The setup
Every account in this study runs on AlgoBoost auto-pilot, which means the model controls bids, budgets, and structure inside the guardrails each advertiser sets. We split the cohort by the bidding objective they handed the model: a flat target CPA, or value-based bidding that predicts the revenue behind each click and bids to it.
We then tracked blended ROAS across a 90-day window, normalizing for spend so a handful of large accounts could not skew the median. No account changed its objective mid-window.
What we found
Accounts on value-based bidding beat target-CPA accounts by a median of 41% ROAS. The gap was widest where revenue per conversion varied a lot — a $40 order and a $400 order look identical to a CPA target, but value-based bidding learns to pay more for the clicks that tend to become the $400 order.
The accounts that gave the model full budget control — rather than capping every campaign individually — saw the largest lift, because the model could move money toward the segments that were actually converting instead of defending a budget line that no longer made sense.
Why it works
A target CPA optimizes for the cost of a conversion. Value-based bidding optimizes for the return on a conversion. Those are not the same goal, and on accounts with any spread in order value the difference compounds every hour the model runs.
The catch is data quality: value-based bidding is only as good as the revenue signal you feed it. Accounts that piped real order value back from their store or CRM pulled ahead of accounts that relied on a static conversion value.
The takeaway
If your conversions are worth wildly different amounts, bidding to a single CPA leaves money on the table. Connect a real revenue signal, hand the model budget control, and let it bid to value. On this cohort that was worth a median 41% lift in 90 days.