Google Showed $45 CPL. The Real Cost Per Customer Was $1,500. Smart Bidding Was Making It Worse.

The situation

A US B2B SaaS company was spending $35K/month on Google Ads. Their dashboard showed $45 CPL per free trial sign-up — well within their $60 target. Their sales team was converting 2–3% of those trials into paying customers. The real cost to acquire a customer was over $1,500, and rising every month.

The problem

The tracking was working — every trial sign-up was counted as a conversion. But Smart Bidding cannot tell the difference between a student, a competitor, and a real buyer. By optimizing for form fills, Google's algorithm bid highest on keywords attracting the easiest form completers: job seekers, researchers, and small businesses who'd never pay enterprise prices. Genuine high-value leads came from a completely different keyword set that Smart Bidding was underfunding. Only 2.3% of all leads were qualified by the sales team as having real purchase potential.

What we did

1. Analyzed 6 months of CRM data — mapped every lead from Google click to MQL to SQL to closed customer, establishing the true quality distribution
2. Connected HubSpot CRM to Google Ads via Offline Conversion Import for Leads — when a sales rep marks a contact SQL, the GCLID is uploaded back to Google
3. Removed form fill as a primary conversion — demoted to observation only so Smart Bidding stopped optimizing for it
4. Set SQL (Sales Qualified Lead) as the sole primary conversion action — algorithm now learns what a real buyer looks like
5. Added Enhanced Conversions for Leads — hashed email from form submissions sent to Google to improve cross-device attribution

Why it worked

Smart Bidding optimizes for whatever you define as a conversion. If you tell Google a form fill is a conversion, it finds humans most likely to fill forms — not humans most likely to pay. By importing only sales-qualified leads, the algorithm learned what a real buyer looks like from first principles. The $45 CPL was never real; it was the cost of teaching the algorithm the wrong lesson.