Growth · 16 min

Referral Programs That Actually Work

The mechanics behind Dropbox / Robinhood-style loops.

A referral program without a viral coefficient math is just a discount code.

tl;dr
  • 1 referral loop beats 5 paid channels when Clay tracks invite quality and Attio closes the handoff.
  • Dropbox-style rewards die without 2-step activation, PostHog events, and a 24h reward claim SLA.
  • Use 11x or Lindy for nudges; humans only handle edge cases, not 1,000 invites.
  • Framer + Relume gives the referral landing page; n8n + Trigger.dev wires the logic.
  • If you cannot attribute 70%+ of referrals in Northbeam or PostHog, the program is noise.
Action plan
Action plan0/3 · 0%
Progress saved locally. Sign in to sync

Design the loop before the reward

A referral programme fails when the reward is the strategy. In 2026, the loop must create 2 outcomes: a useful invite and a faster first value moment. Use Clay to score which users can invite 3 high-fit contacts, then enrich those contacts with Apollo AI or Ocean.io before sending. Store the state in Attio, not a spreadsheet, and trigger the flow with n8n or Trigger.dev. Framer + Relume should host a 1-screen referral page with 1 action, 1 claim rule, and 1 deadline. If the invite cannot be tracked in PostHog within 5 minutes, the loop is too fragile.

  1. 01

    Score who should refer

    Use Clay + Clay Chat to rank users by tenure, product depth, and network fit; only target the top 20%.

    Clay prompt: 'Score this user 1-100 for referral likelihood using tenure, activity, and ICP overlap.'
  2. 02

    Route the offer

    Push qualified users into Attio and let n8n trigger the referral prompt after a threshold event in PostHog.

    If PostHog shows 3 activated sessions, n8n sends the referral CTA through 11x or Lindy.
  3. 03

    Make the page disposable

    Build the referral landing page in Framer + Relume, then ship it on Vercel with 1 CTA and 1 claim rule.

    Framer page copy: 'Invite 3 peers, claim 1 month free, expires in 7 days.'
IfThen
the inviter cannot be enriched in Clay with 3 usable signalsdo not show the referral CTA; wait for 1 stronger activation event.
the referral page needs more than 2 clicks to understandremove 1 step and rewrite in Relume before launch.

Reward behaviour, not volume

The best programmes in 2026 pay for quality, not spam. Robinhood-style cash works only when the referred user completes 1 real activation; Dropbox-style storage works when marginal product value is obvious. Use Attio to tag referrer, referee, and reward state, then let Claude Code or Cursor write the SQL that defines a valid event sequence in PostHog. Keep the reward below 8% of first-year gross margin, and use Northbeam to check blended payback. 11x or Relevance AI should handle follow-up nudges so the team never manually chases reward claims.

  1. 01

    Define valid referral

    Use PostHog or June to require 2 events: signup plus the first value action, not just email capture.

    Valid sequence: invite_clicked -> signup_completed -> first_project_created.
  2. 02

    Set a capped reward

    Use Attio fields for reward tier and cap, then calculate margin-safe payouts in SQL through Cursor.

    Cursor SQL: 'select referral_id, min(award, 0.08 * gross_margin_12m) from referrals;'.
  3. 03

    Automate the chase

    Trigger 11x or Lindy to send claim reminders at 24h and 72h, then close stale cases in Attio.

    Lindy prompt: 'Remind only when claim_status = pending and activation = complete.'
IfThen
a reward can be claimed before first valueblock launch; fraud will eat the programme in 1 month.
the reward exceeds 8% of first-year gross marginswitch to a non-cash offer or cut the headline value by 50%.

Automate the invite layer with agents

Manual referral management does not scale past 500 active users. Use Clay for enrichment, then hand the outreach layer to 11x, Lindy, or Relevance AI, with n8n or Inngest orchestrating the state machine. Common Room can flag community advocates, while Apollo AI can surface lookalike accounts for second-degree invites. The workflow should be simple: identify, validate, prompt, track, reward. Put all state in Attio, and let MCP servers expose referral context to Claude 4.5 Sonnet for reply classification. If a human is touching every invite, the system is already broken.

  1. 01

    Find advocates

    Use Common Room and Clay to surface users with strong product signals and community activity.

    Clay enrichment: 'Pull role, company size, and recent engagement for all users with 5+ activated events.'
  2. 02

    Run the agent

    Let 11x, Lindy, or Relevance AI send the invite nudge, using one approved prompt and one fallback rule.

    Agent prompt: 'Ask for 3 peers who match this ICP and offer the referral link only after reply.'
  3. 03

    Orchestrate state

    Use n8n or Inngest to move users between pending, invited, activated, and rewarded states in Attio.

    Inngest event: 'referral.claimed' triggers reward issuance and Slack alert to finance.
IfThen
the referral process needs a manual spreadsheet updatereplace it with Attio + n8n the same week.
reply classification takes more than 10 minutes of human review per 100 repliesmove the classifier into Claude 4.5 Sonnet via MCP.

Measure the loop like a product surface

Referral programmes are product systems, not marketing campaigns. Track invite creation, open rate, referral signup, activation, reward claim, and 30-day retention in PostHog, then cross-check revenue in Northbeam or June. Use Claude Code or Cursor to write the SQL that joins Attio referral IDs to product events, because most teams still misread source-of-truth drift. Run 1 A/B at a time: incentive type, claim deadline, or invite copy. If the loop cannot beat paid CAC by 30% after 60 days, cut it or rebuild the activation event.

  1. 01

    Instrument every hop

    Send referral events from Vercel to PostHog and mirror business states into Attio for clean joins.

    PostHog events: invite_sent, invite_opened, signup_started, activated, rewarded.
  2. 02

    Join the data

    Use Cursor or Claude Code to write SQL that links referral IDs, revenue, and retention by cohort.

    SQL prompt: 'Join Attio referrals to PostHog activation and Northbeam revenue by 30-day cohort.'
  3. 03

    Run one test

    Change only 1 variable per test so June or PostHog can attribute the delta without noise.

    Test rule: '7-day expiry versus 14-day expiry, nothing else changes.'
IfThen
you cannot tie 70% of referred revenue to a referral_idstop optimising creative and fix attribution first.
the loop underperforms paid CAC by more than 30% after 60 dayschange the activation event or retire the incentive.
Pitfalls

What goes wrong

Mistake

Paying for signup instead of first value

Fix

Require 2 tracked events in PostHog before Attio issues any reward.

Mistake

Using a generic incentive for every segment

Fix

Use Clay to segment by role, company size, and intent before offering cash or credits.

Mistake

Letting finance manually approve every claim

Fix

Use n8n or Inngest to auto-approve valid claims under a hard cap.

Mistake

Measuring referrals only in the CRM

Fix

Join Attio to PostHog and Northbeam so revenue and retention are visible in 1 view.

Mistake

Running invite blasts with no persona fit

Fix

Use Common Room and Apollo AI to target only high-signal advocates and lookalikes.

FAQ
What is the simplest referral structure that works in 2026?

Use 1 invite action, 1 activation requirement, and 1 capped reward. Clay scores eligible users, Attio stores referral state, and PostHog confirms the activation event. If Framer + Relume cannot explain the offer in 1 screen, the structure is too complex. Keep the loop narrow, then automate with n8n or Inngest.

Should referrals pay cash, credits, or product expansion?

Choose the cheapest incentive that matches the product. Cash works for high-frequency workflows, credits work for usage-based products, and expansion works when marginal cost is near zero. Northbeam should prove a <3 month payback, and Cursor can model the margin impact. If the incentive attracts low-fit users, Clay should narrow the segment before launch.

How do you stop referral fraud?

Require 2 events before payout: signup and a meaningful activation in PostHog. Store claims in Attio, then use n8n to block duplicate domains, reused payment methods, and self-referrals. Clay can enrich suspicious accounts, and Claude 4.5 Sonnet via MCP can classify edge-case replies. If the claim can be paid in under 24h without checks, fraud will spike.

When should an agent handle referral outreach?

Once the system exceeds 500 active users or 100 weekly invites, hand the follow-up to 11x, Lindy, or Relevance AI. The agent should only send approved prompts, not invent offers. Use Common Room for advocacy signals and Attio for state. If humans still touch every reminder, the workflow is 1 generation behind.

What kills most referral programmes?

Most programmes die from weak activation, vague attribution, and uncapped rewards. PostHog must show the exact event path, Northbeam must verify revenue, and Attio must own the lifecycle. Clay should qualify who gets asked, not everyone. If the programme needs a weekly meeting to stay alive, the loop is not real.

Get one system like this in your inbox each week

No fluff, no upsell. Unsubscribe anytime.