AI Onboarding Coach — Personalised Day 1 to 7 Coaching for New Users That Actually Reduces Churn
Built by NyxVector ·
About this build
A Claude and n8n powered workflow that monitors new user behaviour in the first 7 days after signup, identifies where they are getting stuck or dropping off, and automatically sends personalised coaching messages that guide them to their first meaningful win inside the product.
What was built
The idea came from looking at our signup to activation data and realising we had a 34 percent activation rate which sounds bad until you look at where people were dropping off. It was not random. There were three specific friction points that accounted for almost all the churn — users who never completed step 2 of setup, users who completed setup but never ran their first action, and users who ran one action but never came back after day 1. Generic email sequences were treating all three groups the same. This build treats them differently. How it works: Every new signup triggers a Segment event that starts a 7 day monitoring window in n8n. At the end of days 1, 3 and 7 the workflow checks what the user has and has not done by querying the event history from Segment. The user's activity state gets passed to Claude with a structured prompt that does two things. First it classifies which friction archetype this user matches — Never Started, Setup Incomplete, Activated Once But Gone, or Active and Progressing. Second it generates a personalised coaching message for that specific archetype that acknowledges what the user has actually done, names the specific next step they need to take, and explains in plain language why that step matters for their goal. The message is written in the product's voice not a generic AI voice — the Claude prompt includes the brand tone guide and three example messages from previous human-written communications as style references. The output goes to SendGrid which sends it as a plain text email from the founder's personal address not a marketing template. This was intentional. Plain text from a name outperforms designed HTML templates for onboarding messages by a significant margin in our testing. What the archetypes look like in practice: Never Started receives a message that acknowledges they signed up, assumes they got busy rather than lost interest, and offers a single concrete 2 minute action to get started. No pressure, no urgency, just a clear invitation. Setup Incomplete receives a message that names exactly which step they stopped at and explains what becomes possible once they complete it. Not a generic nudge — a specific acknowledgement of where they are. Activated Once But Gone receives a message that references what they actually did on day 1 and asks a genuine question about whether it worked. This is the highest performing message in the sequence because it feels personal even though it is automated. Active and Progressing receives a message that acknowledges their progress and introduces one advanced feature they have not discovered yet that is relevant to what they have already been doing. Results after 8 weeks: Day 7 retention improved from 27 percent to 41 percent. Activation rate improved from 34 percent to 52 percent. The Never Started archetype had the highest recovery rate at 38 percent of people who received that message going on to complete setup within 24 hours. What I would build next: A feedback loop where users who churn despite receiving all three messages get tagged in Airtable with their archetype for manual review. The hypothesis is that certain archetypes that do not respond to automated coaching need a real human touchpoint rather than a fourth automated message. Also want to add a success message that fires when a user hits their first meaningful milestone — not a congratulations from the system but a genuine message acknowledging what they just achieved and what it unlocks next.