← Back to Help Center

AI & Automation

How the AI engine works, autonomy levels, and configuring automated optimization rules.

How AI Optimization Works

Scalegrowth uses a heuristics-first, ML-second approach:

  • Day 1 — Heuristic rules: Simple, proven optimizations run immediately (shift budget from high-CPA to low-CPA campaigns, pause underperformers, scale winners). No data threshold required.
  • Week 2+ — ML models activate: After collecting enough data, machine learning models start predicting conversions, optimal bids, and audience segments. Models must beat heuristics in backtesting before going live.
  • Continuous learning: Models retrain weekly with your latest data. Performance is monitored against heuristic baselines — if ML degrades, the system falls back to rules.

Autonomy Levels

Control how much freedom the AI has:

0Manual

AI recommends, you approve every action. Best for new users who want to understand each optimization before applying it.

1Guided

AI executes low-risk actions (bid adjustments <10%) automatically but asks approval for high-risk actions (budget changes, pausing campaigns, creative swaps).

2Supervised

AI executes most optimizations automatically. You receive real-time notifications and can undo any action within a configurable grace period (default: 2 hours).

3Autonomous

Full automation. AI manages budgets, bids, targeting, and creatives. You receive daily summary reports. Best for experienced users with established campaigns.

Creating Automation Rules

Rules consist of three parts:

  • Trigger condition: The metric and threshold that activates the rule (e.g., "ROAS < 1.5x after $100 spend").
  • Action: What happens when triggered (e.g., "Pause campaign", "Increase budget 15%", "Send alert").
  • Scope: Which campaigns the rule applies to (all campaigns, specific platform, specific campaign).

Use rule templates to start quickly, then customize thresholds for your business.

AI Confidence & Transparency

  • Confidence score: Every recommendation includes a confidence percentage. Higher means more data supports the decision.
  • Decision log: All AI actions (both auto-executed and recommended) are logged with the reasoning, data snapshot, and outcome.
  • Undo: Every auto-executed action can be reversed. The system keeps a before/after snapshot.
  • Heuristic flag: Recommendations marked "Heuristic" use rule-based logic; "ML" means a trained model made the decision.

Related Topics

A/B Testing →Campaign Management →Analytics →