B12 can deliver personalized learning paths that optimize the experience for each learner—at scale.
By analyzing learner performance, our AI algorithm identifies each individual's needs and builds a custom learning path, selecting activities that align with their specific strengths and areas for improvement.
The goal of B12's AI-driven adaptative program is to help every learner reach a set of global performance criteria, tailored to their unique learning needs. For example, high-performing learners move quickly through the content and access more advanced activities faster, while those needing additional support follow longer, more gradual learning paths that begin with foundational content.
With the Adaptive Reinforcement program type, B12 personalizes each learner's program in just a few clicks. The app automatically assigns tailored learning sequences from your activity bank, making it possible to deploy personalized programs to thousands of users within minutes.
As with all B12 programs, intuitive dashboards provide real-time tracking and detailed analysis of learner progress.
Key Principles of How Adaptive Programs Work
Scheduling and Sequence Configuration
The timing of learning sequences and the number of questions per sequence are predefined by instructional designers.
Example: Sequences are scheduled for every Monday and Thursday, with 5 questions in each.Question Pool Design
A dedicated question pool must be created in the Studio. This pool should:Cover multiple knowledge topics.
Include questions tagged with one of three difficulty levels: easy, medium, or difficult.
Contain at least ± 50 questions to ensure the algorithm functions effectively.
AI-Based Question Assignment
For each sequence, the AI algorithm selects questions for each learner based on their individual performance data.
Example:On Monday, Mary might receive questions 1, 5, 7, 8, and 14.
Peter might get questions 2, 12, 16, 22, and 36.
Completion Criteria
A learner completes the program when two conditions are met:They have reached the expected performance threshold across all three difficulty levels.
They have answered every question in the pool at least once.
Tailored Progression
The performance threshold is defined in the program settings and may vary from one program to another. All learners finish with results at or above this benchmark, but their journey is individualized:Some may progress quickly and complete the program with fewer repetitions.
Others may require more time and practice to reach the same level of mastery.
How Does the Algorithm Select Questions for a Learner's Sequence?
Step 1 – Predicting the Success Rate of Each Question
For every question in the database, the algorithm estimates the learner’s probability of answering it correctly—this is called the predicted success rate.
If the algorithm is confident the learner will answer correctly, the predicted success rate will be close to 100%.
If it expects the learner to get it wrong, the rate will be closer to 0%.
This prediction is updated every time a new sequence is generated, since learners improve over time and their likelihood of success changes throughout the program.
What Data Is Used to Make This Prediction?
The algorithm relies on data from both the learner and all other learners in the program, using the following key variables:
Success rate by knowledge topic and difficulty level, based on previously answered questions
Confidence level by topic and difficulty, also based on past responses
Number of activities and sequences completed for each topic and level of difficulty
Response time for each question, categorized by topic and difficulty level
These insights help the algorithm tailor each sequence to the learner’s evolving skill level, ensuring a continuously personalized learning experience.
Step 2 – Selecting Questions to Include in the Sequence
When choosing which questions to include in a learner’s sequence, the AI aims to select those that offer an optimal challenge—neither too easy nor too difficult.
Specifically, it looks for questions where the predicted success rate is around 50%, meaning the algorithm is uncertain whether the learner will answer correctly or not. These are the moments where learning is most effective.
To determine which of these 50% prediction questions make it into the sequence, the AI also follows several additional rules, including:
Balancing new vs. repeated questions
Ensuring coverage across different knowledge topics
Maintaining a mix of difficulty levels
These rules help create a well-rounded, targeted sequence for each learner.
For more details on how these algorithmic rules are configured, please contact your B12 Account Manager.
Conditions for Optimal Algorithm Performance
To ensure the algorithm performs effectively and delivers accurate predictions, the following minimum conditions must be met:
✅ At least 50 questions in the activity bank
✅ Questions must be well distributed across the various topics covered in the program
✅ Only supported activity types (questions with feedback), including:
Single answer
Multiple answer
Hotspot image
Open-ended questions
✅ Balanced distribution of questions across all levels of difficulty (easy, medium, difficult)
✅ Minimum of 50 participants in the program
If these minimum thresholds are not met, the program may still function, but the algorithm’s predictions may be unreliable.
🚩Important Notes
Prediction Reliability Failsafe
If the algorithm detects that prediction reliability falls below a defined threshold, B12 will automatically deactivate the AI.
In that case, questions will be selected randomly rather than based on predicted success rates.Unsupported Activity Types
Content-based activities cannot be included in adaptive programs. These include:Text and media
Surveys
Hands-on activities
Key skill and behavior tracking