Performance Max best practices for ecommerce in 2026 (200+ account playbook)

Performance Max in 2026 is not the black box most ecom operators treat it as.

It rewards structure. It punishes laziness. And it has more levers than the average Google Ads agency surfaces - because most agencies treat PMax as untouchable, set a tROAS, ship the asset pack, and pray.

After 200+ ecom accounts and €200M+ in cumulative managed PMax spend, we have a deterministic playbook for what works. This is not theory. Every step has shipped on multiple paying accounts.

Accounts

200+

ROAS lift (median)

+25%

Spend managed

€200M+

Time to lift

9 wks

Best practice 1: Structure Performance Max by margin tier, not category

The single most common mistake on ecom PMax accounts is one campaign for everything.

A $9 squeaky toy and a $200 orthopedic dog bed sit in the same PMax. Smart Bidding chases the easy clicks - the squeaky toy converts cheaply, the orthopedic bed gets starved. PMax bids on the easy product because the conversion volume is there. The margin contribution looks fine on the surface and quietly dies underneath.

The fix is splitting by margin tier.

Tier A: top 20% by margin contribution and velocity. The Champions. Highest tROAS, dedicated asset group, brand-defence Search underneath. Typically 50-60% of spend.

Tier B: rest of the active catalogue. Sleepers. Standard tROAS, broader asset groups, less aggressive structure. Typically 30-35% of spend.

Tier C: tail. Low margin, slow velocity, dead SKUs. Either paused or moved to a low-bid PMax with floor budget. Typically 5-15% of spend.

This is not a hack. It is how PMax wants to be fed. Splitting prevents the algorithm from cross-subsidising your worst products with your best.

The jewelry case study is the canonical example: €120k/month account stuck at 1.8 ROAS for 6 months under a previous agency running one big PMax. After the three-tier restructure (and four other moves), CPA was down 38% in 9 weeks. Same spend. Same algorithm. Different structure.

Best practice 2: Rewrite the top 80 SKU titles to match real search intent

Performance Max bids on the matches Google makes between your feed and incoming queries. The feed is the lever - and titles carry 70% of the matching signal.

Most ecom feeds run brand-first titles: "Madeleine - Solid Gold Necklace - 14k". Google reads "Madeleine" as a generic word, weights "Necklace" and "Gold" as the main signals, and matches you against "gold necklace" - which converts at 0.4% because the searcher wants gold-plated.

The fix: rewrite the top 80 SKU titles by impression share with specs first, intent second.

Before: "Madeleine - Solid Gold Necklace - 14k" After: "14k Solid Gold Necklace, Madeleine, 18 inch, 4.2g, Womens"

Specs in the title do two things. First, they kill match to gold-plated queries because Google reads "solid gold" as a stronger signal than the inferred meaning of "14k". Second, they enable match to long-tail intent like "18 inch solid gold necklace women" - which converts at 4x the rate of "gold necklace."

Time investment: 6 hours of focused writing for the top 80 SKUs. We use the search-terms report as the source for the keywords - not what we think sounds good in marketing copy.

This single move typically lifts ROAS 15-25% on its own. The full Google Shopping feed optimization guide covers the title-rewrite playbook in detail.

Best practice 3: Deploy custom labels for margin band, velocity, and vertical signal

Custom labels are the second feed-side lever. Three labels cover 80% of the value across every vertical.

margin_band: high (over 55%), mid (35-55%), low (under 35%). The single most valuable label - it lets PMax listing-group rules enforce ROAS floors by margin tier.

velocity_30d: hot, mid, slow, dead. Velocity-driven re-tiering. SKUs going hot get promoted to Champions. SKUs going dead get demoted to tail or paused.

Vertical-specific: metal_type for jewelry (solid_14k, solid_18k, plated, silver). season for fashion. fitment for auto. AOV band for furniture. Each vertical has 1-2 high-leverage labels that solve a structural problem in the auction.

The labels feed into PMax listing-group rules so the algorithm bids by margin tier, not blended ROAS. Before labels, PMax averages everything. After labels, PMax bids your 60%-margin SKUs at a 2.0x ROAS floor and your 30%-margin SKUs at a 3.5x ROAS floor. The bottom line typically improves 15-30% with zero change in ad spend.

This is also the work Scaley AI automates on accounts with operator capacity but no senior media buyer. Same labeling logic, automated at $49/mo for the Labelizer tier.

Best practice 4: Exclude branded queries from broad Performance Max

Performance Max with default settings will capture your branded queries (people searching for your brand name) inside broad PMax. ROAS looks inflated because branded queries convert at 8-15x. Smart Bidding sees this and concludes "PMax is amazing" - and bids more aggressively on the non-branded matches that are actually cannibalising your Search campaign.

The fix is one settings change: add your brand terms to the account-level brand exclusion list. Branded queries then route to a dedicated Search campaign at the bottom of the funnel where you have CPC control and can attribute conversions cleanly.

This typically reduces blended PMax ROAS by 20-40% on the surface - but the math on the underlying account improves because the Search campaign now captures the high-intent queries with proper attribution, and broad PMax stops chasing fake signal.

Most ecom operators resist this because the dashboard looks worse. The actual account performance improves. The dashboard is a vanity metric; the back-end Shopify number is the truth.

Best practice 5: Install server-side tracking before any other optimisation

Performance Max bids on the conversion data Google sees. If Google sees half the conversions, Smart Bidding bids for half the value. Every other optimisation in this guide is worth half as much without server-side tracking.

Client-side pixels lose 30-40% of ecommerce conversions to Safari ITP, ad blockers, and iOS updates. The trend is one-way - every iOS release loses more, every browser update loses more. Client-side tracking is structurally broken in 2026.

The fix is server-side tracking via the ZenoX Shopify app for Shopify stores (most ecom is Shopify) or Google Tag Manager Server-Side for non-Shopify. The install is one day of work and pays back every month after.

Server-side tracking also enables enhanced conversions, value-based bidding tuned to real margin, and offline conversion uploads - all of which compound on the same Smart Bidding signal.

Best practice 6: Reset tROAS if the algorithm is starved

The most damaging thing previous agencies do to PMax accounts is raise tROAS when ROAS slips. The intuition is reasonable: "ROAS is falling, target more ROAS." The math is wrong: raising tROAS tells Smart Bidding to bid less, which drops impressions, which drops conversions, which makes the next ROAS reading worse. A doom loop.

If your account inherited a starved PMax (impressions dropped 40%+ over 8 weeks while tROAS was raised), the fix is dropping tROAS to give the algorithm permission to bid again. We typically drop to 220% on starved accounts. Within 5 days impressions double. Let it stabilise 14 days, then nudge tROAS up 5% every 2 weeks if efficiency holds.

This is the only "bid" change in the first 30 days. And it is undoing damage, not making new bets. Real bid optimisation is what you do after the structure is right.

Best practice 7: Vertical-correct asset rotation cadence

PMax asset groups go stale at different rates per vertical. Most agencies refresh quarterly across every client - which is too slow for beauty and too fast for furniture.

The honest cadence per vertical:

Beauty tools: every 2-3 weeks. Trend half-life is 4-6 weeks. Asset packs that lasted 6 months in 2022 die in 6 weeks in 2026.

Fashion: every 6 weeks. Catalog drops drive the cadence. Each new collection ships with its own asset pack.

Home decor: 8-10 weeks normally, weekly in Q4. Gifting season demands fresh creative for every promotional window.

Jewelry: 12+ months on lifestyle photography. Lifestyle imagery on solid gold compounds longer than any other vertical.

Furniture: 6 months. High-AOV, long consideration, decision is rational. Asset stability matters more than freshness.

Auto accessories: 6 months on product shots, refreshed quarterly for seasonal SKUs (winter mats, summer kits).

Stale creative kills CTR which kills Smart Bidding's ability to scale. Cadence mismatched to vertical kills accounts quietly over a quarter.

Best practice 8: PMax + Search + Demand Gen, not PMax alone

Single-PMax setups typically lose 17% ROAS vs the full stack.

The full stack:

Performance Max: broad demand capture. The workhorse. 60-75% of spend.

Standard Shopping (where supported): bottom-funnel branded queries and high-margin SKUs you want manual control on. 5-10% of spend.

Search: branded queries (excluded from PMax), bottom-funnel buying intent, brand-defence against competitors bidding on your terms. 15-25% of spend.

Demand Gen: top-of-funnel visual research phase. YouTube, Discover, Gmail. New product launches and verticals where buyers research visually (furniture, beauty tools, home decor). 5-15% of spend.

The full stack runs into account-level brand exclusions, shared budgets where useful, and attribution-aware bidding. The PMax vs Standard Shopping breakdown covers the split decision in more depth.

What 200+ accounts taught us about scaling Performance Max

Three patterns repeat across every vertical we run.

Pattern 1: structural work outperforms bid optimisation 4:1. Feed engineering, account structure, custom labels, server-side tracking - the four levers that compound. tROAS adjustments are the last 20% of the work, not the first.

Pattern 2: tROAS gets raised too aggressively, dropping starves accounts. The single most damaging move previous agencies make is treating tROAS as a primary optimisation lever. It is a secondary lever. Structure first, asset rotation second, bid third.

Pattern 3: Performance Max wants to be fed clean, structured signal. The algorithm is genuinely capable. It is also genuinely confused when you feed it 14 variant SKUs that are actually one product, mixed-margin asset groups, branded queries muddying intent, and client-side conversion data missing 30% of the picture. Clean the inputs and PMax compounds.

The accounts that scale on Performance Max are the accounts that stopped optimising bids and started optimising structure. PMax was never the problem. The inputs were.

What this means for your account this quarter

Open Performance Max. Look at the listing-group view. Then ask three questions.

One: are all my products in one bucket? If yes, split by margin tier this week.

Two: have my top 80 SKU titles been rewritten with specs that match real search-term intent? If no, that is the highest-leverage move on your account. 6 hours of focused work for a 15-25% ROAS lift.

Three: is my server-side tracking installed and validated against my Shopify back-end? If no, every other optimisation is worth half.

The full playbook is in the free Google Ads eCom Lab community - 740+ ecom operators running the same moves on their accounts. Or if you want senior operators running this on your account, drop the URL on WhatsApp. We pull it up live, walk through the structure first, and tell you on the spot whether the agency math works for your spend level.

The Performance Max accounts that scale in 2026 are the ones that treat PMax as one lever among five - not the only one. Structure, feed, labels, tracking, then bids. In that order.