Business Context: The Tutoring Market and AI's Role
The private tutoring market in the UK is substantial, estimated at over GBP 6 billion annually, driven by parents who believe one-to-one instruction provides outcomes that classroom teaching cannot. AI tutoring tools address this market in two ways: by providing an accessible substitute for families who cannot afford regular private tutoring, and by providing a supplement for families who already use tutors but want additional practice time between sessions. Edtech companies building in this space need to understand which product position they are taking, because the design implications are different. A substitute product needs to cover curriculum breadth, adapt to different learning styles, and provide the reassurance parents need that learning is happening. A supplement product can be more focused on specific subject areas, practice question generation, and progress tracking that integrates with a human tutor's session planning. Both positions have viable businesses. The regulatory context for UK edtech includes GDPR (with specific ICO guidance on children's data under the Age Appropriate Design Code), and any tool used in UK school settings must be compatible with DfE requirements around data sharing and the GDPR exemptions that apply to educational data.
Architecture: Adaptive Conversation, Knowledge State Tracking, and Study Planning
The system is built around three integrated components. The tutoring conversation layer uses GPT-4o with a Socratic tutoring prompt that asks guiding questions rather than simply providing answers, checks understanding before moving on, and adjusts explanation depth based on student responses. The conversation history is maintained in Supabase and feeds the knowledge state model. The knowledge state tracking layer records which topics the student has engaged with, which they answered correctly on first attempt versus which required multiple explanations, and which concepts they return to most often. This data builds a model of the student's current knowledge state that persists across sessions. The study plan generator uses the knowledge state model to produce a weekly study plan: prioritising topics where the student shows the largest gaps relative to their grade level or exam specification, scheduling spaced repetition for topics that need reinforcement, and generating a balanced mix of explanation and practice. PostHog tracks session engagement metrics: average session length, questions answered per session, accuracy rate by topic, and streak days. These metrics feed the parent or teacher dashboard where progress is reported.
AI Components: Socratic Tutoring with Adaptive Explanation Depth
The tutoring conversation is the core AI component and requires the most careful prompt engineering. GPT-4o is instructed to act as a Socratic tutor: rather than giving the answer when a student asks a question, it asks guiding questions that lead the student to the answer themselves, offers hints when the student is stuck, and checks understanding with follow-up questions before progressing. The system prompt includes the student's current year group, the subject they are studying, the topic being covered, and their recent performance history. This context allows GPT-4o to calibrate explanation complexity (a Year 9 student needs different language from a Year 12 student for the same concept) and to reference the specific exam board's specification content for UK students sitting GCSE or A-Level exams. The most important guardrail in the prompt is the accuracy instruction: for subjects like mathematics, science, and history, factual accuracy is non-negotiable. GPT-4o's accuracy on standard curriculum content is high but not perfect. The system includes a confidence-based routing rule: for mathematical computation and factual recall questions, the system uses a verified answer bank alongside GPT-4o, checking the AI output against the known correct answer before presenting it to the student.
Challenges: Safeguarding, Age-Appropriate Design, and Learning Efficacy
Edtech products for children require safeguarding and data protection controls beyond standard adult SaaS products. The ICO's Age Appropriate Design Code applies to online services likely to be accessed by children and requires: privacy by default, data minimisation, no profiling for marketing, no nudge techniques to weaken privacy choices, and clear and accessible privacy information. SpeedMVPs builds these requirements into the product by default: no third-party advertising cookies, data minimisation in the event tracking configuration, parent-controlled account settings, and a privacy notice written in age-appropriate language. Learning efficacy is both a design challenge and a product credibility challenge. An AI tutor that makes learning feel like a conversation but does not improve learning outcomes is not a tutoring product, it is an entertainment product. SpeedMVPs works with the client to define the efficacy metric that the MVP will track: typically a before-and-after assessment on a set of curriculum topics. Building this measurement into the MVP from day one means the product can generate its own efficacy evidence, which is essential for school and local authority procurement.
Outcomes: Engagement Rates and Learning Outcomes
AI tutoring products that achieve strong engagement tend to share design characteristics: short session structures (20-30 minutes) that fit into homework routines, immediate positive feedback for correct responses and encouraging responses for incorrect ones, visible progress indicators, and a subject scope that is relevant to upcoming assessments. Products designed around these principles consistently achieve day-7 retention rates of 40-60% for students who complete their first session, which is significantly above the average for edtech products. Learning outcome measurement requires a more careful study design and longer timeframe than an MVP can provide. The MVP establishes the measurement framework and collects the baseline data that will support efficacy claims as the user base grows.
Lessons: Teacher and Parent Trust Is the Distribution Channel
Direct-to-consumer edtech products targeting students face a significant discovery problem: students do not search for tutoring tools themselves, parents and teachers recommend them. This means the product's design must satisfy the adults who recommend it, not just the students who use it. Parent dashboards showing progress, session summaries, and topic coverage are not secondary features: they are the primary trust-building mechanism that drives word-of-mouth recommendation. Teachers who recommend AI tutoring tools to their students are making a professional judgment about the tool's educational value. This means the product needs to demonstrate curriculum alignment and learning efficacy, not just engagement. The relationship between edtech products and schools is also a regulatory relationship: schools sharing student data with edtech providers must have a Data Sharing Agreement in place under UK GDPR, and the product must be able to operate within the constraints of typical school DPAs.