Automation best practices

Automations turn recurring desk work into reliable flows: qualify leads, escalate handoffs, and protect pipeline stages. This guide is about judgment: when to use a template, how to test, when to stay supervised, and what proactive sends require.

Start from a template when you can

• Common cases (qualify a lead, escalate a high-priority handoff, require approval before Won): start from a gallery template.
• Client-specific tags or fields: customize a template. Do not rebuild most of the same graph from scratch.
• Experiments without a clear name: keep a blank draft, but never publish without a test run.

Starting templates include qualify every new lead, escalate high-priority handoffs, and review before Won. If a template already covers the case, prefer it to a parallel custom maze.

Safe lifecycle: draft, test path, simulate, then publish

1. 1
   Save a draft

   Nothing runs in production while the automation is still a draft.

2. 2
   Test path

   Run the graph on the same engine as live, with actions shadowed so you can inspect the path without real side effects.

3. 3
   Simulate conversations

   Generate visitor scripts in your brand voice and run multi-turn rehearsals. Use this to catch gaps in knowledge and tone, not only graph wiring.

4. 4
   Publish

   Publishing freezes an immutable snapshot. Meaningful edits go back to draft and test before they run live again.

Supervised first, autonomy later

• Add a human approval step on high-impact actions (terminal Won, sensitive outbound, aggressive priority changes).
• When the case is ambiguous, start by notifying the team instead of acting alone.
• After clean runs with no surprises, reduce friction on purpose.
• Keep Disable as the operational stop button. One click stops new runs without deleting the design.

Proactive sends and Terms §5A.5

• Flows that send business-initiated messages (WhatsApp templates, Outbound, follow-ups outside the 24h window) need real recipient consent.
• You own opt-in and opt-out. Concierge keeps an audit trail of send authorizations (Terms §5A.5).
• Prefer a less invasive channel when the goal allows it.
• Do not publish a proactive WhatsApp automation if the channel failed Verify setup or the go-live checklist.

Design guardrails

• One start trigger per automation (the product enforces this).
• Keep conditions readable (priority, tag, stage). Avoid trees only the author understands.
• Add notes and context on handoff and approval actions so the human sees why.
• Do not try to re-fire the same CRM mutation in a loop. The runtime already blocks automation-originated echo.
• Align quiet hours and business hours with how Sessions handles handoff.

Pre-publish checklist

• Clear name and description the team will still understand in three months.
• Correct trigger (lead captured, handoff, card move, visitor message, and so on).
• Conditions cover the happy path and what should be ignored.
• High-impact actions have approval or a human notification.
• Test path passed on the graph.
• At least two or three conversation simulations (happy, angry, asks for a human).
• If it sends messages: opt-in, correct template category, healthy channel.
• A named owner who can disable the flow if it misbehaves.
• Team notified on the day you publish.

After publish

Watch the approvals queue and early run history for a week. Too many handoffs on the same theme usually point to a knowledge or guideline gap, not to “add another automation”. If you see opt-outs or complaints, pause first, then fix.