Google Ads Performance Max vs. manual control: When AI burns your budget (and how to prevent it)

There is currently one dominant buzzword in the world of digital marketing: automation. Google is aggressively pushing advertisers towards „Performance Max“ (PMax) - a type of campaign that promises to deliver better results with less effort with the help of artificial intelligence (AI). However, for many Swiss SMEs, this „black box“ is not proving to be a growth driver, but a budget killer. In this analysis for 2025 and 2026, we show when you can trust the machine and when human control remains indispensable.

The illusion of the „perfect campaign“

The premise of Performance Max sounds tempting: you give Google a budget, a few texts, images and a goal (e.g. „sales“ or „leads“). The AI does the rest. The algorithm decides independently whether your ad will be displayed in Google Search, on YouTube, in Gmail, on Google Maps or in the huge display network.

This is an ingenious model for Google. It fills advertising space that no one else would buy. But is it also brilliant for your company?

The data from our audits at mirostudio.ch show a differentiated picture: While PMax often works brilliantly for large e-commerce brands with thousands of data points, it often leads to massive waste for service providers and smaller budgets. Let's take a look under the hood.

The problem of the „black box“ (black box)

The biggest criticism of Performance Max is the lack of transparency. In contrast to traditional search campaigns, Google does not tell you exactly what you are doing, where your budget has been spent.

Did you pay CHF 500 for high-quality clicks in search? Or did CHF 300 of that go to YouTube channels for children and mobile gaming apps where users accidentally click on banners? Google summarizes this in nebulous categories.

If you don't analyze this iceberg, you are making business decisions based on distorted data.

The 3 biggest budget killers from Performance Max

If you run PMax without strict guardrails, the AI often optimizes for the path of least resistance. This leads to three classic problems:

1. the cannibalization of your own brand (brand cannibalization)

The algorithm is programmed to deliver conversions. The easiest way to do this is to serve ads to people who are already searching for your company name (e.g. „Miro Studio Zurich“). These users are „hot“ and almost always convert. The result: The campaign looks fantastic in the reporting. The reality: You pay for clicks that you would probably have received organically for free. You do not acquire new customers, but buy from your existing customers.

2. the display network and the „sausage finger“ clicks

Have you ever tried to click away an advertisement in an app on your cell phone and accidentally opened it instead? For Google, this counts as a click. For PMax, that counts as a signal: „This user is interested!“ Without manual intervention, PMax tends to push budget to the display network because the click prices (CPC) there are very low. You get a lot of traffic, but the quality is often inferior.

3. the zombie leads (spam)

This is a particular plague in the B2B sector. PMax finds placements where bots fill out forms. Google sees „Conversion“ and thinks: „Great, we'll get more of that!“. Suddenly your CRM is full of requests from „Lksdfj Lsjdkf“ or emails that are obviously fake. If you don't tell Google that these leads are worthless (via offline conversion tracking), the AI learns to scale exactly this garbage.

Manual control (Search Only): When the human wins

Despite the AI hype, there will still be scenarios in 2025 in which a classic, manually controlled search campaign (search only) is vastly superior to automation.

Scenario A: The budget is limited (< CHF 3,000/month)

AI needs data to learn. Google recommends at least 30-50 conversions per month for PMax to work. If your budget or search volume doesn't allow for this, the AI is just guessing. A precise „phrase match“ search campaign is more efficient here.

Scenario B: B2B and complex services

If you offer a specialized service (e.g. „industrial architecture“ or „special software for logistics“), you don't want to appear on YouTube between music videos. You want to be visible exactly when a decision-maker googles the specific problem. The scatter loss of PMax is often too high in the B2B sector.

Scenario C: Strict lead quality

With manual campaigns, you can precisely control negative keywords (exclusion criteria). You can prevent your ad from appearing for search queries such as „free“, „cheap“ or „jobs“. With PMax, this is only possible via detours.

Strategy 2026: How to tame the „beast“

We can't turn back time. Google will continue to restrict manual options. The solution for forward-looking companies is therefore not complete denial, but the hybrid strategy.

When we set up PMax campaigns for clients at Miro Studio, we never do it „out of the box“. We use advanced techniques to guide the AI:

1. brand exclusion lists (brand exclusion)

This is the most important step. We are forcing PMax, not to bid on your own company name. This forces the algorithm to search for new users, even if this initially lowers the ROAS on paper. It is the only honest key figure.

2. audience signals (target group signals)

Don't let the AI guess. We feed PMax with your own data (first-party data): Lists of your best customers, newsletter subscribers or website visitors. We tell the AI: „Search for people who so look like this one.“ Without these signals, PMax is shooting in the dark.

3. feed-only PMax (for e-commerce)

An insider tip for Online storesPMax can be configured so that it does not use videos or generic images, but concentrates purely on the product feed. This eliminates the expensive playouts on YouTube and in the display network and focuses the budget on the shopping tab, where the purchase intention is highest.

4. offline conversion tracking (OCT)

To solve the problem of zombie leads, we connect Google Ads with your CRM. When a lead comes in, Google reports „Conversion“. But only when your sales team marks the lead as „Qualified“ do we send this signal back to Google. This is how the AI learns: „Don't bring me more people who fill out the form, but people who actually buy.“

Conclusion: Trust is good, control is more profitable

In 2026, the role of the marketing manager or agency will no longer consist of manually booking keywords. The task will shift to „Algorithm training“.

PMax is like a very fast sports car without a steering wheel. If the road is straight (lots of data, simple product), you get there quickly. But in the bends (complex B2B markets, small budgets) you fly off course without an experienced pilot.

Do not rely on Google's default settings. These are optimized to increase Google's sales, not necessarily your profits.

Do you want to know whether your Google Ads budget is really working or just wasting away? At Miro Studio, we look deep into your campaign structure. We identify the hidden costs in the „PMax iceberg“ and build a strategy that uses AI but doesn't eliminate common sense.

Short version for those in a hurry (key takeaways):

• PMax is standard: Google Performance Max is powerful, but a „black box“ with little transparency.

• Beware of whitewashing: PMax tends to use brand traffic and retargeting to boost the statistics.

• Display trap: Without control, budget flows into inferior mobile app placements.

• B2B risk: For service providers with small budgets (< CHF 3000), manual search campaigns are often still more efficient.

• The solution: Use brand exclusions, your own customer lists (first-party data) and offline tracking to train the AI on real business goals.