Performance max for ecommerce: the complete guide

You control five things: which products are in, what assets the system can use, what goal it optimises toward, how much it can spend, and a small set of exclusions. Everything else is decided for you, auction by auction.

It helps to be blunt about what that means. You do not choose whether a given impression appears in search results, on a video, in an inbox, in a feed or on a partner site. You do not set separate bids for those placements. You cannot see a full breakdown of where every pound went. The system optimises toward the goal you set, using the assets and products you supplied.

The upside is that a single campaign can follow a shopper across very different contexts, and it will find combinations of product, creative and placement that a person structuring campaigns by hand would never have tried. The downside is that when results are poor, the diagnosis is harder, because the usual questions cannot be answered directly.

This changes where your effort goes. In a campaign type with full control, most of the improvement comes from structure and bids. Here, most of it comes from the quality of the inputs. Better product data, better images, better headlines, a cleaner goal and a sensible product split are the levers. Fiddling with the few settings available is not.

One behaviour worth knowing early: within the same account, this campaign type takes precedence over a standard Shopping campaign holding the same products. You cannot run both on the same items and expect them to share. If you want some products handled with full control, take those products out of this campaign entirely rather than hoping the two will balance.

Finally, the goal you set is not a suggestion, it is the instruction. If you tell it to maximise conversions, it will find the cheapest conversions, which in a store with a wide price range means it will sell your least valuable products enthusiastically. Value-based goals exist for a reason.

This campaign type suits stores with a working feed, enough conversion volume for automation to learn from, and a catalogue where the individual products do not need individual handling. It suits some stores badly, and it is cheaper to work that out before launch.

It tends to work when your catalogue is broad, your margins are reasonably consistent, your product data is accurate, and you have enough orders that the system gets a usable signal within a few weeks. Under those conditions, handing over placement decisions genuinely finds volume that manual structures miss.

It tends to struggle when order volume is very low, because there is not enough feedback for the optimisation to work with and the campaign spends a long time being expensive while it learns. It struggles when margins vary wildly across the catalogue and you have not separated them, because the system optimises revenue, not your profit. It struggles when a large share of your conversions are people who already knew your brand, because those cheap conversions look like success and the campaign will chase them.

There is also a practical question about what you are trying to learn. If you do not yet know which searches your catalogue attracts, what a click is worth in your category, or which products convert, a campaign type that hides those answers is a poor place to find them. Many stores are better served by running a controlled campaign first for a season, taking the knowledge, and then handing the long tail over.

A reasonable default for a growing store is a split. Products that need judgement stay where you can see and steer them. The rest go here, where the volume is and the per-product attention is not worth the hours. That split should follow margin and importance, not catalogue structure.

If you decide against it for now, say so explicitly and write down what would change your mind, such as a certain level of monthly orders. Otherwise the question gets reopened every quarter with no new information.

Get four inputs right before you create the campaign, because changing them afterwards restarts the learning and wastes the spend that produced it.

The first is the feed. Everything about matching still runs through your product data, so titles, images, product types and attributes decide which searches your products can reach. If the feed is weak, this campaign type will not save you. It will simply spend on whatever your data lets it reach.

The second is conversion tracking with real order values. Value-based optimisation needs to know that one order was worth four times another. If your tracking sends a flat value, or if it double counts, or if it fires on the wrong page, every decision the system makes is built on that error. Verify the numbers against your store's own order records before launch, not after.

The third is the goal. Decide whether you are asking for maximum conversion value with no constraint, or value at a specific return target. Calculate the target from your own margin and set it deliberately. If you also care about new customers rather than repeat buyers, there is a setting for that which lets the system treat a first-time buyer as worth more, and it is worth using where your repeat rate is strong enough that a new customer is genuinely more valuable than one order suggests.

The fourth is your exclusion list. Products you never want advertised, countries you cannot ship to, and the searches that are never right for you. Building these in from the start avoids the pattern where the first month is spent discovering waste that was predictable.

One more piece of preparation: decide how you will judge it, and over what period, before it launches. Write the number down. Campaigns of this type look terrible in their first two weeks by design, and a decision made in that window is almost always the wrong one. Knowing in advance how long you will wait removes the temptation.

Group your assets by the way a shopper thinks about your products, not by the way your catalogue is filed. An asset group bundles a set of products with the creative that suits them, and the system mixes those pieces to build ads. If the products and the creative in a group do not belong together, every combination it builds is a compromise.

The practical rule is that everything in one asset group should be describable by the same headline. If you sell running shoes and cast iron pans, one group cannot say anything true about both. If you sell running shoes for road and trail, one group might work, or two might work better if the buyers are genuinely different people.

Inside a group you supply headlines, longer headlines, descriptions, images in several shapes, a logo and video. Fill every slot properly rather than repeating one line in different lengths. The system tests combinations, and it can only test what you gave it. Thin asset groups produce thin ads.

Write the text for the shopper, not for the product manager. Headlines that name the specific benefit, the specific use, or the specific problem beat headlines that describe a category. Include the practical facts that decide a purchase: delivery speed, returns terms, range breadth, what makes yours different. Avoid claims you cannot support and avoid anything that will be untrue in three months.

On video, supply your own. If you leave the slot empty the system may build something from your other assets, and what it builds is rarely what you would have chosen. Even a simple product video shot on a phone, showing the item in use and cut short, does better than an automatically assembled one.

Images need variety in more than shape. Give it a clean product shot, a product in use, a detail shot and something showing scale. Different placements suit different images, and the only way it can pick well is if you gave it real options.

Refresh assets on a schedule rather than when performance drops, because by the time the decline is visible you have already paid for it.

Separate products that behave differently into different campaigns or asset groups, because a single campaign reports one blended result and will quietly fund your weakest line with your strongest one's success. This is the structural decision that most affects profit.

The three splits worth considering are margin, price band and product line. Margin matters because a return target that is comfortable on a high-margin item loses money on a thin one. Price band matters because a campaign told to maximise value will drift toward whatever is easiest to sell, and if that is your cheapest range you end up with volume and no profit. Product line matters when different products attract genuinely different buyers who need different creative.

Use custom labels in the feed to carry these distinctions, then build listing groups from them. Doing the classification in the feed rather than in the campaign means it survives rebuilds and can be regenerated from your sales data on a schedule.

Resist splitting too far. Every split divides the conversion data, and a campaign without enough conversions to learn from performs worse than a slightly blunt one that has signal. For most stores, two or three campaigns is the right number: one for the products that carry the business, one for everything else, and sometimes one for a line with distinct economics.

When you do split, set different targets rather than the same target everywhere. That is the entire point of splitting. If every campaign carries the same goal, you have added complexity without adding control.

Watch the interaction with anything else in the account. Products in this campaign type will win over the same products in a standard Shopping campaign, so a product should live in one place deliberately. Audit this after any feed change, because a new product that lands in the wrong bucket can spend for weeks unnoticed.

Finally, keep new products out of a mature campaign's asset group until they have images and data good enough to compete, or put them somewhere they can be watched. New items dropped into a large campaign often get no exposure at all.

Supply audience signals and search themes as hints about where to start, then set exclusions to keep the campaign out of places it should not go. Signals do not restrict the campaign. Exclusions do, and the difference matters.

Audience signals tell the system who has bought from you before and who resembles them. Your customer list, your site visitors, people who viewed but did not buy, and interest groupings that match your buyers are all useful starting points. The system uses them to make its early guesses better, then goes wherever the conversions are. Feeding it a good customer list is one of the more valuable things you can do, and it costs an export.

Search themes work the same way for queries. They tell the system what kinds of searches you believe are relevant, particularly for products or use cases your feed does not describe well. Use them for genuine gaps rather than for terms your product data already covers.

Exclusions are where you get real control. Brand exclusions keep the campaign off searches for your own name, which matters because those conversions are cheap and would mostly have happened anyway. Left unchecked, the campaign will spend there because it looks efficient, and your reported results will improve while your actual growth does not. Google provides brand controls and account-level negative keyword lists, and the exact location of these settings moves around, so check the current interface rather than assuming.

Also constrain where the campaign can send people. If the setting that lets the system pick its own landing pages is on, it can send traffic to any page on your site, including ones that were never meant to receive it. Turn that down or restrict it to the pages you want.

Geography and language deserve a deliberate decision too. Exclude places you cannot serve profitably, and check whether the location setting is targeting people in a place or people interested in a place, because those are very different audiences.

Launch with the budget you intend to run and then do not touch the campaign while it settles, because every meaningful change restarts the process you are waiting on. The most common way stores waste money here is by intervening during the period the campaign is least able to perform.

Set the daily budget at a level that can produce a readable number of conversions in a reasonable time. A budget too small to generate consistent orders never leaves the unsettled state, and the campaign will look like a failure that was never given the chance to be anything else. If your budget cannot support that, this is not the right campaign type yet.

Expect the first stretch to look bad. Cost per order is high, the product mix looks random, and the return figure is below target. That is the system exploring. Judge it against the period you wrote down before launch, not against last week's other campaign.

During that period, allow yourself only two kinds of intervention: fixing something that is factually broken, such as a disapproved product or a tracking error, and stopping the campaign entirely if spend is running somewhere clearly wrong. Do not adjust the target. Do not swap assets. Do not restructure.

Afterwards, change one thing at a time. Target adjustments should be modest, because a large move sends the campaign back into exploration and you lose the stability you just paid for. Budget increases are gentler than target changes in their effect, so if you want more volume, that is usually the safer lever first.

Keep a dated log of every change. Because this campaign type shows so little of its workings, your own record of what you did and when is often the only way to explain a shift in results three months later. Write down the date, the change and the reason, and check it before you conclude that something the system did was responsible.

Build your own reporting layer, because the campaign will not hand you the diagnosis. There is more available than the default view suggests, and combining it with your own store data closes most of the gap.

Start with what the platform does show. Asset group performance tells you which combination of products and creative is carrying the campaign. Listing group reporting shows spend and revenue by product, which is the most useful view in the whole account, because it lets you see whether the campaign is selling what you want it to sell. Asset-level ratings tell you which headlines and images are being used and which are being ignored. Search category insights give a rough sense of the query themes involved, without the detail a search terms report would give.

Then add your own numbers. Export spend and revenue by product, join it to your cost of goods and your return rate, and calculate contribution after ad spend for each line. That is the number the campaign cannot give you and the only one that answers whether it is working.

Watch the product mix over time as its own metric. A campaign drifting toward cheaper or thinner-margin items is a common failure that never shows up in the headline return figure, because those items still convert. Tracking the share of spend going to your top-margin products catches it early.

Compare against a baseline outside the campaign. Total store revenue, new customer count and blended spend efficiency across all channels tell you whether the campaign added anything or moved existing demand around. This matters more here than elsewhere, because a campaign that reaches every placement is well positioned to claim credit for orders that were coming anyway.

Set a monthly reporting rhythm and keep the same view each time. The value of this reporting comes from the trend, and a report rebuilt differently every month has no trend in it.

When results need to improve, work through the inputs in order of impact rather than adjusting the target and hoping. There are five moves worth making, and the target is the last of them.

First, fix the feed. Products with no impressions, weak titles, poor main images or missing attributes are limiting the campaign's reach before any bidding happens. This is usually the largest available gain and it is entirely under your control.

Second, replace the weakest assets. Look at which headlines and images the system is actually using and which it has effectively ignored, and replace the ignored ones with genuinely different options rather than variations on the same idea. Creative refreshes matter more than they look here, because the system's ability to find a good combination depends on having good pieces.

Third, revisit the product split. If the campaign is selling the wrong mix, the fix is usually structural: move the products you want prioritised into their own campaign with their own target, so their performance is no longer averaged with everything else.

Fourth, tighten the exclusions. Brand traffic, irrelevant search themes, products that should not be advertised, geography you cannot serve. Each of these silently improves the average without any change to the campaign's real reach.

Fifth, and only then, adjust the target. Do it in small steps and wait between them. If you need more volume, loosening the target adds spend at lower efficiency by design, so decide in advance whether you want the extra gross profit at that lower return.

If several rounds of this produce nothing, consider that the campaign type may not fit the catalogue. That is a legitimate conclusion, not a failure. Moving the products into a structure you can steer directly is sometimes the right answer, particularly for stores where a small number of products carry most of the profit and each one deserves individual attention.