Sequences fire. Activation stalls. How we'd fix an AI app's lifecycle.

Concept project: fictional brand, built to show our process. We won't invent client results, so we show the whole process instead. No client data was used. Every brand number below is invented and labeled as such, every benchmark is sourced, and every outcome is a target, never a claimed result.

Meet Notewell

Notewell is an invented AI meeting assistant that turns every meeting into decisions, owners and follow-ups. It is a Series A company with about $2.8M in ARR, 6,500 signups a month, a 14-day no-card Pro trial at $18 per user, and a Customer.io account that is barely used. Its event stream exists and sits silent. The story is simple: sequences fire, activation stalls.

What we found

1. Everything is a countdown, nothing is a trigger. Three calendar emails, no product events. Triggered emails earn roughly 76% higher opens and 152% higher click-through than timed sends.

2. One sequence for two different users. A user with 14 summaries and a user with zero meetings get the same day-13 upgrade email.

3. The setup cliff has no fence. 34% never connect a calendar and are never contacted again.

4. The trial-ending email wastes its best moment. A generic countdown instead of a recap of what the user would lose. Loss-framed copy converts 21 to 32% better.

5. Expansion signals fire nothing. Teammate invites and 80% plan usage are invisible. Product-qualified leads convert at about 17.4% against 4.6% for pure self-serve.

6. Involuntary churn writes itself off. No dunning sequence, even though dunning recovers 20 to 40% of failed payments.

7. The distribution loop is unused. The shared recap email to non-users is the category's proven acquisition loop, and Notewell sends none of it.

8. Reporting runs on opens, not outcomes. The fix is a measurement frame built on activation, trial-to-paid, NRR and reactivation.

What we'd build

An event-triggered lifecycle map: verification recovery, a welcome that sets the expectation that the first meeting is the test, setup recovery for anyone without a calendar after 24 hours, a milestone email on the first summary, a loop email to non-users who view a shared recap, a teammate-invite expansion email, and a usage-limit email at 80%.

The trial-ending email splits into two tracks. Activated users get a loss-framed recap of what they keep. Stalled users get a value rescue with no upgrade pitch and a trial-extension fallback. Dunning, re-engagement using the user's own last recap, and win-backs segmented by churn reason complete the program. Guardrails: never more than two emails in the first 48 hours, and time is a guardrail while events are the engine.

What we'd measure

All of these are targets, not results.

• Signup to activation: 31% to 42% (Benchmark: median 37 to 38%, AI and ML products about 55%)

• Trial to paid, blended: 4.1% to 9%, tracked per track (Benchmark: no-card median 4 to 6%, good 8 to 15%)

• Net revenue retention: 98% to 105% (Benchmark: median 106%)

• Calendar connected: 66% to 80%

• Dunning recovery, reported as a percentage of failed revenue

Opens and clicks are reported as diagnostics, never as goals.

How to read this

Three kinds of numbers appear on this page. Invented numbers are part of the fictional brief. Benchmark numbers come from published sources and are attributed. Target numbers are what we would design toward. When we have real client results, they get their own pages, held to the same standard.

What a real engagement looks like

A real engagement runs in three steps: the audit, the foundation build, then a monthly retainer. The audit is free for a limited time, and you keep the written report either way. It covers deliverability, flow gaps, segmentation, list health, and what closing the gap is worth in revenue.

Get the audit