See where learners stall, and why they leave
DecisionBox reads enrolment, lesson activity and completion records overnight, and points to the exact module, group or moment where learners drop off, before you plan the next intake.
Module 3 of Intro to Python collapses session-to-session retention from 87% to 16%.
8,400 active learners · the OOP-basics lesson is the cliff point. 4× the drop of the next-worst module.
Built for VP Student Success, Heads of Growth, and Heads of Product at platforms where learners enroll, progress, and either finish or leave.
What stays hidden when the questions don't get asked.
Three benchmarks every learning-platform leader already accepts. The discovery loop runs against all of them in a single pass — and surfaces where your numbers diverge from the curve.
Average certification rate across HarvardX and MITx courses over four years. The platform-wide number that buries the cohort variance worth investigating.
Source: HarvardX/MITx — Person-Course Academic Year 2015-16Roughly half of MOOC enrollees never return after their first lesson. The activation cliff is sharper than any post-week-1 retention curve a dashboard plots.
Source: Class Central — MOOC Engagement TrendsMedian monthly churn for consumer-subscription learning apps — well above the SaaS median, and skewed by a long tail of single-month learners no dashboard segments by intent.
Source: Recurly — Subscription Churn BenchmarksFrom a finding to a plan
One overnight run. DecisionBox picked the analysis areas, ran them against the platform's data, checked its findings, and ranked what to fix. Three shapes of finding — one cliff in a single course, one pattern across acquisition and retention, one anomaly the catalog average was hiding.
Module 3 of Intro to Python collapses session-to-session retention from 87% to 16%.
- 8,400 active learners · OOP-basics lesson is the cliff · 71-point drop in one module · 4× the next-worst module
Restructure module 3 around checkpoint exercises before the next cohort intake.
- Expected
- 25-40% improvement in Intro to Python completion if module 3 abandonment closes by half · ~8,400 learners in scope
Meta paid-acquisition cohorts activate 2.4× faster — and churn at 3× the rate by week 4.
- $94,200 monthly spend · 1.2× signup volume · 2.4× D1 activation · 0.34× D30 retention · 12,800 acquired
Re-tier paid spend by 30-day retention, not D1 activation.
- Expected
- 35-45% reduction in 30-day churn cost on paid cohorts · no measurable signup volume loss
Aggregate completion holds at 64%. The three most-enrolled courses dropped from 71% to 38% in two quarters.
- 3 flagship courses · 47,100 enrollments (41% of total) · 33-point drop in completion · 12 newer courses at 85%+ completion masking the trend
Refresh content and instructor on the three flagship courses before the next intake.
- Expected
- lifting the three flagship courses back to 60% completion moves the platform aggregate by 11 points
Same data, five sets of questions.
One discovery run produces findings ranked for the team that owns the metric. Each panel is what DecisionBox investigates in your data — questions it picks on its own.
Where the funnel quietly breaks.
Your acquisition dashboard shows volume by channel. It doesn't show which channels feed cohorts that activate and never return. DecisionBox reads source quality against every downstream stage — and ranks where the volume is dying. Pair with Find growth opportunities.
- Which channels feed signups that never start lesson 1
- Where the funnel narrows between signup and first paid course
- Which segments show the steepest trial-to-paid drop-off
- Where landing-page variant × cohort retention diverges
- Which referrals carry the strongest 30-day stickiness
Three properties your next cohort review can defend.
DecisionBox is built so the findings on the cohort-review call are not the kind that fall apart under a follow-up question.
Sees what nobody had time to look for.
DecisionBox picks the questions. It reads every cohort against every lesson, every channel against every retention curve, every flagship course against the catalog underneath. The investigations nobody had time to formally request happen anyway — overnight, ranked by morning.
Every number you'd defend in the cohort review.
A verifier agent checks each finding against your data, and a refuter agent tries to break it. Every finding gets a verdict, such as confirmed, partial, or rejected, and stays visible with it. Every step of DecisionBox’s reasoning stays attached to the finding for review.
Your learner data stays where it already lives.
Run DecisionBox on your own infrastructure, and student names, email addresses, and learning records never leave your environment. The platform is open source, so your team can see exactly how it works and what it touches.
What Education & EdTech security teams ask first
Does our data move?
No. DecisionBox only reads it where it is, and can run inside your own environment, on-prem or air-gapped.
What can the AI see?
Only what you allow. Personal and sensitive data is hidden before the AI sees anything.
Which AI does it use?
The one your company approved, in your cloud account or on your own servers.
Can we audit it?
Every action is recorded: who, what, when and under which rule.
See DecisionBox at work
Your next cohort review question, already on the table.
Let's run it on your data.