> For the complete documentation index, see [llms.txt](https://connectyai.gitbook.io/documentation/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://connectyai.gitbook.io/documentation/getting-started-self-serve-plan/system-overview-for-meta-ads.md).

# System Overview (For Meta Ads)

Part of the Self-serve plan. This page describes how Connecty connects to your Meta ad account and delivers daily recommendations in the app.

Connecty is an AI-native decision platform for Meta Ads, built as an agentic system, not a chatbot on an API.

At its core is the **Autonomous Semantic Graph**: a proprietary, living model of your business that a coordinated set of agents reasons over continuously.

Connecty connects to your Meta ad account and ecommerce sources, hosts and processes the data on its own infrastructure, and delivers recommendations through the Connecty app. Ingestion, modeling, graph construction, and statistical grounding all happen automatically: no warehouse, no pixel, no setup project.

### How it works

**1. Connect your accounts.** Authorize your Meta ad account via OAuth, then connect Shopify, Stripe, WooCommerce, or Google Analytics. No pixel or tracking tag is installed.

**2. Select campaigns.** Selection happens at the campaign level: every ad set and ad under a selected campaign is included. Entities can't be cherry-picked in or out.

**3. Sync and host.** Connecty ingests performance, creative, and revenue data and keeps it in sync on Connecty-hosted infrastructure. Metrics are computed with consistent definitions, so the same question always returns the same answer.

**4. Analyze.** Creative signals are extracted, interpreted in context, and joined with audience, performance, and revenue data. See Core technology below.

**5. Recommend.** The output is sized, ranked, guardrailed recommendations, each with its reasoning and evidence, surfaced in daily reports and chat.

**6. Review and act.** You decide what to apply. Every recommendation is explainable; every approved change is logged.

### Core technology

The foundation is the **Autonomous Semantic Graph**: a living graph of your business that Connecty's agents reason over.

A standard semantic layer stops at metrics: what happened. The graph adds four layers above them (Signals, Scenarios, Actions, and Goals) so agents know what to weigh, which moves exist, and what winning means for your business.

The graph builds and updates itself as data syncs. Agents run scenarios against it daily; no static rules to wire.

On that foundation, the analysis runs in three layers:

**Layer 1: Creative signal extraction.** Every surface the customer sees, read in full: video frame by frame (hook, pacing, product visibility, proof, CTA timing), image and carousel ads, and the landing page after the click. What's listed is a sample; the full signal taxonomy runs considerably deeper.

**Layer 2: Contextual interpretation.** A reasoning layer decodes the role each element plays: entity detection sees that the red scarf is visible; Connecty sees whether it's acting as the hook that improves retention. Every judgment is grounded in the account's own data, against statistically comparable ads only; how comparability is determined is part of Connecty's core methodology.

**Layer 3: Signal fusion and decision.** Interpreted creative signals are joined with the audience and funnel stage watching (TOFU, MOFU, BOFU), the full journey from engagement through conversion to retention, and real revenue and margin, with no pixel. That three-way join produces the output: a sized, ranked, guardrailed recommendation, not a chart.

### Interacting with Connecty

Everything happens in the Connecty app, in three modes:

* **Reports.** Daily reports surface what changed, why, and what to do about it. They arrive proactively; you don't have to ask.
* **Chat.** Ask questions in plain language and get deterministic answers grounded in your synced data.
* **Review.** Scan pending recommendations, inspect the reasoning, and approve or dismiss each one. Rate any output with a thumbs up or down.

### Agentic operation

Connecty runs as a coordinated set of agents over the graph, not a single chatbot. Agents track signals, evaluate scenarios, and rank candidate actions continuously as new data syncs.

This can't be replicated by pointing a general-purpose LLM at an ad account. An LLM gives you perception: an opinion from general priors. Connecty's infrastructure solves the problems underneath a trustworthy decision:

* Deterministic metric computation: the same answer, every time
* Statistical comparability: no ad is ever judged against the wrong baseline
* Revenue joined across ad and commerce sources, without a pixel
* A semantic graph that stays correct as the account evolves
* Agent actions that are sized, guardrailed, and fully traceable

Each of those is purpose-built infrastructure, not a prompt.

### Human control

Connecty recommends; you decide. No configuration exists where changes happen to your ad account without your review.

Every recommendation shows the inputs and reasoning that produced it. Every approved change is logged.

Every answer, report, and recommendation carries a thumbs up and thumbs down control. That feedback refines how Connecty analyzes and recommends for your account.

### Data ownership and security

Your data stays yours. Connecty accesses your Meta and commerce accounts through their official APIs, with the permissions you grant.

Synced data is hosted securely and used only to power your workspace. Disconnecting an account stops the sync.

No pixel or tracking script is ever placed on your site or your customers' browsers.