AI Transparency & Literacy at Cogny

Why this page exists

Cogny is an AI-assisted marketing platform. Article 4 of the EU AI Act (Regulation (EU) 2024/1689) asks organisations that build and operate AI systems to make sure the people using them have a working understanding of what the AI does, where it can go wrong, and how to stay in control. This page is our plain-language answer to that — for our team and for our customers.

It also documents, in one place, how we meet the AI Act's transparency obligations (Article 50) and how the system is designed to stay out of the Act's high-risk and prohibited categories.

What Cogny is — and is not

What it is: a decision-support and analytics tool. Cogny connects to your marketing and business data (Google Ads, Meta, GA4, Search Console, BigQuery, CRMs, email tools) and uses AI to analyze it, surface findings, write reports, and draft recommended actions ("Growth Tickets"). It can also help draft and execute marketing operations — but only with a human approving each action.

What it is not: Cogny is not designed or intended to make automated decisions that produce legal or similarly significant effects on individuals. It must not be used for credit scoring, hiring, worker management, insurance eligibility, access to essential services, biometric identification, emotion recognition, or anything else listed in Annex III of the EU AI Act or prohibited under its Article 5. Those uses are out of scope and not permitted under our terms.

Our role under the EU AI Act

Cogny is a provider of a limited-risk AI system that we place on the market under our own name.
Cogny is a deployer of third-party general-purpose AI models — currently Anthropic's Claude and OpenAI's GPT. We do not train or fine-tune foundation models, and our model providers do not train on customer content submitted via their APIs.
Cogny is not a provider of a general-purpose AI model, so the model-tier obligations of the Act fall on our upstream providers, not on us.

Because the system is used for marketing analytics and human-supervised marketing operations, it falls outside the Act's high-risk categories (Annex III) and prohibited practices (Article 5). The obligations that do apply to us are the transparency duties in Article 50, which we meet as below.

How we keep you informed (transparency)

AI interaction is disclosed. The chat assistant identifies itself as AI, and every chat surface carries a reminder that responses are AI-generated and may contain mistakes.
AI-generated output is labelled. Reports (in-app and by email) and AI-generated images and video in the Ad Studio are marked as AI-generated.
Recommendations are reviewable, not automatic. Growth Tickets and any action that writes back to a connected platform require explicit human approval before they run.
Actions are logged. Tool calls, approvals, denials, and execution transcripts are recorded so you can reconstruct what the AI did and when.

How we protect data

EU data residency: primary storage in Stockholm (Supabase, AWS eu-north-1), compute in the EU (GKE europe-west1).
Cogny Shield can mask personal data (emails, phone numbers, names, IPs and more) using an EU-hosted model before any frontier model sees it. It is fail-closed: if masking fails, the raw data is withheld.
A GDPR Article 28 Data Processing Agreement is available, with a signed audit trail and a current subprocessor list.

What every user should know (the literacy basics)

AI can be wrong. Treat findings and recommendations as a well-informed draft, not ground truth. Check the underlying numbers before you act — especially for anything with budget or customer impact.
You are in the loop on purpose. Approvals exist so a human decides. Don't approve actions you don't understand; ask the assistant to explain or show its query first.
Keep personal and special-category data in its lane. Don't use audience or targeting features to target people based on protected characteristics (health, ethnicity, religion, sexual orientation, political views, etc.) or to target minors. This is both a legal line (GDPR Art. 9, DSA Art. 26) and a Cogny policy line.
Don't repurpose Cogny for high-stakes decisions about people. It is a marketing tool. Using its output to decide credit, employment, eligibility, or similar is prohibited.
Mistakes and surprises are reportable. If the AI produces something biased, wrong, or unexpected, tell us at support@cogny.com — that feedback is part of how we keep the system safe.

Questions

For data-protection, AI Act, or DPA questions, contact support@cogny.com. This page is informational and is not legal advice; for your own compliance obligations as a deployer of marketing actions, consult your counsel.