Built for edtech apps

The behavior that predicts retention in your edtech app. Found, then driven.

For one edtech app, we found the early behavior that predicts whether a trial user sticks. 34% to 39.5% retention, tested across 24,118 trials. The same 48-hour analysis runs on your cohort.

Book a 48-hour analysis → No commitment · No setup before the call
Baseline retention
34.0%
12-week · n=24,118
After Sendlyr signal
39.5%
+5.5pp · p < 0.001
Behaviors tested
24
1 winning signal
Time to insight
48h
from data access
The Typesy finding

Here's what surprised us. And why it matters for your app.

We compared 24 candidate early behaviors against 12-week retention for Typesy. One milestone stood out on lift, coverage, and significance. It wasn't the one their team would have picked. Your milestone will be different — but the method is the same.

One milestone reached in the first 14 days retained at 39.5% — a 5.5pp lift over baseline. We mask the other signals to protect customer data.

Candidate signal Retention Decision
Key action B in 14 days 37.9% 81% cov · drop
Key action A in 14 days 36.1% 82% cov · drop
Complete ≥1 course within 4 days 39.5% 79% cov · primary

n = 24,118 trials · 12-week window · p < 0.001 · baseline 34.0%

Most edtech teams would have picked "complete first lesson" — we tested it, and it didn't predict retention. The winning behavior was different. Read the full case study →

12-week retention · Typesy
34%
Before
→
39.5%
After
Typesy Typing app · eReflect · n = 24,118
Why guessing fails

The activation milestone most edtech teams pick. Usually wrong.

These are the activation metrics edtech teams default to because they sound right. We've tested all three against retention data. None of them predicted who stays — for Typesy or for the others.

Myth 01 · The obvious one
"Complete your first lesson"
Almost every new user does this — coverage is high but the lift is near zero. It measures intent, not commitment.
Myth 02 · The engagement one
"Sign in 3 times in week 1"
Catches the high-intent users — who would have stayed anyway. Lifts engagement metrics, not retention.
Myth 03 · The onboarding one
"Finish onboarding flow"
Onboarding completion is a proxy at best. Users who finish onboarding aren't the users who pay.
For your app

We don't guess. We test these against your retention.

For an edtech app, we typically test 15–25 candidate behaviors against your retention or trial-to-paid outcome. The winner is whatever has real lift + coverage in your cohort. Examples we test:

Completion
First lesson finished
% complete in 24h / 72h / 7d
Depth
Course progress threshold
≥1 course done in 4 / 7 / 14d
Consistency
Practice streak
3 / 5 / 7 consecutive days
Performance
Quiz / test score reached
≥80% on first assessment
Time-on-task
Cumulative session minutes
≥60 min in first 7 days
Return
D2 / D7 return visit
at least one return session
Milestone
Certificate / badge earned
first achievement unlocked
Habit
Notification opt-in + open
reminder followed within 48h
Breadth
Feature exploration count
≥3 features used in 7d
Social
Friend / classroom invite
at least 1 invite sent
Personalisation
Goal / level configured
profile setup completed
+ 4–14 more
Whatever's in your data
behaviors unique to your product
We test all of them in parallel against 12-week retention (or trial-to-paid). The winner is the milestone — different for every product, never a formula.
After we find it

Every trial user gets a state. Every day.

Once the milestone is set, each user is classified daily based on where they are relative to it. The state determines which email they get — and whether anything goes at all.

NEW_LEARNERdays 0–1
FIRST_LESSON_DONEday 1–3
MILESTONE_REACHED≤ 4 days
STUCK_BEFORE_MILESTONEdays 3–7
HIGH_INTENTpaid before signal
CHURN_RISKno session 5+ days
DORMANT_3Dsilent 3 days
DORMANT_7Dsilent 7 days
TRIAL_EXPIREDpast day 14
How it works

Three steps. One milestone.

No new SDK, no schema changes, no rebuild. We read from the behavioral data you already capture — and you stay in control of every send.

01 · Connect
Amplitudelesson · quiz · return
Mixpanelsession · streak
Firebaseprogress · badges
Read your existing learner events
Lessons, sessions, quiz attempts, badges — whatever you already track. We add nothing to your stack. Read-only access to a DB user is the entire integration.
02 · Analyze
lesson_done62%
d2_return71%
course_done92%
streak_3d48%
quiz_score55%
Find your activation milestone
We score every candidate edtech behavior against your retention outcome. The winner becomes the milestone. 48 hours from data access to the report — your milestone, your data, your call.
03 · Drive + approve
Milestone reached · day 3sent
Stuck on lesson 2 · day 5staged
First lesson donesent
No session 5dqueued
State-matched emails, your approval
Each user state gets its own template — drafted by us, approved once by you, then personalized at send time. Launches in monitoring mode. Nothing leaves the queue until you flip the switch.

Want the full 5-step detail? How it works →

FAQ for edtech founders

Objections we've heard from edtech teams.

Our users are minors — does this work under COPPA/FERPA?
Yes. By default we work with anonymized user IDs only — no names, emails, or PII pass through our analysis. If personalization is needed at send time, PII is fetched from your system and never logged. DPA available.
Our trials are 7 days, not 14 — does the method still work?
Yes. The lookback window is configurable. For short trials we look at days 0–3 or 0–5. The signal still emerges — sometimes more cleanly, because users have less time to drift.
We're B2B / school district sales — is this for us?
Partial fit. If you have an individual learner trial layer (teacher pilots, free seats, freemium), yes. Pure enterprise PO sales cycles aren't what Sendlyr is built for.
What's the minimum cohort size you need?
Practical minimum is ~10,000 trial users in the last 90 days. Below that, the analysis can't reach statistical significance and we'd be guessing. We'll tell you on the call if you're under-scoped.
School year seasonality skews our retention. Can you handle that?
Yes. We bucket cohorts by signup month and look at matched-cohort retention, not raw averages. The September spike and the summer dip don't break the analysis.
We already use Customer.io / Iterable / Braze. Do you replace it?
No. We find the milestone and generate the drafts. You can route the sends through your existing ESP or have us handle delivery — whichever is less work.
What if the analysis finds no strong milestone?
It happens. If there's no behavior with meaningful lift and coverage, we tell you that. You keep the report and the methodology. We part as friends. No obligation.
How long until we see lift?
Milestone in 48 hours. Templates approved and live in 1–2 weeks. A/B reaches significance typically in weeks 4–6, depending on volume.
◆ Get started

Find your edtech app's activation milestone.

We found it for Typesy. The same 48-hour analysis runs on your cohort. No commitment, no setup before the call. Free for qualified teams.

Book a 48-hour analysis →
30-min call No setup before 48-hour turnaround