Ecommerce is where Google Ads problems compound. Performance Max hides its queries, the feed breaks silently, brand traffic leaks into paid results you were getting free, and Q4 rewrites the rules of every campaign for six weeks. An AI copilot connected to the live account does not remove the complexity — it reads and runs the whole account on your command, in seconds, so one person holds all of it, weekly, without the account running them.
The ecommerce triad: PMax, Shopping, and brand protection
Three structures carry most stores. Performance Max takes the catalogue everywhere Google serves ads, and tells you little about how. Standard Shopping keeps control and visibility where margins deserve it. And somewhere underneath, your brand terms — the cheapest, highest-converting traffic you own — need defending from your own campaigns as much as from competitors. Managing the triad means reading across all three every week: what PMax absorbed, what Shopping converted, what brand actually cost. That cross-read is tedious by hand and one prompt for an agent with live account access.
Brand leakage is a margin eater — check it weekly
The quiet failure: PMax or a broad-matched Search campaign starts answering brand queries that your brand campaign — or organic — used to catch. Nothing alarms. ROAS even looks good, because brand always converts. But you are paying auction prices for customers who typed your name. The weekly check is mechanical: where did brand queries get served, at what cost, and is the brand campaign losing impression share to a sibling? The agent runs the read, drafts the negative keywords that wall brand off from non-brand Search, and verifies your PMax brand exclusions are holding. You approve the additions in one click. The weekly PMax review has a fuller checklist; leakage is its first item for a reason.
Seasonal operations: pacing without the whiplash
Ecommerce budgets are not lines — they are curves with cliffs. The agent's job across a season: track pacing against the month's plan, flag campaigns that will exhaust budget before the weekend, build sale campaigns ahead of time as paused drafts that cannot spend, and propose the launch-day budget raises with last season's numbers attached. When the sale ends, the revert is another one-click proposal, not a note you hoped to remember. It will also flag when demand shifts are sharp enough that a seasonality adjustment is worth setting, so Smart Bidding is not surprised by your best week. The Black Friday preparation guide turns this into a dated checklist.
SKU-level questions your dashboard answers slowly
The questions that matter in ecommerce are product-shaped, and dashboards make you dig for them. Which products spent the most this month without a single sale? Which SKUs converted last quarter but have no impressions now — availability problem or bid problem? What share of PMax spend went to the ten products that cannot make money at current CPCs? Asked in chat, each of these is a live read of the account, answered in a minute with the product IDs listed. No approval needed, because reading changes nothing. The follow-up actions — pausing a bleeding campaign, rebalancing budget toward what sells — arrive as proposals with the evidence already attached; approve, and the agent applies the change in seconds.
ROAS targets by margin band, not account-wide
One account-wide target ROAS is a polite fiction. A store selling both accessories at strong margins and electronics at thin ones needs different targets per campaign, derived from margin, not from habit. Walk the agent through your margin bands once, and it can propose per-campaign targets that translate "profitable" into the number bidding actually optimises for — then check monthly whether performance justifies moving them. Setting a target is a bidding change, and the connector makes it on Search and Performance Max campaigns as a proposed write: the agent argues the number and the reasoning in the thread, where you can interrogate it, and the target changes only after your yes. A standard Shopping campaign's target is still set in the Google Ads UI. Each move resets some of what bidding has learned, so move targets in considered steps, not weekly.
The peak-season daily check in five minutes
From November onward, weekly is not enough and daily is unaffordable by hand. The five-minute version, in one conversation: today's spend and revenue against plan; campaigns hitting budget caps; disapprovals or product drops since yesterday; anything spending with zero conversions since the sale started; pacing forecast for the week. Five reads, no approvals, done before the second coffee. The agent surfaces exceptions; you spend judgement only where something moved.
Where the approval line sits for a store
Reads change nothing — allow them once (in Claude, set the read-only tools to Always allow) and the daily check and the SKU questions stop prompting. Negatives and paused campaign builds are quick approvals: they cannot spend. The firm line covers what moves money during your highest-stakes weeks: budget raises, enabling sale campaigns, pausing anything live. Each proposal shows the exact change and the evidence, and you click once. Deletion is not gated — it is absent: the connector's remove tools are never exposed, and a REMOVED mutation is refused server-side. During Q4, that distinction is the difference between delegating and gambling.
Where its reach ends: feed and storefront
It does not see or edit Merchant Center — feed titles, GTINs, disapprovals and promotions are fixed at the source, with the agent pointing you to which fixes matter most. It does not touch your storefront, your pricing or your inventory system. It reads Google Ads live; Shopify or WooCommerce analytics are not in its view, so margin data enters through you. And it cannot make PMax fully transparent — nobody can — it can only extract everything the API reports and make the opacity legible enough to manage.
First week on an ecommerce account
Start where the money is. Day one: connect and run the full account audit — expect the geo settings, brand leakage and spend-without-sales findings to pay for the week by themselves. Day two: the margin conversation, ending with ROAS targets per band. Days three to five: approve the negative sweep, review a paused rebuild of your weakest campaign, and run the first weekly triad read. The trial is 7 days on the full Pro plan for a new workspace, no card — plenty of room for one store before peak season, and it drops to the Free plan afterward rather than cutting off. Start free while the season is still yours to plan.