Adferry is a deterministic, ML-assisted ad decisioning stack — VAST + OpenRTB + Prebid in one unified waterfall, with Auto-Pilot dynamic floors and the Sentinel yield engine making sub-millisecond calls on every request.
From the ad call to a rendered impression — a single pipeline, fully observable at each stage.
Built for sub-100ms decisions at global scale.
VAST wrappers, OpenRTB bidders, Prebid Server adapters, and PMP/PG deals compete in priority order inside a single rule — not forked across tools.
Drag-and-drop demand ordering, priority + weight tuning, and per-partner floor overrides. Every change previews before activation.
Dynamic floor CPM resolution with percentage and absolute variance modes, plus geo and device-level overrides. Floors update live — no republish.
Last Known Good Config snapshots the last fill-producing waterfall. On no-fill incidents, traffic automatically reverts — no on-call page.
First-class Prebid.js and Prebid Server support. Client-side, server-side, or hybrid per partner. Unified timeouts and transparent per-bidder logs — win, no-bid, timeout.
PMP, Preferred, and Programmatic Guaranteed as first-class deal entries in the same waterfall — per-deal floors, buyer seats, always-on vs. auction-override.
Control how often the same user sees an ad — across devices, hourly or lifetime windows, without slowing the decision path.
Production ML that predicts fill rate for the next 15-minute window and adjusts the floor in real time — with guardrails, audit trails, and staged rollout.
Sentinel learns from your fill history, bid behavior, demand mix, and seasonality — then adjusts floors for the window ahead. Guardrailed, bounded, and auditable per decision.
New model versions run in shadow first, then promote through explicit statistical gates — never flipped at once, never without evidence.
Real-time deviation tracking against historical baselines. Alert thresholds configurable per tag; auto-mute and suppression for known incidents.
Action cards capture the feature vector, model version, prediction, action taken, and outcome — queryable per tag, per hour. No black box.
Requests, bids, impressions, bid-level and impression-level fill rate, revenue CPM — broken down by tag, demand partner, and time range. Role-scoped so each team sees only the inventory they own.
Every auction event retained at row level — fill rate by tag, bundle, state, and geo at arbitrary depth. Export via API or hook directly into your data warehouse.
Programmatic control for every resource your team actually touches — tags, campaigns, line items, analytics, and yield audits.
Drop-in JS SDK and raw VAST endpoint. Works with IMA-compatible players across Web, Mobile-Web, and CTV.
Full CRUD for campaigns, line items, creatives, day-parts, budgets, and tags — scoped to the resources your team owns.
Role-scoped analytics plus a Sentinel audit surface — every ML decision captured with inputs, outputs, and outcome.
A 30-minute session with a solutions engineer — waterfall logic, Sentinel dashboard, and your tag integration mapped live.