Responsible AI

Adsperty uses AI to watch, reason and recommend, under rules that keep a human firmly in charge of money.

The AI never touches the ads APIs

This is an architectural rule, not a policy promise. The language model has no code path to Meta, TikTok or Google. Everything it wants to do becomes a proposal that must pass a deterministic policy engine, your guardrails, and, for risky actions, your explicit approval, before separate non-AI code executes and verifies it.

Hard limits the AI cannot cross

Daily spending caps, maximum budget increases, CPA and ROAS floors are enforced by plain deterministic code that evaluates every action, whoever proposed it. A prompt injection or model error can at worst produce a proposal that the policy engine rejects. The Emergency Stop halts all autonomous activity in one tap, and its release requires identity re-verification.

You choose the autonomy level

Manual, Copilot (AI recommends, you approve), Shadow (AI decides on paper, executes nothing) and Autopilot (AI executes within your guardrails). Autopilot must be earned: two-factor authentication, configured guardrails and demonstrated shadow-mode judgment come first, and every autonomous action is logged, verified and reversible by pausing.

Honest memory

The AI reads a memory of what worked for each brand, but it cannot write its own "facts". Insights are promoted by deterministic statistics with visible sample sizes and confidence, and you can confirm, dispute or delete any of them.

Your data and AI providers

Prompts sent to AI providers contain campaign metrics and settings, never your ad-platform credentials, never other workspaces' data. Per-workspace AI budgets cap what the AI may spend thinking; safety systems and monitoring never depend on an AI call to function.