QORA continuously analyzes traffic quality, routing behavior, source performance and conversion signals. It adjusts bids, payouts and routes per impression, every tick.
Every impression passes through the same loop: score the source, predict the conversion, balance the bid, settle the payout, then feed the outcome back into the model.
There is no static rules layer. There is no manual repricing every morning. The system decides, learns, and decides again.
loop time · 0.8ms · 2,400+ features · 14 fraud vendors
Every publisher, app, list and SDK is scored per geo × format × hour on historical CPA, fraud, and revenue contribution.
Behavioral entropy, device fingerprinting, session depth, post-click conversion windows. Composite score per request.
Each impression is routed to the campaign × creative × payout combination with the highest predicted profit, not the highest bid.
Per-impression CR estimate. Retrained hourly on live postbacks across every advertiser, format, geo and time slice.
Bid ceiling and payout floor are co-optimized per source × campaign. The spread closes toward a target margin band, not a fixed take rate.
Pre-auction screening across 14 fraud vendors plus internal signals. Click-injection, device-farm, IP-reputation, behavioral entropy.
OpenRTB 2.6, first-price clearing. Bids equalize across PMP, PG, and open lanes so no campaign is starved on edge inventory.
M1 → M2 → M3 → M4 → M5 → M6 → M7, every tick. The outcome of each decision feeds back into M1 at the next refresh.
Every reroute, bid update, score change and block is a live event. The terminal on the right is a slice of the live event stream, sampled at one tick per second.
Per-campaign CR prediction, creative rotation, frequency caps and payout-tier balancing. All decided per impression.
Each publisher gets a quality score per geo × format. Floors and demand routing follow the score, not the placement label.
Geo-time-of-day surfaces are scored as their own market. T1 / T2 / T3 supply is priced against the geo's CR curve, not a global average.
In-app, push, native, email, banner, video. Each format has its own latency profile, fraud profile and CR curve. The engine treats them as separate markets.
No campaign starves on edge inventory; no campaign over-spends a geo. Pacing is updated per tick against a target burn curve.
CPM / CPC / CPL / CPA all clear against the same expected-revenue ladder. Publishers see the payout tier that maximises their expected revenue, not the loudest bid.
Plug your demand or supply into QORA. The CPA engine is on by default. Every impression is scored, routed, bid-balanced and audited from the first tick.