EdTech
Build learning products that adapt to how people actually learn
Learning platforms generate a rich stream of behavioral data, but much of it never becomes part of the product experience. We help EdTech teams use that data to improve retention, personalize learning, match people and content more effectively, and scale assessment and feedback workflows.
Discuss your learning product or retention challenge
Retention intelligence
Know who is drifting away before they disappear
Learner trajectories are illustrative and do not represent production users, probabilities or measured outcomes.
Most learning platforms can tell you who completed a lesson or opened the app. The harder question is who is quietly losing momentum and what to do about it.
We build the data pipelines, engagement models and product workflows that turn learner activity into useful signals: declining participation, changing learning pace, incomplete journeys and cohorts that behave differently from the average.
That creates a foundation for targeted interventions instead of treating every inactive learner the same way.
What this can include
- Learner engagement and retention scoring
- Dropout or churn prediction
- Cohort and behavior analysis
- Intervention triggers and re-engagement workflows
- Funnel analysis from registration through payment and active participation
Personalization
Replace static learning paths with behavior-driven recommendations
Many products describe themselves as adaptive while relying on a growing set of manually maintained rules. That works until the product, curriculum and learner population become more complex.
Static rules
- Finish lesson A → unlock lesson B
- Score below X → repeat the exercise
- Manually maintained conditions
Behavior-driven recommendations
Actual learner behavior, performance, pace, progression and engagement patterns can shape what happens next.
The model does not replace curriculum design. It gives the product a better way to apply it to individual learners.
Matching
Better matches between learners, tutors, groups and content
For social and marketplace-style learning products, matching often determines whether the experience works at all. Language, level and availability are only the obvious variables. Good matching may also depend on goals, behavior, engagement patterns, group dynamics and previous outcomes.
We treat matching as a scoring and recommendation problem rather than a collection of rigid filters.
- Language
- Level
- Availability
- Learner goals
- Behavioral fit
- Engagement patterns
- Group dynamics
- Previous outcomes
Matching applications
- Learner-to-tutor matching
- Learner-to-group formation
- Peer or conversation-partner matching
- Learner-to-course or content recommendations
- Ranking when several acceptable matches are available
Our team brings production experience with scoring and recommendation systems from other large-scale, data-intensive platforms and applies the same engineering principles to learning products.
Teaching workflows
Scale teaching workflows without removing the teacher
Use AI where language and feedback are the bottleneck
Assessment, practice and personalized feedback are valuable precisely because they require attention. That also makes them expensive to scale.
LLMs can handle bounded parts of this workload without pretending that every educational decision should be automated.
- Conversational language practice
- Structured feedback on learner responses
- First-pass assessment and classification
- Explanation and practice-generation workflows
- Teacher or tutor copilots
- Content transformation and adaptation
Where accuracy or educational judgment matters, the workflow keeps a human reviewer involved rather than treating the model as an unquestionable answer engine. AI helps scale the workflow; it does not replace the teacher.
The foundation
Useful AI starts with useful product data
A recommendation model or AI tutor is only the visible part of the system. Underneath it are event tracking, learner identity, content metadata, payments, enrollment states, evaluation datasets, integrations and product workflows. If that foundation is inconsistent, adding a model rarely fixes the problem.
- 01
Product engineering
Learning flows, onboarding, payments, cohort and group logic, tutor and learner interfaces.
- 02
Data engineering
Event pipelines, behavioral datasets, analytics and model-ready data.
- 03
Applied AI/ML
Scoring, recommendations, prediction, matching and LLM-based product features.
- 04
Production integration
Connecting models to an existing product, production workflows and maintainability — without forcing a platform rewrite.
Engagement
Add a focused delivery stream without rebuilding your organization
EdTech teams often have product knowledge in-house but lack enough senior engineering capacity for a new data or AI-heavy roadmap.
Forma Pro can join as an embedded senior team or take ownership of a bounded delivery stream. Product direction and educational decisions stay with your team. We take responsibility for the engineering implementation needed to turn them into a working, maintainable product.
See how we workBounded streams we can own
- Learner retention and engagement intelligence
- Recommendation or matching systems
- AI-assisted learning workflow
- Analytics and event infrastructure
- New product flow
- Major product rebuild
A practical starting point
Start with one learning-product problem
You do not need an “AI strategy” before talking to us.
Bring us a retention problem, a matching problem, an assessment bottleneck or a product flow that is not performing the way it should. We can look at the existing product and data together and determine what is worth changing.
Discuss your learning product or retention challenge