Governance and guardrails
The promise is simple. So is how we keep it.
Nothing is published, sent or spent in Brandvyo until a person has agreed to exactly that wording, at that revision, for that amount. These are the controls that make the promise hold.
Step 1
Draft
Written by a person, or by AI from your brand's own data.
Step 2
Compliance scan
Your rules run at submission. Blocking rules stop it; advisory ones flag it.
Step 3
Human approval
One stage or several: brand, then legal. Each stage names who can clear it.
Step 4
Scheduled
Only the approved wording, at the approved time.
Step 5
Published & measured
Results flow back to analytics, Kara and the next draft.
1 · The gate
Human approval, enforced where it can't be skipped
Enforced in the database
The approval rule lives in the database itself. A post cannot move to approved, scheduled or published except through the one approval function, so a bug in the app produces a failed action, not an unapproved post.
Approval is for exact wording
An approval is tied to one revision. Edit the post afterwards, even slightly, and the approval is gone and it returns to review. Inbox replies and ad campaigns follow the same rule.
No auto-publish lane
We deliberately do not offer a mode where an AI check can publish for you. Pre-approval exists only as a named person's explicit, expiring, revocable choice, never as a model's verdict.
2 · Approval chains
As many stages as your organisation needs
Multi-stage approval
Require a brand review and then a legal review. Each stage names exactly which roles can clear it, how many different people must, whether the author may, and what counts as late.
Conditional stages
Apply a stage only when it matters: a particular network, AI-drafted posts only, posts with links or media, posts a compliance rule flagged, or posts mentioning certain words.
Separation of duties
Stop authors approving their own work. A legal role can read everything and clear only the stages that name it. Roles are exact, with no hidden hierarchy, so an admin cannot clear a legal stage just by outranking it.
Requirements are frozen at submission
What a post needed is fixed when it is submitted. A policy you change this afternoon does not alter what a post submitted this morning required, and an edited post is re-frozen so earlier approvals have to be earned again.
Schedules can't dodge the rules
Recurring posts and advance approvals go through the same stage check, and a stage can refuse pre-approval, so putting a post on a schedule is never a way around legal review.
Agency layer, client sign-off protected
An agency administrator acts as the brand owner and only as owner. A stage reserved for the client's approver cannot be cleared by the agency that wrote the post.
3 · Compliance rules
Say what you never say, once
- Five kinds of rule: phrases you never use, wording every post must carry, claim words that need substantiating, blocked link domains, and an allow-list of link domains.
- Plain phrases, not regular expressions, matched whole-word and case-insensitively, so “sale” does not fire on “wholesale” and nobody can accidentally write a rule that blocks everything.
- Blocking or advisory. Blocking rules stop a post being submitted, in your own words. Advisory rules let it through with the finding shown to the approver, and can route it to legal.
- The author keeps their work. A post that fails the check is saved, with exactly what tripped it.
- Automation drafts are flagged, not hidden. A blocked automated draft would sit unseen, so it reaches an approver with the finding shown loudly.
- Over-length posts are stopped before approval using each platform's own limits.
4 · Approvers
Approving shouldn't need a login or a licence
Approve from an email link
A signed link for one person, one post, one revision, one stage. It is single-use, expires after seven days and is withdrawn automatically if the post is edited or resubmitted.
Opening a link decides nothing
Email scanners and link previewers open links automatically. Here, opening only shows the post. The decision happens only when a person presses the button, so a scanner cannot approve anything.
A link authenticates, it does not authorise
The same rules apply on a link as in the app: stage roles, separation of duties, revision checks and the number of distinct approvers. And a rejection requires a reason.
Reminders that never decide
Slow reviews are chased once, then escalated once to whoever looks after the brand. Escalation emails carry a working link only to people allowed to clear the stage.
5 · AI and money
The same gate for AI, replies and spend
AI never approves or publishes
AI can draft, and an automation can submit its draft for review. It can never approve or publish anything. Kara is read-only. Learned brand voice and suggested rules stay out of drafts until a human has reviewed them.
Replies need approval too
Direct-message, comment, mention and review replies wait for approval, with the reply window visible. Posts themselves are never configurable. The setting is a fixed row that says so.
Ad spend is double-checked
Approval covers budget, schedule, targeting and creative together, the approver types the budget back, a monthly cap is checked separately, and pausing never needs approval.
6 · Access and data
Every brand walled off from every other
- Tenant isolation in the database. Row-level security keeps each brand's content, accounts, knowledge and people separate. AI knowledge and Kara are scoped the same way.
- Role protections in the database. Triggers stop admins promoting themselves to owner and keep agency links system-managed.
- Single sign-on. SAML for your organisation, only after the email domain is verified by a DNS record. Auto-provisioning never grants owner or admin.
- Closed beta gating. Sign-up is invitation-only while we harden the product.
- Hardened web fetching. Anywhere the platform reads a URL for you, it refuses private, internal and metadata addresses, on every redirect.
- Integration security. API keys are stored hashed and shown once. Webhooks are signed. CSV exports are protected against formula injection.
- Abuse controls on public forms: hidden trap fields, timing checks, hashed connection addresses (no raw IP stored) and rate limits.
- Audit trail. Drafts, edits, approvals, rejections, attachments and imports are recorded with who and when.
What we haven't built yet
Compliance-minded buyers can forgive a missing feature, not a misleading one. These are on the roadmap, not in the product today:
- • Compliance records archive and per-post approval certificates
- • SOC 2 / ISO 27001 certification and a standard signed data-processing agreement
- • Multi-market localisation workflows
If your evaluation depends on one of these, talk to us before assuming a timeline.
Governance you can explain to your auditor
Early access to Brandvyo is open. Register your interest, tell us about your brands, and we will invite you as beta spaces open. Nothing goes live without your sign-off, ever.
Brandvyo is a product by A1 IQ AI.