Continuous Prebid optimization without revenue share
BidTune watches your live auction, finds safe revenue tests, and deploys the winners inside your guardrails. No stack migration. No bidder lock-in. No percentage of revenue.
Header floor optimization
Revenue lift observed against holdback traffic. Ready for promotion under current caps.
JSONLint beta proof from Prebid traffic
The autonomous researcher generated 23 optimization hypotheses, tested them as Bayesian experiments, and only promoted changes with evidence.
Autonomous research, constrained execution
BidTune is built around a tight loop: observe the auction, propose a test, measure against holdback traffic, then ship only when the evidence clears your rules.
Observe
Bid-level telemetry starts without changing your auction.
Research
The AI researcher proposes experiments from live patterns.
Experiment
Bayesian tests split traffic and stop when evidence is clear.
Deploy
Winners ship inside your caps. Losers never reach full traffic.
You choose which changes make it to production
Autonomous does not mean uncontrolled. BidTune keeps each recommendation tied to scope, measurement state, and rollback evidence.
Scope
Site, geo, ad unit, and bidder limits are applied before launch.
Decision
Observed, estimated, and projected metrics stay labeled.
Rollback
Every change keeps an audit trail and a fast recovery path.
Ready to promote
GumGum timeout extension
Manual approval mode keeps the final decision with your team.$0.03 CPM. No revenue share.
Most managed optimization asks for a percentage of revenue. BidTune charges a fixed CPM because the research loop is automated.
Built for publishers who already run Prebid
Do I have to migrate my stack?
No. BidTune works alongside your existing Prebid.js setup and tests changes against your current demand partners.
What does it cost?
The beta price is a flat $0.03 CPM. No revenue share, no annual commitment, and no percentage of upside.
Can I review changes before they go live?
Yes. Manual approval mode is available, and every experiment keeps a visible diff, result, and rollback path.