Skip to main content

Trust & compliance

Nothing runs without guardrails

Automation that touches live budget and customer data has to be defensible — to your CFO, your legal team, and your board. This page is the whole answer: what we commit to in writing, how the architecture enforces it, and the privacy and AI rules your marketing data already lives under.

Read-only by defaultNever trains on your dataYou own everythingWe sign your DPA

Four things we put in writing.

Not aspirations — these are commitments we make contractually, and every one of them is a question your security team is going to ask anyway.

Read-only by default

We start with read-only credentials on your ad and analytics accounts. Write access is requested per workflow, only where an automation genuinely needs it, scoped to that job — and revocable by you at any time, without breaking what we have already built.

Your data never trains models

Your campaign data, CRM records and strategy documents run your workflows and nothing else. We do not use client data to train or fine-tune models, and we never pool one client’s data into another client’s environment.

You own everything, and it exports

Workflows, guardrail configs, prompt packs, agent instructions and audit logs are yours. They live in your environment, they export in full, and if the engagement ends the system keeps running without us. No seat licence to cancel, no dashboard to lose.

We sign your paperwork

We will sign your DPA, MSA and security addenda, and complete vendor security questionnaires — SIG, CAIQ, or whatever format your procurement team requires. Send it over and we fill it in.

A human tower of castellers, each level standing on the shoulders of the one below

Nobody climbs a tower they don’t trust.

Guardrails are the product, not a setting.

Trust is not a promise made in a sales meeting — it is a property of how the system is built. Every automation we stand up runs inside limits your team defines before it is switched on, and writes down what it did afterwards.

  • Spend caps — a hard ceiling on what any automation can move, per day and per account.
  • Change ceilings — limits on how far a single setting can shift in one step, so nothing swings wildly overnight.
  • Approval thresholds — routine changes execute and get logged; large or unusual ones stop and wait for a human.
  • Audit trail — every action recorded as Trigger, Action, Impact: what fired it, what changed, what moved as a result.
  • Reversibility — changes are designed to be undone cleanly, not excavated out of a change log.
  • Human judgment where it matters — creative, cultural and strategic calls stay with people, by design.

The mechanics are documented publicly rather than held back as proprietary mystery: bounded autonomy, how guardrails work, the audit trail, approval thresholds, and why every change must be reversible.

The red tape your marketing data already lives under.

Most of this applies to your marketing whether or not you automate it — automation just makes it visible, and makes non-compliance scale faster. Here is the landscape we design around, and what each one actually asks of a marketing team.

Orientation, not legal advice — your counsel owns the final call on what applies to you.

Privacy & data protection

GDPR / UK GDPR
If you touch EU or UK personal data: a lawful basis for processing, honouring access and deletion requests, data processing agreements with every processor, and a valid mechanism for international transfers. Article 22 also constrains decisions made about people by automated means alone.
PIPEDA (Canada, federal)
Meaningful consent for collection and use, access on request, and safeguards proportionate to the sensitivity of the data. The baseline for most Canadian commercial activity.
Quebec Law 25
Stricter than PIPEDA: explicit consent, privacy impact assessments, transparency when decisions are made by automated processing, a designated privacy officer, and rules on moving personal information outside Quebec. Penalties reach $25M CAD or 4% of worldwide turnover.
CCPA / CPRA and the US state patchwork
Opt-out of sale and sharing — which captures a great deal of ordinary ad-tech data flow — tighter handling of sensitive personal information, and a growing list of states running their own variations.

Marketing communications

CASL (Canada)
Among the strictest anti-spam regimes anywhere: express consent before commercial electronic messages, clear sender identification, and a working unsubscribe in every message. Consent has to be requested honestly and proved later.
CAN-SPAM and TCPA (US)
Accurate headers and a functioning opt-out for email; separate and stricter consent rules for calls and SMS, where automated dialling and texting carry real per-message exposure.
PECR (UK)
Sits alongside UK GDPR and governs electronic marketing and cookies specifically — consent for tracking, and rules for unsolicited messages.

AI-specific rules

EU AI Act
The Article 50 transparency obligations — telling people when they are dealing with AI, and marking AI-generated or manipulated content — apply from 2 August 2026, which is precisely where AI-written marketing creative lands. Obligations for Annex III high-risk systems were set for the same date, but the European Parliament has voted to push them toward December 2027 and that change still needs Council agreement, so treat this timeline as live rather than settled.
Automated decision-making transparency
Quebec Law 25 and GDPR Article 22 both give people rights when consequential decisions about them are made automatically. If a system decides who sees what, who gets contacted, or who gets an offer, that is in scope.

Platform & contractual rules

Ad platform terms and API policies
Google Ads and Meta both restrict what customer data may be uploaded, in what form, and how it may be matched — hashing requirements, prohibited data categories, and API conditions that a badly built automation can breach quietly and at scale.
Your own customer contracts
Often the tightest constraint of all: the MSAs and DPAs you have already signed with your own clients, which frequently limit sub-processors, data location and retention more strictly than the law does.

The frameworks your buyers audit against

SOC 2
A controls attestation across security, availability, processing integrity, confidentiality and privacy. The default ask in North American enterprise procurement.
ISO/IEC 27001
The international standard for an information security management system — certification that security is run as a system rather than a set of habits.
ISO/IEC 42001
The AI management-system standard, and the one moving fastest. It is the only framework here that produces an audited certificate specifically for how an organisation governs AI, and it is starting to appear as a hard vendor requirement rather than a nice-to-have.
NIST AI Risk Management Framework
Voluntary and uncertifiable, but the common vocabulary — govern, map, measure, manage — that most AI risk conversations and vendor questionnaires are built on.

Primary sources: EU AI Act (European Commission) ↗ · PIPEDA (OPC) ↗ · CASL ↗ · CCPA (California AG) ↗ · NIST AI RMF ↗

We will not claim a badge we have not earned.

We are actively working toward SOC 2 and ISO certification. We are not certified today, and you will not find a logo on this page pretending otherwise — which is exactly the standard we would want you to hold any vendor to, including us.

In the meantime we build to the same control areas those audits examine: least-privilege access, change management, logging and monitoring, incident response, and vendor management. We will walk your security team through how each one works today, in detail, and complete your questionnaire in your format.

Ask for our current status →
What procurement gets
Signed DPA and security addenda
Completed SIG / CAIQ questionnaire
Access control and review process

The underlined items are published at /trust/data-practices — read them before you ask.

The questions your security team will ask.

Do you need access to our ad accounts?

We start read-only. The Automation Audit runs entirely on read-only access to your ad and analytics accounts — we can look, we cannot change anything. Write access is requested per workflow, only when a specific automation needs it, scoped to that job, and revocable by you at any time.

Is our data used to train AI models?

No. Your campaign data, CRM records, strategy documents and business context are used to run your workflows and nothing else. We do not use client data to train or fine-tune models, and we never pool one client’s data into another client’s system.

Are you SOC 2 or ISO certified?

Not yet, and we will not claim otherwise. We are actively working toward SOC 2 and ISO certification, and in the meantime we design to the same control areas those audits cover — access control, change management, logging and monitoring, incident response, and vendor management. We are happy to walk your security team through exactly how each control works today and to complete your vendor security questionnaire.

Who owns what you build?

You do. Every workflow, guardrail configuration, prompt pack, agent instruction and audit log we produce is your property, documented in your environment and exportable in full. If the engagement ends, the system keeps running without us — there is no seat licence to cancel and no dashboard you lose access to.

Will you sign our DPA and complete our security review?

Yes. We will sign your Data Processing Agreement, master services agreement and security addenda, and complete standard vendor security questionnaires such as SIG and CAIQ. If your procurement team has a required format, send it over and we will fill it in.

Built by partners you can check.

Start with the free Readiness Score to see which parts of your marketing can safely run autonomously — then the Audit maps it against your real accounts, read-only, nothing changed.

Get your free Readiness Score → Start with the Audit →