Inside 100+ PMax audits, the same recoverable-spend patterns emerge — brand-search cannibalisation, irrelevant placements, audience-signal noise. Here's where to look.
Google’s PMax campaign type is intentionally a black box. The default reporting hides channel-level data — you see “PMax: $X spend” without knowing how much went to Search vs Shopping vs YouTube vs Display vs partner placements.
That opacity is the source of most wasted spend.
You can fix this. The Mike Rhodes channel-split script (and several public variants) pulls placement-level data PMax doesn’t surface in the UI. Run it weekly. The output usually surprises:
This is the #1 recoverable-spend category. PMax bidding on your brand terms is buying clicks you’d already get organically — at full CPC. The brand-exclusion setting in PMax is leaky; it doesn’t fully prevent this.
Fix: explicit brand campaign with -1 negatives in PMax, monitored monthly via search terms report.
Most accounts ship a single audience signal and call it done. The compounding work is testing 5–8 audience signals, retiring underperformers, doubling down on top 2.
Lookalikes from your existing customer list typically outperform interest-based signals 2–3×.
Asset groups are PMax’s version of ad groups. Most accounts ship 1–2; we typically see ROAS lift from running 4–8, each tightly themed.
Standard sprint: separate brand campaign with -1 negatives in PMax → exclude Display + irrelevant partners via placement controls → audience signals A/B test → asset group split by product theme → CAPI + Enhanced Conversions deployed.
Typically 4 weeks from audit to restructure go-live. ROI shows in week 2 after restructure as account retrains on cleaner signal.
Across 100+ PMax audits we’ve run, median recoverable spend is around 35%. Range 18–62%. We quote the actual number per account, not a range.
If you’re spending $50k/mo on PMax, that’s typically $15–18k/mo of recoverable spend. Pays back our $1,199 audit + $3,499 restructure sprint in week 1 of the new structure.
PMax's bidding algorithm optimises against whichever conversions are cheapest — and brand-search clicks (people typing your brand name into Google) are the cheapest, highest-converting traffic available. Without explicit negative-keyword brand terms in PMax, the algorithm claims credit for traffic Google would have given you free via organic brand-search. The fix is two-step: add brand terms to the PMax negative-keyword list, then run a separate dedicated brand-search Search campaign you control. Most accounts recover 20–40% of measured PMax 'success' as actually-cannibalised free traffic.
Three layers. (1) Customer lists: upload existing-customer email/phone (hashed) so PMax knows what your converted customers look like — not what Google guesses. (2) High-LTV segments: split customer list by lifetime value tier, feed top-quartile separately for higher-CAC tolerance. (3) Site visitors with intent signals (cart adds, pricing page visits, demo requests). Without these, PMax's lookalike modelling defaults to whoever Google's lookalike algorithm decides 'looks like a converter' — typically retargeting traffic that would have converted anyway.
Smart Shopping was deprecated in 2022 — there is no choice. The relevant question is whether PMax should be the only campaign type or whether to run PMax alongside dedicated Search + Shopping campaigns. For most mid-market DTC brands above $25k/mo Google Ads spend, the answer is: run PMax with strict brand-term exclusions + first-party audience signals + value-weighted conversions, AND run a separate dedicated brand-search Search campaign, AND a manual Shopping campaign for high-margin SKU-level control. PMax-only setups consistently underperform this hybrid by 18–28%.
Configure conversion-value modelling so each conversion type has a value reflecting actual margin contribution: a $50 first purchase from a new customer is not the same value as a $50 repeat purchase from an existing customer (existing customers have higher LTV but lower marginal acquisition value). Use Enhanced Conversions for value to send the actual transaction value alongside the conversion. For new-vs-returning, configure a custom 'new-customer-only' conversion action and feed it as a separate optimisation target. Without value-weighting, PMax optimises for cheapest conversion, not most valuable.
Material. iOS 14+ ATT + browser tracking restrictions mean GA4 + ad-platform pixels miss 11–22% of conversions. Without server-side tracking (GTM Server-Side + Enhanced Conversions + offline-conversion imports), PMax's bidding algorithm makes decisions on incomplete data — under-bidding on high-intent users whose conversions are invisible to it. Server-side rebuild typically takes 2–4 weeks and recovers 11–22% of attributed conversions, which compounds with PMax bidding accuracy improvements over the next 60–90 days.
Three scenarios. (1) Pure-brand DTC brand where 80%+ of revenue is brand-search — PMax cannibalisation is the dominant pattern; classic Search + Shopping is more efficient. (2) High-margin B2B with 5–10 commercial intent terms total — PMax's broad-match logic destroys precision; manual Search is more efficient. (3) Pre-launch / new product where you have no conversion history — PMax cannot optimise without conversion data; start with manual Search + manual Shopping, layer in PMax once you have 30+ conversions/month.
