CRM Leads - Automated Follow-up (n8n)

Context

I maintain a professional-leads CRM in a Notion database (status, interest level, next step, notes, next-message date). Until now: manual drafting of each follow-up plus manual addition of a TickTick reminder task. Repetitive, time-consuming, risk of forgetting.

Business goal: a single action (updating the next-message date in Notion) automatically triggers the drafting of a personalized message and the scheduling of a reminder task.

My role: end-to-end design and implementation of the workflow, from requirements analysis to production deployment.

Key achievements

End-to-end CRM follow-up automation

Event-driven n8n workflow that listens to Notion DB updates and orchestrates 3 steps:

1. Smart trigger filtering (only generates when the follow-up date is in the future, avoiding unnecessary LLM calls)
2. Message drafting by an AI agent (Claude) producing a personalized follow-up tailored to the lead context (status, interest, history)
3. Automatic scheduling of a TickTick task that reminds the follow-up at the right time

Technical challenges: generating messages that are genuinely personalized (tone, context, length of the relationship) without falling into empty sales clichés.

Solutions: advanced prompt engineering (strict rules on tone, relative dates, concise style), structured output for reliability.

Notion → Claude → TickTick pipeline without duplicates

Connecting Notion ↔ Claude ↔ TickTick in a single pipeline. The Notion "Page Updated in Database" trigger only returns the row's DB properties, never its markdown body. Yet that's where the rich narrative context of a lead lives (mission, timeline of exchanges, next action).

Solution: a Notion MCP Server attached to the AI agent via an MCP Client Tool node, calling notion-fetch on the page ID at the start of execution to fetch the enriched markdown body before drafting.

Handling edge cases: repeated edits on the same lead (no duplicates), TickTick API without native upsert (match by title → update or create), network errors.

Self-hosted deployment

Workflow deployed on self-hosted n8n (VPS + Dokploy) and versioned in a dedicated repo for backup and change traceability. Dedicated alerting workflow that notifies on execution failures (crash, timeout, API error).

Results

• Time savings: automatic follow-up in ~ 30 seconds vs 3-5 minutes manually
• No more forgotten follow-ups: the TickTick reminder guarantees that no scheduled follow-up is missed
• Quality: contextualized messages consistent with the relationship history

Takeaways

• Advanced prompt engineering for reliable and contextualized AI outputs
• Multi-tool orchestration in n8n (upsert patterns without a native API, filtering)
• Designing internal tooling that saves time (ROI-first approach)