AI Advertising Optimization Checklist for Every New Account

Start here, always. Automated bidding treats every recorded conversion as truth. Duplicate events, test purchases and soft actions counted as primary conversions all teach the system to chase the wrong thing.

• In Google Ads, open Goals, Conversions, Summary and check which actions are set as primary. Only real business outcomes should be primary.
• In Meta Events Manager, check event match quality and confirm the Conversions API deduplicates with the pixel.
• Compare platform conversions with backend orders or CRM leads for the last 30 days.
• Remove or downgrade events like page views, scroll depth or time on site as optimisation goals.
• Enable enhanced conversions in Google Ads and advanced matching in Meta where your privacy setup allows.

A gap between platform conversions and real orders usually points to tracking setup, not to the platform being wrong. Fix it before changing bids.

Value-based bidding, such as Target ROAS in Google Ads or highest value in Meta, optimises toward the values you send. If every lead is worth 1, the system cannot tell a large enterprise prospect from a student download.

For ecommerce, send order value, and consider sending gross profit instead of revenue if margins vary widely across products. For lead generation, assign values by lead type and import offline conversions from your CRM so the system learns which leads became customers.

An example of value signals for lead generation

For example, a B2B software company might assign a value of 50 to a newsletter sign-up, 300 to a demo request, and then import a closed deal from HubSpot with its real contract value weeks later. Google Ads and Meta can both accept these offline uploads. The values here are placeholders; derive yours from close rates and average deal size.

AI bidding needs volume per campaign to learn. Twenty campaigns with a few conversions each give the system little to work with, while three or four consolidated campaigns with clear goals give it plenty.

• Group campaigns by goal and margin, not by every product line or keyword theme.
• Avoid overlapping campaigns that bid on the same users or queries.
• Use shared budgets or campaign budget optimisation only where goals match.
• Keep a separate brand search campaign so brand queries do not inflate results elsewhere.
• Document naming conventions so reports read clearly.

Consolidation is not always the answer. If two product lines have very different margins, separating them lets you set different targets.

Systems like Performance Max and Advantage+ assemble ads from your assets. If you give them three images and two headlines, they have little to test. Give them a full set.

• Google Performance Max: up to 15 headlines, 5 long headlines, 5 descriptions, images in landscape, square and portrait, logos and at least one video.
• Meta Advantage+: several images and videos in different aspect ratios, multiple primary texts and headlines.
• Product feeds in Google Merchant Center and Meta Commerce Manager with accurate titles, prices, availability and images.
• Feed titles written with the words people search for, not internal product codes.

If you do not supply a video for Performance Max, Google may generate one from your images, and those auto-generated videos often look basic. Upload your own.

Automation needs limits. Without them, a campaign can drift into irrelevant searches, low-quality placements or audiences you never meant to reach.

• Negative keywords at account level for search and, where available, for Performance Max.
• Brand exclusions in Performance Max if you want brand queries handled elsewhere.
• Placement exclusions and content suitability settings in Google and Meta.
• Audience exclusions for existing customers when the goal is new customer acquisition.
• Target CPA or ROAS set from historical data, not from hope.
• Budget caps and automated alerts for spend spikes.

New automated campaigns go through a learning period. Google suggests allowing time for Smart Bidding to calibrate after major changes, and Meta shows a learning phase status on ad sets until they gather enough optimisation events. Judging results in the first few days usually leads to bad decisions.

Set a review date before launch, typically two to four weeks out depending on conversion volume. Avoid large changes to budget, targets or creative during that window, since each one can restart learning.

What to watch during learning

• Spend pacing: is the campaign spending as expected?
• Search terms report and placement report for obvious waste.
• Conversion lag: some conversions arrive days after the click.
• New customer share, if you track it.

Run controlled tests where possible. Google Ads offers experiments for bidding strategy changes, splitting traffic between the original and the variant. Meta offers A/B tests in Experiments. Compare cost per conversion, conversion volume and, where possible, profit.

Also look at incrementality. Automated campaigns are good at finding people who were going to buy anyway, especially through retargeting and brand queries. A conversion lift study, available on Meta and Google for qualifying spend, or a simple geographic holdout test, gives a better view of whether ads create new demand.

Keeping a change log

Write down every change to automated campaigns with a date: new target, new assets, a feed update, a budget move. Automated systems react to many inputs at once, and without a log it becomes impossible to explain why results moved. A shared spreadsheet is enough. Note what changed, why, who made the change and the date to review its effect.

When a result shifts sharply, the log is the first place to look. In many accounts, a sudden drop traces back to an unrelated change such as a site release that broke a tag, or a feed error that disapproved products, rather than to the bidding system itself.