Use cases · Marketing

Campaign pipelines where brand and legal approve before anything publishes

Workflows 4 patterns
Human oversight built into every one
All departments /use-cases

The reality in marketing

Marketing runs on volume with a review bottleneck in the middle. Every asset needs brand review; anything customer-facing needs legal too. Generic AI writing tools speed up the draft, then dump it into the same Slack-thread approval chaos — no record of who changed what, no record of who signed off. Meanwhile enquiries arrive as Instagram comments and Messenger DMs at all hours, answered inconsistently or not at all, and competitor research happens in browser tabs that produce a doc with no sources.

The uncomfortable part: the more content AI produces, the more review load humans absorb. Speeding up drafting without structuring review just moves the pile.

What changes with governed agents

Drafting gets grounded and review gets a mechanism. Agents pull brand voice and product facts through a read-only search_kb tool whose scope is fixed server-side — an agent granted the brand knowledge base can search that base and nothing else. Guardrails and PII redaction run before inference on anything customer-facing, and every workflow carries a spend ceiling, so a runaway generation loop hits a budget wall, not your invoice.

Review becomes a node, not a thread. A draft parks durably at an approval gate — single approver or N-of-M consensus, with SLA timeouts and escalation chains — and the run waits days if it has to, burning nothing. Approve, and the publish step executes through connected integrations with credentials the agent never sees. Reject, and the rejection is recorded with reasons.

On customer channels, agents get a restricted tool surface: only tools explicitly marked channel-safe are available to anyone talking to an anonymous Instagram visitor. Per-conversation auto-reply spend is captured and surfaced. Everything — drafts, decisions, publishes, guardrail events — lands in an append-only audit log your security team can query.

What stays human

Brand judgment. No agent decides what your brand sounds like; agents draft against the guidelines humans wrote, and humans decide whether the draft clears the bar. Legal sign-off on claims stays with legal — the platform makes the gate durable and visible, it does not make the call. Live conversations stay one click from takeover: an operator can step into any channel thread from the shared inbox and hand back when done. And campaign strategy — what to say, to whom, when — remains yours. Agents propose; humans decide.

The workflows
01

Campaign content pipeline

Trigger A campaign brief arrives via webhook from your intake form, or the content-calendar schedule picks up the next planned piece.
Flow An agent drafts copy grounded in the brand knowledge base through read-only search_kb; guardrails and PII redaction run before inference; the draft lands as a versioned artifact.
Human gate Brand and legal sign off as an N-of-M approval with SLA timeouts and escalation; the run parks durably until decided, and a rejection sends the draft back with reasons.
Output Approved copy publishes through the connected CMS or social integrations; every version and decision lands in the audit log.
02

Instagram and Messenger enquiry capture

Trigger An enquiry lands as a DM on the connected Instagram or Messenger channel.
Flow A channel agent with a channel-safe tool surface answers from a pinned public-facing knowledge base, extracts contact and intent details, and scores fit against your qualification criteria.
Human gate Operators watch the Conversations inbox and can take over any thread mid-conversation, then hand back to the agent when done.
Output Qualified enquiries land as rows in a Data Tables lead list, ready for the sales pipeline; per-conversation spend is captured.
03

Competitor and market brief

Trigger A weekly schedule with an overlap policy, or an on-demand run before a launch or planning cycle.
Flow A sandboxed agent researches sources inside a network egress allowlist — every outbound domain captured on the run page — and drafts a brief with citations.
Human gate The marketing lead reviews and approves the brief before it circulates; nothing distributes unreviewed.
Output A versioned, cited artifact with a stable URL, plus the exact list of domains the research touched.
04

Content repurposing run

Trigger An approved long-form piece is marked ready — a webhook fires, or a scheduled run picks up rows marked ready in the Data Tables tracker.
Flow Fork-join branches draft channel variants (email, social, newsletter) from the approved source; each variant is checked against brand guidelines retrieved from the knowledge base.
Human gate The channel owner approves each variant batch before anything publishes; a spend ceiling caps the whole run.
Output Approved variants publish through integrations or queue as drafts; each version is tracked as an artifact.
Platform pieces this leans on
Channel agents (Instagram/Messenger/email) Conversations inbox (takeover and handoff) N-of-M approvals with SLA timeouts Vector knowledge base + search_kb Data Tables PII guardrails + spend ceilings Egress-controlled sandboxes Versioned artifacts + audit log

Start with one workflow. The free tier is enough to run a real one — agents propose, your team decides.

Other departments