THE THESISThe bidding algorithm will optimize the value you send. If revenue is a poor proxy for profit or lead quality, scale will magnify the mistake.
Choose the economic target
Move from gross revenue toward contribution margin after product cost, fulfillment, payment fees, discounts and expected returns. For lead generation, assign values from observed stage conversion and expected margin—not arbitrary scores.
Build a value pipeline
Create stable transaction or lead IDs, capture consented first-party signals, reconcile outcomes in the CRM or warehouse and return qualified values to platforms with clear latency rules. Validate match rates and deduplication before changing bidding.
Protect exploration
High-value bidding can over-concentrate on familiar customers or easy products. Add new-customer value rules, inventory and margin guardrails, and explicit exploration budgets. Monitor value distribution, not only averages.
Read marginal return
Average ROAS hides saturation. Increase spend in controlled steps, compare incremental margin with the next-best use of capital and stop when marginal return falls below the commercial hurdle rate.
Practical questions
Can lead values be estimated?+
Yes, if estimates are tied to observed stage conversion and margin, refreshed regularly and checked for bias.