Four models
Four interconnected models, each answering a different question. Data is the thread that connects them. The Mentoring Model describes how data is generated. The Assessment Model describes what the data reveals about where a startup stands. The Value Growth Model describes what it reveals about where the startup should go. The Aggregation Model combines assessment data across startups into cohort, portfolio, and ecosystem views.
Assessment Model
Evaluates the current state of a startup. Backward-looking. Every pillar is scored and independently evidence-graded on a five-level scale from E1 (assumption) to E5 (transactional). The critical rule: a high pillar score at E1 evidence cannot pass a gate. Confident assumptions are still assumptions. This single rule eliminates the most common failure mode in pitch-based evaluation — mistaking conviction for validation. The model produces a quantified valuation range by combining pillar scores with evidence multipliers through a three-layer architecture: evidence-weighted pillar valuation, traditional method triangulation for early-stage startups (Berkus, Scorecard, Risk Factor Summation, First Chicago, CCA), and dynamic factors including coachability premium, evidence velocity, and self-awareness adjustment.
Value Growth Model
Forward-looking. Where the Assessment Model produces a snapshot, the Value Growth Model produces a trajectory. It identifies the highest-impact value gaps in the assessment, converts each gap into a specific evidence discovery task, sequences those tasks by dependency and impact, and projects the value growth that would result from successful completion. It also handles pivot, persevere, and refine decisions: when evidence discovery produces negative results, it determines which pillars carry forward, which reset, and what the value implications are. Negative discovery is treated as an asset — the founder now knows something they didn't.
Mentoring Model
The engine that produces the raw material the other models consume. It defines how diagnostic data is generated through structured coaching — adapting approach in real time based on founder archetype, sentiment, and response patterns. Eight coaching styles shift fluidly within a single session: Socratic questioning, direct challenge, supportive reframing, educational explanation, storytelling, curious scepticism, provocative framing, and silent reflection. The model detects deflection, tracks coachability as diagnostic data, monitors founder wellbeing, and manages multi-session continuity including mentor trust capital that accumulates across sessions.
Aggregation Model
Combines assessment data across multiple startups into three organisational views. A cohort dashboard serves time-bounded programmes — accelerator batches, university semesters — answering "which teams need help now, and will they be ready by demo day?" A portfolio dashboard serves open-ended collections — VC funds, incubator rosters — answering "where should I deploy attention and capital?" An ecosystem dashboard spans institutions across a geography — answering "where is the pipeline weakening, and what structural interventions would help?" Global reach → Each view surfaces different patterns: peer comparison within a cohort, stall detection across a portfolio, systemic gaps across an ecosystem.
No model operates in isolation. The assessment informs the value growth guidance. The guidance produces evidence discovery tasks. Those tasks generate new data. New data updates the assessment. The cycle repeats — and every repetition makes the system more accurate.
See the 14 documents these models produce → · What are the 16 value growth pillars? →
How it was built
The methodology behind The Startup Mentor™ existed as years of accumulated mentoring experience — pattern recognition, situational judgment, calibrated intuition about which founders would succeed and why. This is tacit knowledge: expertise that experienced practitioners possess but cannot easily articulate, document, or transfer.
Traditional approaches to making tacit knowledge explicit — interviews, case studies, process documentation — capture fragments at best. The expert knows more than they can say. The gap between what a mentor does in practice and what they can describe in theory is enormous. This gap is why mentoring quality varies so dramatically and why scaling mentorship has proven so difficult.
Symbiosis: a new method for tacit knowledge transfer
The system was built through a process we call symbiosis: sustained, iterative co-evolution and mutual learning between the domain expert and the AI system. Rather than attempting to document the methodology upfront and then implement it, the domain expert worked directly with the AI across dozens of structured sessions. Each session surfaced new layers of tacit knowledge. Each layer was formalised, tested, challenged, and refined through the collaboration itself.
The technique works because the AI asks the right kind of wrong questions. When the system misapplies a principle, the expert's correction reveals an implicit rule that was never written down. When the system produces an output that "feels off," diagnosing why exposes calibration knowledge that the expert didn't know they had. The errors are not failures — they are extraction events. Each one makes the previously invisible visible.
Symbiosis matters because it solves a problem that has resisted every previous approach. Tacit knowledge — the expertise that makes one mentor dramatically more effective than another — has historically been untransferable at scale. You could shadow the expert, apprentice under them, or interview them exhaustively, and still capture only a fraction of what they actually do. Symbiosis captures what no interview can: the implicit rules that the expert follows without being able to articulate them, because those rules only surface when the system gets them wrong.
The implication extends well beyond startup mentoring. Any domain where expertise is tacit — medical diagnosis, investment judgment, executive coaching, craft skills — faces the same transfer problem. Symbiosis offers a repeatable method: put the expert in sustained iterative dialogue with an AI system, let the system's errors drive extraction, and build the knowledge architecture layer by layer. The resulting system is not a replacement for the expert. It is the expert's knowledge made explicit, transferable, and scalable.
This is not prompt engineering, nor is it fine-tuning. It is both an explicit and tacit knowledge architecture — a structured transfer of tacit expertise that would otherwise remain locked in one person's head. The AI brings reasoning capability, structural rigour, and implementation power. The human brings domain expertise and calibrated judgment. Neither could produce the system alone.
Bootstrapping through simulation and symbiosis
Before the system was tested with real founders, it was bootstrapped through simulations. Detailed startup personas were constructed — complete with business models, founder archetypes, evidence profiles, and psychological characteristics. The AI mentor then conducted full assessment sessions with these simulated founders, applying the complete methodology: pillar scoring, validation assessment, gate evaluation, coaching adaptation, deflection detection, and value growth guidance.
A critical property of this approach: the AI mentor's responses during simulations were exactly what they would have been when confronted with a real person. The system used the same functionality, the same decision logic, the same evidence standards, and the same coaching dynamics it would apply in a live session. The simulations were not simplified test cases — they were full-fidelity exercises of the complete system.
The simulation phase revealed structural gaps in the explicit knowledge that only became visible under pressure. Edge cases in tarpit detection, unexpected interactions between founder archetypes and coaching styles, evidence quality ambiguities that required new classification rules — all surfaced through simulation and were resolved before real founders encountered them.
The symbiosis phase revealed gaps in the tacit knowledge that only became visible when the expert detected that something was not quite right but couldn't quite articulate why. These gaps would be clarified and formalised in deep exploratory discussions with the AI system — conversations where the act of explaining why an output felt wrong made the implicit rule explicit for the first time. See the solution → · See a sample assessment → · Sample value growth report →
The result
What emerged is not a chatbot that has been told to act like a mentor. It is a comprehensive, internally consistent system of expertise that operates at the same diagnostic rigour as the expert it was extracted from — but at scale, with perfect memory, and with structured data outputs that the human mentor could never produce alone.
Every session generates data that flows through all four models simultaneously. The Mentoring Model adapts the coaching in real time. The Assessment Model scores what it observes. The Value Growth Model identifies gaps and generates tasks. The Aggregation Model combines individual assessments into cohort, portfolio, and ecosystem intelligence. And the cycle continues — because every new session generates new data that further refines the system.