Google Ads Optimization Engine

4-Tier Conversion Value Framework

ConversionOS feeds ML-derived conversion values to Google Ads, enabling Smart Bidding to optimize for revenue rather than raw conversion count. This is the system that powered 34K qualified leads from 2.3M prospects.

The 4-Tier Framework

Instead of sending a binary conversion signal (converted = yes/no), ConversionOS assigns a dollar value based on propensity tier:

Tier	Propensity Score	Conversion Value	What It Tells Google
Tier 1	≥ 0.70	$240	"This is a premium prospect — bid aggressively"
Tier 2	0.40 – 0.69	$50	"Moderate quality — bid competitively"
Tier 3	0.15 – 0.39	$10	"Low probability — bid conservatively"
Tier 4	< 0.15	$1	"Near-zero value — minimal spend"

Why This Works

Google's Smart Bidding (tROAS, value-based bidding) optimizes toward the conversion values you provide. By feeding ML-derived values instead of flat conversions:

• Google learns which audience signals correlate with high-value prospects
• Budget naturally shifts toward prospects your model predicts will convert
• ROAS improves because spend concentrates on quality, not quantity

Implementation

Step 1: Enhanced Conversions Setup

# Upload offline conversion values to Google Ads
from google.ads.googleads.client import GoogleAdsClient

def upload_conversion_values(client, customer_id, conversions):
    """Upload propensity-scored conversion values."""
    conversion_upload_service = client.get_service(
        "ConversionUploadService"
    )

    click_conversions = []
    for conv in conversions:
        click_conversion = client.get_type("ClickConversion")
        click_conversion.gclid = conv["gclid"]
        click_conversion.conversion_action = (
            f"customers/{customer_id}/conversionActions/{conv['action_id']}"
        )
        click_conversion.conversion_value = conv["propensity_value"]
        click_conversion.currency_code = "USD"
        click_conversions.append(click_conversion)

    request = client.get_type("UploadClickConversionsRequest")
    request.customer_id = customer_id
    request.conversions = click_conversions

    response = conversion_upload_service.upload_click_conversions(
        request=request
    )
    return response

Step 2: BigQuery Scheduled Export

-- Daily export: scored prospects with GCLIDs for offline upload
CREATE OR REPLACE TABLE `project.conversionos.gads_upload` AS
SELECT
  s.user_id,
  c.gclid,
  s.propensity_tier,
  s.conversion_value,
  CURRENT_TIMESTAMP() AS upload_timestamp
FROM `project.conversionos.scored_prospects` s
JOIN `project.analytics.click_ids` c USING(user_id)
WHERE s.conversion_value > 1  -- Skip $1 tier (suppression)
  AND c.gclid IS NOT NULL
  AND c.click_date >= DATE_SUB(CURRENT_DATE(), INTERVAL 90 DAY);

Step 3: Campaign Configuration

Campaign Type	Bidding Strategy	Value Signal
Performance Max	Target ROAS (tROAS)	4-tier conversion values
Search	Maximize Conversion Value	4-tier + keyword intent signals
Demand Gen	Target ROAS	4-tier + audience signals

Performance Benchmarks

From production deployment:

• ROAS improvement: 2.8x vs flat conversion bidding
• CPL reduction: 40% lower cost per qualified lead
• Budget efficiency: 70% of spend concentrates on Tier 1-2 prospects
• Lead quality: Tier 1 converts at 4.7x the rate of Tier 3