How an E-Commerce Brand Cut Wasted Spend 34% and Grew ROAS 2.1x After Switching to Google Ads AI Max

The Google Ads AI Max E-Commerce Case Study: Brand, Problem, and Stakes

The brand in question — a UK-based online apparel retailer selling mid-price casualwear — was spending approximately £42,000 per month across Google Ads by Q3 2025. Their catalogue sat at around 3,400 active SKUs, split across men's, women's, and kids' ranges, with seasonal peaks at Easter, back-to-school, and the November–January sale period. Revenue from paid search was the single largest acquisition channel, accounting for roughly 58% of all new customer orders.

The problem was not that campaigns had stopped working. The problem was that they had plateaued badly. After two years of incremental Smart Bidding adjustments, the account had become a complex tangle of 47 ad groups, 900+ keywords, 14 separate Shopping campaigns, and a Performance Max campaign that nobody on the internal team fully trusted. The manual overhead was significant: the in-house PPC lead was spending an estimated 22 hours per month on bid adjustments, negative keyword pruning, and asset testing — time that was producing diminishing returns.

Three specific metrics were flashing red heading into Q4 2025:

• ROAS: Blended account ROAS had slipped from 3.4x to 2.8x over six months, despite CPCs holding roughly flat.
• Wasted spend: A spend audit identified £15,200 per month — roughly 36% of monthly budget — flowing to search terms with zero recorded conversions over 90 days.
• Click-through rate: Average CTR across text ad campaigns sat at 3.1%, well below the 5–7% range many practitioners report as achievable in apparel with well-optimised creative.

The stakes were real. The Q4 peak (October through January) historically accounted for 41% of annual revenue. Entering that window with a deteriorating ROAS and a bloated account structure was not an option the commercial director was willing to accept.

"The account had become a monument to decisions made two years ago. Every layer of complexity was there for a reason that no longer existed."

After an external audit in late August 2025, the recommendation was unambiguous: consolidate, migrate to Google Ads AI Max campaigns, and build proper guardrails before the Q4 rush. The team had eight weeks to execute before the seasonal window opened.

Strategy: What Was Decided — and Critically, What Was Not

The strategic brief was written around three non-negotiable requirements: preserve brand safety, maintain human oversight on budgets, and have a clean migration path that could be reversed within 48 hours if performance deteriorated sharply in the first three weeks.

The decision to use AI Max — rather than simply consolidating into standard Shopping or a conventional Performance Max setup — came down to one factor: creative breadth. With 3,400 SKUs and limited creative production resource, the team needed a campaign type that could generate and test asset combinations at a scale no human team could match. AI Max's ability to synthesise ad copy, adapt assets across placements, and optimise in real time against a target ROAS signal was judged to be the right fit for a catalogue this large.

What was deliberately not done is equally instructive:

• No full automation handover on brand terms. Branded keywords were carved into a separate, manually-controlled exact-match campaign with its own budget and a brand exclusion list applied to the AI Max campaign. This was a firm line.
• No removal of negative keyword lists. The audit had produced a consolidated negative list of 1,847 terms. These were imported in full before launch.
• No broad asset uploads without review. Every image asset, headline, and description string was reviewed against the brand's tone-of-voice guide before being added to the asset library. The team rejected roughly 30% of AI-suggested creative in the first two weeks.
• No single campaign for all product categories. The decision was made to run three AI Max campaigns — one per gender range — rather than one monolithic campaign, to preserve cleaner performance signals and budget control by category.

For teams unfamiliar with the levers available inside AI Max, the Google Ads automation controls documentation covers exactly which settings give human teams meaningful oversight without undermining the model's optimisation capability.

Implementation: Campaign Structure, Timeline, and Tools Used

The migration ran across eight weeks in September and October 2025. The timeline was deliberately front-loaded with audit and preparation work, with the actual campaign launches happening in week five — leaving three weeks of live performance data before the Q4 peak began in earnest.

The tools used were deliberately minimal. Google Ads Editor handled bulk campaign configuration and negative keyword imports. Google Merchant Centre required a product feed audit — 340 SKUs had missing GTIN data, which was fixed before launch to avoid feed disapprovals during the peak. Google Analytics 4 was configured with enhanced e-commerce events to give the AI Max bidding model clean, granular conversion signals beyond just transactions, including add-to-cart and checkout-initiated events with appropriate conversion values.

One structural decision that proved critical in retrospect: the team set the initial tROAS target at 3.2x — only modestly above the previous blended 2.8x — rather than immediately targeting the aspirational 5x figure. This gave the model enough room to gather data without being constrained into low-volume, hyper-selective bidding from day one. The target was stepped up in 0.4x increments over weeks six through eight as volume held.

Total migration cost: approximately 28 hours of external consultancy time plus 14 hours of internal resource across the eight weeks.

Results: Before-and-After Metrics Across the Q4 Peak

The comparison window used is Q4 2024 (October–January, legacy structure) versus Q4 2025 (October–January, post-AI Max migration). This is a like-for-like seasonal period, which controls for the natural revenue uplift Q4 typically produces in apparel e-commerce.

The wasted spend reduction of 34% — the headline figure — came from two sources in roughly equal measure: the consolidated negative keyword list eliminating low-intent traffic, and the AI Max bidding model itself becoming progressively more selective about which auctions to enter as conversion signal quality improved over weeks six through ten.

The ROAS improvement of 2.1x above the starting point was not uniformly distributed across the three campaigns. The Women's campaign outperformed significantly, reaching 7.2x ROAS by January. The Kids' campaign was the laggard, finishing at 4.1x — still a meaningful improvement over the legacy 2.8x blended figure, but one that required additional negative keyword work in week seven when the model began surfacing irrelevant gifting-related queries.

"The most surprising result was not the ROAS lift — it was that Shopping impression share jumped 23 percentage points without any increase in CPCs. The model was simply entering better auctions."

Key Learnings: What Worked, What Failed, and What Nobody Expected

What worked decisively: The phased budget scaling approach — starting at 40% allocation and stepping up — was the single most important structural decision. It prevented the model from entering a cold-start data vacuum with a large budget, which industry practitioners frequently flag as the primary cause of poor early AI Max performance. The enhanced e-commerce conversion signals (not just purchase events) also proved material: the model's bidding improved noticeably in week eight after add-to-cart signals hit sufficient volume.

What failed and required correction: The Kids' campaign required a manual intervention in week seven. The AI Max model had begun allocating significant budget to broad seasonal gifting queries — terms like "Christmas gifts for kids under £30" — which generated clicks but virtually no apparel-specific conversions. This was addressed with a 34-term negative list addition and a tROAS floor adjustment. It resolved within five days, but it underscores that AI Max is not a set-and-forget system, particularly in categories with seasonal search behaviour drift.

What nobody expected: The reduction in PPC management hours — from 22 to 9 per month — was larger than projected. The original estimate had been a reduction to around 14–15 hours. The consolidation from 47 ad groups and 14 Shopping campaigns to three AI Max campaigns eliminated an enormous volume of routine maintenance work. This freed the in-house PPC lead to spend time on feed optimisation and landing page testing, both of which contributed incremental gains beyond the AI Max migration itself.

One finding worth flagging for teams considering a similar migration: the brand exclusion configuration took longer than expected. Ensuring that AI Max was not cannibalising branded traffic required careful setup of brand exclusion lists, URL exclusions for competitor pages, and ongoing monitoring of the "brand vs. non-brand" split in the Search Terms report for the first six weeks. Teams that skip this step frequently report inflated ROAS figures that include branded conversions the algorithm has claimed credit for — obscuring the true incremental performance picture.

How to Replicate This: An Actionable Migration Checklist

The following checklist is sequenced to match the eight-week implementation structure used in this case. Adjust timelines based on account complexity and catalogue size.

Phase 1: Audit and Prepare (Weeks 1–3)

• Run a 90-day search term report and identify all terms with spend and zero conversions — these form the foundation of your consolidated negative list.
• Audit your Merchant Centre feed: fix missing GTINs, resolve disapprovals, and ensure custom labels are configured to allow campaign segmentation by margin or category.
• Map your product catalogue to campaign groupings — aim for three to five AI Max campaigns maximum, segmented by meaningful business logic (category, margin tier, or audience intent level).
• Create a dedicated branded keyword campaign (exact match only) with its own budget before AI Max goes live. Configure brand exclusions in AI Max from day one.
• Build your asset library with human review: reject any AI-suggested creative that doesn't meet brand tone-of-voice standards.

Phase 2: Launch and Monitor (Weeks 4–6)

• Set your initial tROAS target no more than 15–20% above your current blended ROAS. Aggressive targets from launch starve the model of learning volume.
• Launch at 40–50% of your intended budget allocation. Scale only once you have at least 30–50 conversions per campaign per week.
• Review Search Term reports three times per week for the first four weeks. Add negatives promptly — don't wait for the weekly review cycle.
• Ensure GA4 enhanced e-commerce events are firing correctly and that micro-conversion signals (add-to-cart, checkout-initiated) are visible to the bidding model.
• Monitor the brand vs. non-brand traffic split weekly to confirm brand exclusions are holding.

Phase 3: Scale and Optimise (Weeks 7–8+)

• Step tROAS targets up in 0.3–0.5x increments once volume stabilises. Do not increase targets and budgets simultaneously — change one variable at a time.
• Review asset performance at the two-week mark and pause any asset rated "Low" for two consecutive weeks. Replace with new variants.
• Run a cannibalism check: compare branded CPC trends before and after launch to confirm AI Max is not inflating brand costs.
• Schedule a full account review at the eight-week mark. Assess which campaigns are underperforming relative to tROAS targets and investigate root causes before scaling further.

Frequently Asked Questions