a UX change is dragging conversion rate down, no bid strategy in the world fixes it.
The evidence, and it is unusually consistent across datasets:
- As mobile page load time increases from 1 to 10 seconds, the probability of a bounce increases by 123%. (Google, mobile page speed study)
- For each 1-second delay beyond 2.5s LCP, conversions drop about 7%, and mobile sessions are roughly 2× more sensitive to speed delays than desktop. (Digital Applied, 2.1M-session study, 2026)
- Landing pages loading in 1 second convert at roughly 9.6% vs 3.3% for 5-second pages, an almost 3× gap. (Genesys, 2025–2026)
- Across a 50,000-page dataset, moving from 0–2s to 4–5s load time drops average CVR from ~3.8% to ~1.9% and pushes bounce from ~32% to ~58%. (DollarPocket landing-page benchmark)
- Mobile now drives about 65% of landing-page traffic but converts at ~58% of the desktop rate. (Digital Applied, 2026)
Target: mobile LCP ≤ 2.5 seconds, hard ceiling 3.0. Above that, you are burning CVR on every click you buy. Pull PageSpeed Insights against the exact landing URL your ads point to (not the homepage) and check the last 28-day field data, not just the lab score.
The offer-side twin question: did anything on the page change in the last 30 days? Hero copy, form fields, price display, above-the-fold layout, a new consent banner, a re-added chat widget. Small UX regressions are how CVR quietly slides 15% while nobody notices.
5. Audit the query and audience mix
In a world where 78% of Google Ads spend runs through Smart Bidding or PMax, the auction routes you into inventory you didn't explicitly pick. Two of the most common CPA-inflating drifts:
- Broad match creep. Broad match expands aggressively when Smart Bidding sees any conversion signal. Pull the search terms report for the last 30 days: how much spend went to queries you would not have chosen? A 20–30% "off-intent" share on non-brand keywords is common and it is a direct CPA tax.
- PMax cannibalization. PMax will cheerfully harvest brand and remarketing traffic at low CPA, inflating its reported performance and starving your Search/Prospecting campaigns of budget. Segment PMax reporting by placement and asset group; separate brand exclusions if you haven't already.
On Meta, the parallel is Advantage+ audience expansion pushing spend into cheap-CPM but low-intent inventory. Symptoms: CTR steady, CVR down, cost per landing-page view down, cost per purchase up. That is a mix problem, not a creative problem.
The fastest triage: pull the last 30 days by network / placement / asset-group / audience and rank each segment by CPA. The top 10% of segments almost always own the bulk of the drift.
Which lever, in what order
| Symptom you actually see | Likely cause | Where to look first |
|---|---|---|
| Backend revenue steady, Meta CPA up | iOS attribution decay | Backend vs. Meta ratio, iOS only, last 30 vs. prior 30 |
| CTR up, CVR down, CPC up | Auction inflation + query drift | Search terms report; PMax asset-group CPA |
| Frequency > 3.0, CTR down 15%+ | Creative fatigue | Rank ads by CTR vs. rolling baseline; refresh top-spender |
| CPC flat, CVR down, mobile worst | Landing page / UX regression | PageSpeed field data on the exact ad URL; last 30-day page diffs |
| Everything moved, all at once | Two or more of the above stacked | Do the checks in this order; do not batch-fix |
The single most common mistake we see in audits: someone cuts spend or lowers Target CPA on a measurement problem. That converts a reporting issue into a real revenue loss, then the new lower-volume state looks like "the algorithm is broken" and the loop restarts.
Run the five checks in order. Subtract the market first. Then attribution. Then creative. Then landing page. Then query mix. Most 40% CPA blowouts we open resolve to 12% market + 15% attribution + 10% creative fatigue + a slow mobile page, four small things, not one big one.