Signal Architecture: How to Feed Google's AI the Right Conversion Data
Google’s AI doesn’t know your business. It knows what you tell it. And most advertisers are telling it the wrong things.
When you optimize for “form fills” or “page views,” you’re training a multi-billion-dollar machine learning system to find you the cheapest, fastest, lowest-quality conversions possible. The algorithm is doing exactly what you asked—you’re just asking the wrong question.
The Signal Hierarchy
Think of your conversion data as a hierarchy of truth. The closer your signal is to actual revenue, the smarter your campaigns become:
| Signal Level | Example | Truth Quality |
|---|---|---|
| Level 1: Vanity | Page views, time on site | ❌ Nearly useless |
| Level 2: Engagement | Form fills, chat starts | ⚠️ Directional |
| Level 3: Qualification | MQL, discovery call booked | ✅ Getting warm |
| Level 4: Pipeline | SQL, proposal sent | ✅✅ Strong |
| Level 5: Revenue | Closed-won, contract signed | ✅✅✅ Truth |
Most advertisers operate at Level 2. They optimize for form fills and wonder why their sales team complains about lead quality. The fix isn’t better targeting or different keywords—it’s better signals.
Building the Feedback Loop
Step 1: Map Your CRM Pipeline
Before touching Google Ads, document every stage in your sales process. What happens after someone fills out a form? Who qualifies them? What percentage advance to each stage? How long does each stage take?
This pipeline map becomes your signal blueprint.
Step 2: Implement Offline Conversion Imports
Google Ads supports offline conversion tracking through the Google Click ID (GCLID). When a lead fills out your form, capture the GCLID alongside their contact info. When that lead progresses through your pipeline—say, from “New Lead” to “Qualified Opportunity”—push that event back to Google Ads with the GCLID attached.
Now the algorithm knows which clicks led to real business outcomes, not just form fills.
Step 3: Assign Conversion Values
Not all conversions are equal. A $50K enterprise deal and a $500 SMB trial are both “qualified opportunities,” but they have radically different business value. By assigning estimated values to each conversion event, you unlock value-based bidding (tROAS), which tells the algorithm: “find me more clicks that look like the ones that generate the most revenue.”
Step 4: Set Appropriate Conversion Windows
This is where most practitioners get it wrong. If your average sales cycle is 90 days, but your conversion window is set to 7 days, you’re training the algorithm on incomplete data. It literally can’t see the conversions that matter most.
Match your conversion window to your actual sales cycle. For B2B, that often means 60–90 days. For consumer, 7–30 days.
The Compound Effect
Here’s what happens when you get signal architecture right:
- Week 1–4: Algorithm begins learning on higher-quality signals. CPA may appear to increase as it deprioritizes cheap, low-quality clicks.
- Week 4–8: Smart Bidding recalibrates. You start seeing fewer total conversions but higher conversion quality.
- Week 8–16: The flywheel effect kicks in. The algorithm has enough data to find patterns invisible to human analysts. Cost per qualified lead drops. Pipeline velocity increases.
The patience required in weeks 1–4 is why most advertisers never get here. They see CPA increase and panic. They revert to optimizing for form fills. And they stay stuck.
Signal Architecture Is a Competitive Moat
Here’s the strategic insight most people miss: your signal architecture is a durable competitive advantage.
Your competitors can copy your keywords. They can copy your ad copy. They can even copy your landing pages. But they can’t copy the quality of the conversion data you feed your algorithms.
When your campaigns are trained on Level 5 signals and your competitor is still at Level 2, you’re operating a fundamentally different—and superior—system.
The algorithm isn’t the weapon. The signal is.