What AI automation handles now (and what it cannot)
The automation boundary in ecommerce advertising has shifted materially since 2023. Platforms now handle bidding, audience targeting, creative selection, and cross-channel budget allocation as defaults rather than optional features. Manual campaign management is declining in most channel categories because human bid management cannot process as many auction-level signals as the platform's AI.
What platforms now handle automatically
- Bid optimisation -- Smart Bidding on Google and equivalent systems on Meta and Amazon adjust bids at auction level using conversion probability signals no human can match in real time
- Audience targeting -- Advantage+ Shopping on Meta and Performance Max on Google build their own audience segments from conversion signals; you do not set targeting manually
- Creative rotation and testing -- both platforms serve creative variants and suppress underperformers automatically
- Cross-surface placement -- Performance Max spans Search, Shopping, YouTube, Display, Gmail, and Discover from one campaign; ad delivery allocation is AI-managed
What still requires human expertise
- Product feed architecture -- The quality of your product data feed determines the ceiling of AI performance. Title structure, attribute completeness, category mapping, and price accuracy are human decisions made once that affect AI output for months.
- Conversion event hierarchy -- Defining which conversion events to optimise toward (add-to-cart, checkout initiation, purchase, repeat purchase) and what relative weight each carries is a strategic decision, not an algorithm output.
- Incrementality testing design -- No platform AI tests whether its own channel is incremental. Geo-holdout tests, conversion lift studies, and multi-touch attribution models require external design and execution.
- Channel budget arbitration -- When Google, Meta, and Amazon all report positive ROAS, no algorithm tells you how to allocate across them. That arbitration requires incrementality data the platform does not share.
- Margin-aware optimisation -- Platform AI optimises toward revenue (or proxy signals). Brands with variable product margins need to feed margin data as custom signals or segment campaigns by margin tier manually.
leapbuzz's AI performance marketing service for ecommerce starts with the attribution architecture -- defining the right conversion hierarchy and incrementality testing methodology before touching campaign structure or creative.
Google Performance Max for ecommerce: feed quality is the actual product
Performance Max (PMax) is Google's unified AI campaign type that uses your product feed, creative assets, and audience signals to serve across every Google surface. For ecommerce, Shopping inventory within PMax is typically the primary driver of volume, with other surfaces adding reach.
The critical insight that most agencies underemphasise: PMax performance is gated by product feed quality. The AI selects products, writes responsive ad combinations, and allocates budget across surfaces -- but all of those decisions flow downstream from the structured product data you provide. A feed with incomplete attributes, stale pricing, missing images, or incorrect availability codes means the AI is optimising on bad inputs.
Feed audit before campaign audit. If a PMax campaign is underperforming, the first diagnostic is the product feed -- not the bid strategy, not the creative assets, not the campaign settings. Product disapprovals, low search impression share on high-intent queries, and poor Shopping placement often trace to feed attribute problems, not campaign configuration.
What a capable ecommerce partner manages in PMax
- Feed attribute optimisation: product titles with high-intent modifiers in the right position, supplemental feeds for sale pricing and custom labels
- Campaign segmentation by product margin tier or strategic category (so high-margin products can have a different target ROAS from commodity SKUs)
- Asset group structure to ensure relevant creative reaches relevant product categories
- Audience signal lists built from first-party data (purchasers, high-value customers, email list) to accelerate learning
- Negative keyword lists applied at account level to filter irrelevant traffic PMax's broad matching surfaces
- Search term transparency: extracting search term data from Insights and using it to inform both negative lists and organic content strategy
leapbuzz's search advertising service includes product feed management as a core deliverable for ecommerce clients, not an optional add-on. Feed quality work directly precedes any campaign optimisation engagement.
Meta Advantage+ Shopping: where it wins and where it loses control
Meta's Advantage+ Shopping Campaigns (ASC) use AI to manage prospecting and retargeting audiences in a single campaign, allocating budget dynamically based on predicted conversion probability. For ecommerce brands with substantial catalog depth and strong pixel data histories, ASC often outperforms manually segmented campaigns on blended ROAS.
Where ASC performs well
- Brands with 5,000+ SKUs where manual audience-product matching is impractical
- Accounts with 18+ months of Meta pixel purchase data providing a strong signal base
- Seasonal products where the algorithm needs to shift prospecting and retargeting weights quickly
- DTC brands where most purchases are single or low-frequency, making retargeting pool management less critical
Where ASC creates problems
- Brands with high repeat purchase rates: ASC tends to allocate heavily toward existing customers (lowest cost per conversion) rather than new customer acquisition, inflating ROAS while new customer acquisition cost rises invisibly
- Brands with specific audience exclusions (competitor suppression, geographic exclusions) where ASC's automated audience expansion overrides manual targeting
- Brands running brand awareness objectives in parallel: ASC budget can bleed into top-funnel placements if the conversion signal is weak
The diagnostic metric is new customer acquisition cost (nCAC) tracked separately from blended acquisition cost. If ASC shows improving ROAS while nCAC rises, the campaign is optimising toward your cheapest audience (existing customers) rather than your highest-value outcome (new customers).
leapbuzz's paid social service for ecommerce clients separates nCAC reporting from blended ROAS from the first engagement, preventing the common pattern where Meta performance looks strong in platform reporting but retention economics tell a different story.
Retail media as a buyer: Amazon, Shopee, Lazada, and the margin question
Retail media networks sell advertising adjacent to product listings on their platforms -- effectively, shelf placement for digital products. For ecommerce brands, the question is not whether retail media works (it does, in the sense that it drives purchases) but whether the margin it returns justifies the spend, independently of your Google and Meta budgets.
The structural problem with retail media accounting is co-mingling. When a brand buys Sponsored Products on Amazon, Shopee, or Lazada and also runs Google Shopping and Meta, the purchase event is claimed by multiple channels. Without a holdout structure, the brand cannot determine which channel drove the incremental unit versus which channel claimed credit for an organic purchase.
| Network | Primary market | Ad product | Key consideration |
|---|---|---|---|
| Amazon Ads | US, CA, AU | Sponsored Products, Brands, Display, DSP | Highest intent signal; self-serve accessible; DSP requires $50k+ minimum. Marketplace fees plus ad spend must both fit in margin model |
| Walmart Connect | US, CA | Sponsored Products, on-site display, off-platform DSP (post-Vizio acquisition) | Growing grocery and general merchandise reach; VIZIO acquisition adds CTV data layer; competitive with Amazon for household brands |
| Shopee Ads | SG, MY, and broader SEA | Search Ads, Discovery Ads, Affiliate Marketing | Dominant in SG and MY for FMCG, fashion, and electronics. Commission plus ad spend model requires careful CAC modelling. No verified ROAS benchmarks from primary source available |
| Lazada Sponsored Solutions | MY, and broader SEA | Sponsored Products, Keyword Ads | Still significant in MY; lower market share than Shopee in SG following platform consolidation. Commission structure and ad spend both affect margin |
| Coles 360 / Cartology (Woolworths) | AU | On-site search, display, offsite programmatic | Both networks growing strongly (Cartology reported +19.5% FY, Coles 360 +13.5% FY per published reports). Primarily for CPG/FMCG brands stocked in Woolworths/Coles stores |
For brands in Singapore and Malaysia: published performance benchmarks for Shopee Ads and Lazada Sponsored Solutions comparable to Amazon's published data do not exist in verifiable form. Practitioners work from internal benchmark ranges. Any agency quoting specific ROAS figures for Shopee or Lazada without describing the category and price point context should be pressed on the source.
Our retail media network buyer's guide goes deeper on ROAS benchmarks by market, the incremental ROAS gap, and how to evaluate retail media as a distinct budget line.
AI shopping agents: why product data matters more than ad spend
A new layer of product discovery has emerged alongside traditional paid search: AI shopping agents. ChatGPT's shopping mode, Perplexity's product recommendation cards, and Google's Shopping Graph integrations now surface product recommendations in response to conversational queries. These are not ad slots -- they are editorial recommendations generated from structured product data.
The selection criteria these agents use to rank products differ from search ranking factors:
- Product data completeness: titles, descriptions, specifications, and attributes that are parseable by AI crawlers. Incomplete or inconsistent product data is invisible to these systems.
- Review signal density: volume and recency of reviews from third-party sources. Products with sparse reviews are underweighted in AI recommendations.
- Price and availability freshness: stale pricing data causes AI agents to recommend products that are out of stock or incorrectly priced, which platforms penalise.
- Structured markup: Product schema (Schema.org/Product) with complete price, availability, review aggregate, and brand markup makes products parseable without page rendering.
The implication for ecommerce marketing investment is that structural product data work -- feed optimisation, review acquisition programs, Schema markup -- now serves double duty: it improves paid shopping campaign performance AND it improves visibility in AI agent recommendations. A partner that treats feed management as a technical commodity rather than a strategic priority is undervaluing one of the two places where ecommerce product discovery is growing.
leapbuzz's AI visibility service includes ecommerce product discovery audits that assess Schema.org completeness, review signal gaps, and AI agent citation rates for product categories, alongside standard paid search management.
Budget allocation by channel: the toggle guide
The right channel split depends on your customer repeat rate, margin structure, and current attribution quality. These splits are practitioner starting points -- they should be tested and adjusted using actual incrementality data from your account.
DTC brands with low repeat rates prioritise prospecting on Google (high purchase intent) and Meta (broad reach with ASC). Retail media is kept lean to preserve margin until incrementality can be measured.
High-LTV subscription brands need new customer acquisition, not retargeting efficiency. Meta ASC is avoided because it over-weights existing customers. YouTube and CTV build the awareness layer that sustains subscription growth at scale.
Marketplace-first brands (primarily selling through Amazon, Shopee, or Lazada) concentrate media spend in-platform to defend search ranking and win buy-box position. Off-platform spend focuses on category awareness and new product launches.
Five-market ecommerce context
Singapore
Singapore's ecommerce market is compact geographically but high-value per transaction. Shopee and Lazada dominate FMCG and fashion volumes; DTC brands compete for a more affluent segment that shops through brand websites and on social commerce. Same-day and next-day delivery is table-stakes for most categories. Google and Meta CPCs are among the highest in APAC, reflecting the density of competition and purchasing power. The leapbuzz ecommerce-retail industry page covers sector-specific engagement approaches for Singapore-based brands.
United States
The most developed retail media ecosystem globally, with Amazon, Walmart Connect, Target Roundel, Kroger Precision Marketing, and Instacart all offering scaled advertising. PMax and ASC maturity is highest in the US market. Customer acquisition costs are high; LTV modelling and repeat rate management are essential to profitable growth. AI shopping agent penetration is growing fastest here.
Canada
Close to the US in platform mix and behaviour, with the added complexity of French-language requirements in Quebec (OQLF regulations mandate French advertising copy in certain categories). Amazon Canada is the dominant marketplace. Shopify's home market, so DTC infrastructure is advanced. CASL compliance applies to email marketing, with opt-in requirements stricter than CAN-SPAM.
Australia
Geography makes shipping economics central to any ecommerce model -- the cost of fulfilment from major city hubs to regional Australia affects viable product categories and price points. Afterpay's buy-now-pay-later penetration is among the highest globally and materially affects checkout conversion. ACCC consumer guarantees apply to digital purchases and affect return policy design and advertising claims. GST applies to imported goods over AUD 1,000 with the vendor responsible for collection, a compliance issue for cross-border DTC brands.
Malaysia
Shopee leads ecommerce volume; Lazada remains significant. Facebook and Instagram are primary social discovery channels. Mobile-first purchasing behaviour means desktop-optimised checkout flows underperform relative to regional benchmarks. SST (Sales and Services Tax) applies to most ecommerce categories at 8 to 10 percent since March 2024. Lower average order values than SG or AU mean that high-CPM advertising compresses margins quickly unless the product economics absorb the cost.
