ML-Based Bidding Automation

At GetYourGuide, I co-developed an ML-based bidding model that replaced the existing rule-based CPC scripts for placement campaigns managing 5 million whitelisted placements. The model predicts the best possible bid per placement and updates bids automatically every week.

The challenge

With 5 million placements across multiple accounts and markets, manual bid management was impossible. The rule-based script was too rigid. It couldn’t account for the nonlinear relationship between bid price and conversion probability.

What I did

• Worked with the Growth Data Product team to design and validate the ML model
• Provided the marketing domain expertise: which features matter, the business constraints, how bidding interacts with Google’s auction dynamics
• Managed the transition from the old script to the ML model across all active accounts
• Built monitoring dashboards in Looker to track model performance vs. the old approach
• Documented the entire system in a 44-page handover the team still references

The result

Cost down 54%, ROAS up 81%. Net revenue decreased 16%, an acceptable trade-off that was explicitly modeled and agreed upon, because the efficiency gain produced seven-figure annual cost savings. This is the kind of decision that requires a marketer who understands both the business model and the math.

Why this matters for you

Automation is only as good as the judgment behind it. Most bidding tools optimize the metric you point them at, even when chasing it quietly costs you money. The value here wasn’t the model; it was knowing which trade-off was worth making and being able to say so out loud. That’s the thinking I’d bring to your account before I let any algorithm touch your budget.