Within twelve months, anyone will be able to generate slick research decks with AI. But not everyone can guarantee, on a 20-million-euro decision, that the numbers are right, every assumption is on the table, and a human signed off on every claim. That's exactly what we build.
Are the numbers right? Are the sources current, independent, and credible? We validate at source level, not at claim level.
What assumptions sit behind each statement? What is fact, what is estimate, what is judgement? Every assumption sits explicitly in the Assumption Register, not between the lines.
Every statement in the report links to an Evidence Card with source, confidence level, methodology, and human sign-off. Down to the original sentence in the PDF.
If a regulation shifts six weeks after delivery, the report updates. Versioning and a change log are part of the product.
Every 3ya engagement delivers not just a report, but the full audit trail behind it. That's what sets these decks apart from ordinary AI output.
The standardised IC format: Executive Summary, market overview, competitive landscape, regulatory context, scenarios, clear investment implication.
Every key statement as a traceable card: claim, sources, confidence level, reviewer, last check. Quotable straight into the IC memo.
Every assumption listed explicitly, ranked by sensitivity × impact. With risk, monitoring trigger, and basis. When an assumption breaks, the IC knows immediately whether the thesis still holds.
| Assumption | Sens. | Impact |
|---|---|---|
| BfArM reimbursement Q3 | high | high |
| Reimbursement drift | med | high |
| Staff costs +4% | med | med |
| M&A multiples stable | low | med |
AI makes each stage 4–6× faster. But every stage ends with a human sign-off. No output leaves the process without a named reviewer signing it.
15–25 independent sources, < 18 months old, at least one contrarian perspective. Coverage map exposes blind spots.
Extraction, dedup, tagging, separation of fact / estimate / opinion. ≥ 80% confirmed relevant by reviewer.
Claims drawn from signal clusters. Confidence levels (★1–5). Assumptions made explicit. Causation vs. correlation checked.
Standard IC template populated. Charts, market maps, executive summary. Narrative consistency checked.
Four independent levels: Facts & Sources · Logic & Counterpoints · Numbers & Models · IC-Readiness.
Versioned asset, change log, continuous monitoring. New evidence triggers an update proposal.
Before each claim, the IC sees how solid it is. Nothing gets inflated into certainty when it's actually an assumption. Assumptions go into the register, not into the main text.
Not for marketing slides, not for "strategy refreshes". For engagements where quarterly results, board decisions, or capital allocation hinge on a single analysis.
Over two decades of healthcare and financial-investor mandates. Formerly BCG. Has spent his career on decks that had to hold up in Investment Committees at KKR, Equistone, and in due diligence processes for Bain & Company. He knows every place where weak research breaks down.
3ya was born from a single realisation: the next wave of AI-generated research doesn't solve the trust problem, it sharpens it. The answer isn't less AI, it's visibly better quality assurance.
We'll send you a redacted sample asset including Evidence Cards and Assumption Register. You see the methodology before we discuss your engagement.
Response within one business day · NDA on request