AI Quality
Built to be trusted. Not to sound clever.
You have used AI tools that confidently present made-up facts. So have we. myAIstrategy is designed from the ground up to prevent that: deterministic scoring, verified sources, multi-layer quality checks and a human strategist who signs off before anything reaches you.
6
Trained advisors
curated knowledge per role
7
Quality layers
before you see results
30+
Sanitisation rules
strip AI artifacts
0
AI-generated scores
all deterministic code
The Problem
Why most AI output cannot be trusted
A general-purpose AI assistant asked to assess your business will generate plausible text. It will sound confident. It will cite sources that do not exist. It will invent statistics. It will describe competitors that were acquired three years ago as active threats. It will give your government agency advice about “customers” and “revenue”.
This is not a criticism of the underlying models. It is a design problem. A single prompt with no structure, no grounding data, no validation and no domain expertise will produce output that reads well but does not hold up under scrutiny.
myAIstrategy is built differently. Every stage of the analysis pipeline has specific guardrails. Scores are calculated, not generated. Sources are verified against the research corpus. And every AI Strategy deliverable is reviewed by a senior strategist before it reaches you.
Trained Advisors
Six advisors trained on a curated knowledge corpus
The six advisors are not personality wrappers over a general-purpose chatbot. Each one is loaded with a specific library of consulting frameworks, methods and reference material relevant to their discipline. When Danni discusses strategic positioning, she works from a curated set of strategy frameworks. When Marcus assesses your technology stack, he applies build-versus-buy methods and AI adoption playbooks. Jordan builds business cases using the same financial modelling approaches a CFO would expect.
Danni
Strategy Advisor
Strategic positioning, business model design, portfolio strategy, value chain, McKinsey 7S, PEST analysis
Marcus
Technology Advisor
AI adoption, hire-vs-agent decisions, role redesign, rapid innovation, pre-mortem analysis, risk radar
Miles
Revenue Advisor
Customer persona development, negotiation, initiative prioritisation, cost-benefit analysis
Alex
Leadership Advisor
Executive decision-making, change readiness, team building, delegation, difficult conversations
Taylor
People Advisor
Workforce strategy, role redesign, capability gaps, organisation impact, team pulse, debriefs
Jordan
Financial Advisor
Business case modelling, investment sizing, ROI stress-testing, financial governance
The advisors also share a common reference library covering AI strategy, transformation methodology and Australian regulatory context. For government scans, this includes the live policy documents agencies are actually required to comply with.
Reference library for government scans
National Framework for Assurance of AI in Government (Finance, 2024)
Policy for Responsible Use of AI in Government v2.0 (DTA, 2025)
Australia’s 8 Mandatory AI Ethics Principles (DISR, 2025)
AI and the Trustworthiness of Public Service Delivery (PM&C, 2023)
Guidance for AI Adoption (DISR)
The reference library is curated by hand, not scraped. It is the same body of material a senior strategist would draw on when advising a board. The corpus is loaded directly into each advisor’s context so their analysis is grounded in current frameworks rather than recalled approximations from a training dataset.
Scoring
Your scores are calculated, not generated
The most important design decision in the platform: the AI Disruption Score, quadrant scores, dimension scores and all deep-dive financial figures are calculated in deterministic code. Industry-specific weight tables, substitution factors and benefit-case arithmetic are fixed formulas. The AI researches and analyses. It does not pick numbers.
10
Industry weight tables
sector-specific scoring
18
Benchmark organisations
calibration anchors
±2
Max score adjustment
with mandatory evidence
Why this matters: When a general-purpose AI is asked to rate something on a scale of 1 to 10, it produces a number. It will not produce the same number twice, and it cannot explain why it chose 7 instead of 6. Our scoring model uses calibrated rubrics with hard boundaries. A workforce replaceability score above 7 requires evidence that more than 50% of the work is knowledge work. The arithmetic is exact.
Research
Grounded in live research, not training data
Every assessment starts with real-time web research: your website, recent news, financial filings, job postings, peer activity and industry trends. Peer companies are identified from live search results, not recalled from training data. ABN details are verified against the Australian Business Register.
EVIDENCEDVerified from public data: website content, financial filings, job postings, news, ABN records.
INFERREDLogically derived from available evidence. Revenue estimates from headcount and industry benchmarks, for example.
HYPOTHESISA reasoned assessment where direct evidence is unavailable. Clearly marked so you know what needs validation.
When you provide validated data through the strategy workshop, confidence tags upgrade and the underlying analysis changes across every deliverable.
From URL to Deliverable
Six stages, each with its own guardrails
This is not a single AI prompt. It is a structured pipeline where each stage receives only the context it needs, produces structured output and passes through independent quality checks before feeding the next stage. Tap any stage to see its guardrails.
Quality System
Seven layers between the AI and you
Every output passes through multiple independent checks. Some are deterministic code that runs in milliseconds. Some are AI reviewers more capable than the AI that wrote the content. The final layer is a human.
CODEDeterministic checks that run in millisecondsAI REVIEWIndependent AI reviewerHUMANSenior strategist
Guardrails
What the platform will not do
Constraints are enforced architecturally, not by suggestion. These rules are built into every generation call, validated by the quality system and cannot be bypassed by prompt variation.
No fabricated statistics
Every specific number, percentage or dollar figure is either sourced from research data, derived from a deterministic calculation or clearly tagged as an estimate.
No unsupported claims about people
Defamation discipline rules prevent negative assertions about named individuals without sourced citation. Absence language uses "no public evidence of" rather than definitive negatives.
No wrong-jurisdiction contamination
Analysis is anchored to Australian regulatory and market context. US and UK references (HIPAA, GDPR, FDA) are automatically substituted with Australian equivalents.
PII filtered before AI processing
Document uploads and workshop conversations pass through a filter covering Australian-specific identifiers (ABN, TFN, Medicare) and universal patterns (credit cards, bank accounts) before any AI call.
Entity-aware language
Government sees "citizens" not "customers". NFPs see "impact" not "revenue". Education sees "students" not "clients". Every metric adapts to your entity type.
Confidence tags on every finding
Evidenced (verified from public data), Inferred (logically derived) or Hypothesis (reasoned assessment). You always know what is grounded and what needs validation.
Architecture
Three AI tiers, each matched to the task
A single model used for everything means either overpaying for simple tasks or underpowering complex ones. We use three tiers of Anthropic's Claude, each selected for the cognitive demand of the task.
HIGHEST REASONING
Strategic analysis, financial modelling, quality review, initiative generation, dimension refinement
SYNTHESIS
Research synthesis, narrative generation, deliverable content, cross-sector analysis
ROUTING
Industry classification, lightweight validation, quick categorisation
The quality reviewer always operates at a tier at least as capable as the writer. Content generated by the synthesis tier is reviewed by the highest-reasoning tier. This is a deliberate design choice: a less capable reviewer cannot reliably catch the mistakes of a more capable writer.
Comparison