Agentic AI: Performance Pilot in Google Ads 360 for 2026

In 2026, martech is all about intelligence, not just basic automation. The biggest thing happening in martech innovation is agentic AI, a massive change that lets systems reason, plan, and act on their own to hit goals in performance marketing. This whole shift changes what we think of as efficient and effective, but the real question is, how do you actually put it into practice?

Setting Up Your First Agentic AI Campaign in Google Ads 360’s Performance Pilot

The “Performance Pilot” module in Google Ads 360 (version 2026.3, from the Q1 release) is a huge jump forward for agentic AI. It’s built to take your high-level strategy and break it down into tiny, real-time campaign adjustments. You need a methodical setup process to make sure the AI actually understands your business goals.

1. Accessing Performance Pilot and Defining Campaign Scope

First, log into your Google Ads 360 account. Find “Agentic Tools” in the left-hand navigation and click it, then pick “Performance Pilot” from the menu. The platform will ask if you want to create a new campaign or apply these agentic strategies to one you already have running. We’ll start a new one for this walkthrough. Click the blue “New Agentic Campaign” button. You’ll get options for Search, Display, Video, and App campaigns. I always suggest starting with Search because it’s so structured, which gives the AI cleaner data to learn from.

Pro Tip: Don’t go all-in at first. Carve out a small, experimental budget for your first agentic campaign, maybe 10% of what you’d normally spend in a month, so you can watch how it learns without risking a ton of money while it’s still gathering data.

2. Configuring Performance Objectives in Goal Architect

After picking your campaign type, you’re sent to the “Goal Architect” interface. This is mission control for the AI. Goal Architect gives you a few presets like “Maximize Conversions” or “Target ROAS,” but for most of our performance work, you’ll want “Target CPA” or “Target ROAS.” Let’s pick “Target CPA”.

1. Specify CPA Target: Put in your target Cost Per Acquisition. If your business is profitable at a $50 CPA, type “50.00.” The system can work with a range, and I like to give it a 0.5% margin of error, which lets it bid up to $50.25 if it finds a really good conversion.
2. Define Conversion Actions: You have to tell it what a conversion is. Click “Select Conversion Actions” and choose the right events from your Google Analytics 4 property, like “Lead Form Submission” or “Purchase.”
3. Budget Allocation: Set your daily or monthly spend. Performance Pilot will handle the bidding and budget pacing to stay within that number. I’d suggest a starting daily budget of around $200 for a new campaign, which is usually enough for the AI to get its bearings without blowing the budget in the first week.

Common Mistake: People often try to force an unrealistically low CPA target from day one. This just starves the AI of data, so you either get no volume or really poor leads. Look at your historical campaign data to find a realistic starting point, and maybe even set it a little higher than you’d like so the AI has room to explore.

3. Providing Creative Assets and Audience Signals

The Performance Pilot needs raw materials. In the “Creative & Signals” section, you feed it ad copy and audience information. Here, the AI starts combining these pieces and testing thousands of variations, way more than a person ever could.

1. Ad Group Creation: You need at least three separate ad groups, each with a clear theme. If you’re selling running shoes, for example, your ad groups could be “Trail Running Shoes,” “Road Running Shoes,” and “Race Day Footwear.”
2. Headline & Description Upload: In each ad group, give it a good mix of creative, at least 15 different headlines (30 characters) and 4 different descriptions (90 characters). The AI uses these as building blocks. I’ve seen much better performance when these short pieces include strong calls to action and what makes the product unique.
3. Audience Signals: This is critical. Under “Audience Signals,” connect your best audience lists: first-party data like customer lists, Google’s in-market segments, and any custom intent audiences you’ve built. The better your signals, the faster the AI hones in on the right people. For example, if you’re going after small business owners, plugging in a custom intent audience of people searching for “CRM software for startups” gives it a powerful starting point.

Expected Outcome: Don’t panic if things look a little weird at first. In the first 72 hours, Performance Pilot will start spinning up and testing tons of ad combinations. You’ll see impressions and clicks, but don’t touch anything yet. The AI needs time to find what works, and I’ve noticed the CPA is often a bit high for a few days as it explores before it starts to dial in its optimizations.

Monitoring and Refining Agentic AI Performance

Once the campaign is running, your job isn’t about pulling levers anymore. It’s about strategic oversight and refinement. This constant feedback is what makes or breaks the AI’s effectiveness.

1. Daily Review of the Anomaly Detection Dashboard

Go to “Agentic Tools” > “Performance Pilot” > “Anomaly Detection”. This dashboard is where you’ll see how the AI is doing and spot anything that’s going off the rails. It uses machine learning to figure out what’s normal for your campaign and then flags anything that looks statistically odd.

1. Conversion Rate Fluctuations: Watch for big dips or jumps in conversion rates, especially if traffic hasn’t changed. The dashboard flags this. A sudden drop can mean your landing page is broken or the AI has started targeting a completely different audience.
2. Budget Pacing Alerts: The AI tries to pace the budget, but a sudden market shift can make it spend way too fast or not at all. Anomaly Detection will warn you. If it’s always underspending, your CPA target might be too tight for it to find anyone.
3. Spend vs. CPA Trend: Look at the daily spend versus the actual CPA it’s achieving. If you’re spending a lot but the CPA is way off target, you need to dig into which ad variations and audiences the AI is throwing money at. The dashboard lets you drill down and see this.

My Opinion: Trusting the AI completely without a human checking in is a fast way to burn through your budget. Yes, the system is agentic, but it’s not perfect. The AI only knows the data you give it, and if that data’s bad, or if a competitor suddenly launches a huge sale, a human needs to step in and point the AI in the right direction.

2. Providing Feedback via the Strategy Adjustment Module

Performance Pilot has a “Strategy Adjustment” module, which you can get to from the Anomaly Detection dashboard. This is how you give the AI direct feedback. It’s for course-correcting its strategy, not for micromanaging individual bids.

1. Adjusting CPA/ROAS Targets: If the AI is consistently beating your CPA target, try tightening the leash a bit (e.g., from $50 down to $48) to see if you can get more efficient. On the flip side, if it’s struggling, you may need to loosen the target slightly to give it more room to learn.
2. Pausing Underperforming Creative Elements: The AI will eventually stop using bad creative, but you can speed it up. In the “Creative & Signals” section, look at the metrics for each headline. If one is getting tons of impressions but zero conversions, just pause it. This sends a very clear signal to the AI about what isn’t working.
3. Excluding Irrelevant Audiences: If you see from the Anomaly Detection dashboard that a certain audience segment is eating up budget with junk leads, go into “Audience Signals” and exclude it. The AI learns from this explicit instruction.

Here’s what nobody tells you: The AI learns best from clear, decisive feedback. Constantly making tiny, hesitant changes just confuses it. You should only make an adjustment when you have clear data showing that something is really underperforming or overperforming and needs your guidance. For example, if a certain keyword is just bringing in garbage traffic no matter what, you might change its match type or pause it. The AI will adapt.

3. Analyzing Performance Reports in the Optimization Insights Hub

The “Optimization Insights Hub” in Google Ads 360 (under “Reports”) is where you get deeper analytics to see what the AI is really doing. This is where you go to prove its value and find new opportunities. A 2025 IAB report on advanced automation actually found that businesses using these platforms well saw ROAS go up by an average of 15% in the first year (IAB.com/insights/agentic-ai-impact-report-2025).

1. Attribution Modeling: Use the “Model Comparison Tool” in here to see how the agentic campaign is influencing conversions across the whole customer journey. These campaigns are often great at getting that first touchpoint that leads to a conversion later.
2. Keyword Performance Deep Dive: You still have to check your keywords, even with agentic bidding. The AI might uncover some great new high-performing keywords you can use in other campaigns. It also might be wasting money on some, which you should then add to your negative keyword list.
3. Segmented Performance Analysis: Slice the performance data by device, location, and time of day. This will show you if the AI is over-spending on certain segments that don’t deliver quality, even if the overall CPA looks okay. You might discover mobile conversions are cheap but have a low LTV which means you need to adjust your goals.

Working with agentic AI in performance marketing is a constant cycle of setting it up, watching it, and tuning it. It requires you to shift from controlling every little detail to providing strategic guidance, letting the AI sweat the small stuff while you focus on the big picture. Getting this right gives you efficiency gains along with a level of campaign sophistication that was impossible before.