A job AI is built for
That loop is autoresearch. Watch the bid landscape. Form one question. Spend a slice of traffic on a controlled test. Write down what happened. Come back when demand moves again.
This is the kind of work AI can do 24/7. The people and partners already on the account can be excellent at it. BidTune does not grade their work. It applies the same discipline continuously, with the arithmetic written down.
Small tests, on your setup
BidTune works with the Prebid, bidders, ad server, and partners you already use. The install is one script URL. The rest of the stack stays.
A permanent 10% holdback keeps your current configuration untouched. Tests run on 30% of optimized traffic. Each test needs at least 72 hours and 2,000 sessions in each arm. A change is kept only when it is measurably better. Anything else rolls back.
Every change, with its measured lift
The experiment ledger is the record: every change BidTune made, the traffic it used, and the lift that measurement supports. A rollback is recorded too. Individual tests are not added together and called a total. The holdback is the comparison for the whole setup.
We publish a revenue number when the ledger supports it. A point estimate without a window, a sample, and a comparison is not a result. Read how we measure.
Alongside the people already in place
If a team or a vendor already manages your Prebid, they stay. BidTune runs beside that setup. Their relationships and fees remain theirs. You can pause the work or undo a change.
Our experience building Optimize at BuySellAds is why this measurement matters to us. Software can keep the experiments going. The publisher still owns the stack.
Proof, when we have it
JSONLint is our first live site. The public case study is a set of diagnostics from the auction data. We will publish an overall revenue result when the holdback comparison supports it.