A deterministic engine that scores every campaign, allocates budget against Hill response curves, executes nightly, and explains every decision. No black boxes. No vague AI magic.
Link your Google, Meta, and Microsoft Ads accounts with a single OAuth flow. ROASt pulls campaigns, budgets, bidding strategies, conversion actions, and impression share — automatically. No CSVs, no API keys, no manual setup.
ROASt scans your conversion tracking and account structure automatically. Auto-fixes what it can, flags what it can't. Most accounts have issues they don't know about — and those issues silently inflate your ROAS.
The Strategic Mix Allocator fits Hill response curves per platform — given a total budget, where should each pound go? These same curves feed the live nightly allocator, blending marginal-return scores into campaign scoring so budget flows toward genuine headroom.
Organise your campaigns into portfolios — by client, brand, funnel stage, or however you work. Set a total budget, allocation constraints, and an optimisation goal. Mix platforms freely in the same portfolio.
The engine scores every campaign, allocates budget within each portfolio, and executes nightly or intraday. Platform adapters enforce safety constraints at every step.
After attribution lag clears, the calibration system compares predictions to actuals. EWMA bias tracking derives nine parameters that feed back into the next run. Shrinkage protects against weak evidence; 14-day staleness decay reverts dormant portfolios to safe defaults.
One OAuth flow per platform. ROASt pulls campaigns, budgets, bidding strategies, conversion actions, and impression share — automatically.
Scans conversion tracking for double-counts, stale tags, zero-value events, and attribution conflicts. Auto-fixes what it can.
Hill response curves map spend to revenue at every level. The same curves feed the live nightly allocator, steering budget toward genuine headroom.
Set a total budget, constraints, and a goal. Each portfolio is its own optimisation unit with pacing, reserve, and safety constraints.
The engine scores every campaign, allocates budget, and executes nightly or intraday. Platform adapters enforce safety constraints.
Calibration compares predictions to actuals. EWMA bias tracking adjusts nine parameters, with shrinkage and staleness decay for reliability. The engine improves with every cycle.
Updated curves feed the next model. Better calibration produces better allocations. The system improves autonomously.
Flume is your ROASt Labs assistant. Ask about performance, compare campaigns, generate reports, or manage portfolios — all in natural language. It uses the engine's real data, not guesswork. Every action requires your confirmation.
ROASt fits a Hill response curve to each platform from your own data, then re-allocates budget across Google, Meta and Microsoft to maximise total revenue. You see the uplift before you apply it — and the math behind every number.
The engine scores every campaign, proposes per-campaign budgets, and pushes them to the platform — nightly or intraday. Every change is staged for review, executed atomically, and logged with a full reasoning trail.
Upload your landing pages, brand briefs, dated promos, and client notes into the Context Hub. When an anomaly lands, click Explain this. ROASt cross-references the drop against everything it knows — events, platform changes, creative refreshes, URL changes — and writes a one-paragraph narrative you can paste into a client reply.
Upload a brand brief, dated promo, or competitor note. When something shifts, the narrative points at the exact document and date — no guessing which change caused the dip.
Every Monday at 07:00 in your client’s timezone, a one-page brief lands in your inbox. KPIs, what changed, what to act on, context drift — ready to forward to the client.
Connect Claude Desktop, Cowork, or your own Claude Agent SDK integration to ROASt’s 19 read-only MCP tools. Have an agent run your Monday check across every client before 9am.
The engine is platform-agnostic. Platform adapters are the final safety layer — each one knows exactly what that platform can and can't handle.
No black boxes. No vague AI magic. A deterministic engine that scores, allocates, executes, and learns — across Google, Meta, and Microsoft from one place.