Continuous Prebid optimization
without revenue share
BidTune runs controlled experiments on your existing Prebid.js setup and deploys winners inside your guardrails. No stack migration. No bidder lock-in. No percentage of revenue.
An autonomous AI researcher analyzes your bid stream, generates optimization hypotheses, and tests them using Bayesian A/B experiments against live traffic. Winners deploy automatically. Losers get killed. The system never stops iterating.
The cost is a flat $0.03 CPM. No revenue share. No minimums. No annual commitment.
Built by the team behind
BuySellAds. Powering ads on developer sites since 2008.
During our beta on jsonlint.com (a developer-tools site running 8 demand partners on Prebid.js), BidTune's autonomous researcher generated 23 optimization hypotheses and tested all of them as controlled Bayesian experiments. Eight produced statistically significant lifts. Ten losers were caught and terminated before causing sustained damage. Five were inconclusive and queued for retesting.
| Experiment | Result | Lift |
|---|
| Sonobi/OpenX exclusion in tier-1 geos | WINNER | +31.5% |
| Bid caching for repeat impressions | WINNER | +17.3% |
| Header ad unit floor optimization | WINNER | +13.0% |
| GumGum per-bidder timeout extension | WINNER | +12.1% |
| Footer premium floor pricing | WINNER | +12.0% |
| GumGum restrict to tier-1 countries | WINNER | +11.9% |
| Sidebar floor optimization | LOSER | -41.7% |
| India floor price raise to $0.10 | LOSER | -27.9% |
| Global timeout reduction to 2000ms | LOSER | -19.1% |
The losers matter as much as the winners. One experiment—raising sidebar floor prices—would have decreased revenue by an estimated 42% if deployed without testing. The Bayesian engine caught it and killed it. Another—reducing the global timeout to 2000ms—would have cost 19%.
Full case study with experiment-level data →
How it works
BidTune consists of three components:
Observe
The Observer
A lightweight JavaScript snippet (<2KB gzipped) that rides alongside your existing Prebid.js installation. It reports bid-level telemetry to BidTune's analytics pipeline without modifying the auction or injecting demand.
Research
The Researcher
An autonomous AI agent that consumes your bid data and generates optimization hypotheses. It operates continuously, noticing patterns across geos, devices, ad units, and bidder behavior that humans would miss.
Experiment
The Experiment Engine
A Bayesian A/B testing system that runs at the edge. Traffic is split, the variant is tested against live traffic, and the system auto-terminates when confidence thresholds are reached. Winners deploy automatically. Losers are killed.
Safety
Autonomous doesn't mean uncontrolled. BidTune runs changes inside your guardrails with observation-first defaults, traffic caps, and full auditability. Built around three core safety principles:
- Observation first. BidTune starts in observation-only mode. No changes to your config until the research engine has a baseline and a hypothesis worth testing.
- Constrained by default. Configurable traffic caps, per-geo and per-ad-unit constraints, approved-bidder lists, and minimum confidence thresholds before any change deploys.
- Reversible and auditable. One-click rollback on any change. Full change log with config diffs. Manual approval mode available. Every decision is visible and exportable.
BidTune charges $0.03 per thousand impressions. No revenue share. No minimums. No annual commitment. Getting started is free during beta—you can probably tell from this website that the product speaks for itself.
| Managed services (rev share) | 15–25% of revenue |
| SaaS tools (flat CPM) | $0.05–0.10 CPM |
| BidTune | $0.03 CPM |
Use the calculator for your numbers.
We can offer this pricing because the autonomous researcher replaces the most expensive part of the managed-service cost structure: the human ad ops team. That labor is what the 20% revenue share pays for. BidTune's AI researcher does the same work—faster, more thoroughly, around the clock—without the headcount.
Features
Everything is included at $0.03 CPM. No tiers. No feature gates. No upsells.
- Bayesian A/B testing engine with auto-termination
- Automated floor price optimization
- Per-bidder timeout and latency tuning
- Bidder win-rate analysis and exclusion testing
- AI-powered continuous hypothesis generation
- Real-time analytics dashboard with full bid-level data
- Smart alerts and health checks
- Bid density and auction analysis
- Config diff log and one-click rollback
- Unlimited experiments, unlimited sites
- Month-to-month. Cancel anytime. No lock-in.
Get started free — connect your site in 5 minutes