You type in a business idea. Fifteen minutes later, you get a score, a verdict, and a report that separates what's real from what isn't.
The question nobody asks about AI analysis tools: how does the machine actually decide?
Most tools won't tell you. They show a score and ask you to trust it. Here's what happens inside Cortex AIF — step by step, no black boxes.
Stage 1: Understanding Your Input
Submit a URL — the system scrapes the website and extracts structured data: company name, pricing, team, services, testimonials, social media links.
Submit text — the system classifies the project type, geography, audience, and monetization model.
Two context questions shape the analysis: Is this your business or someone else's? What stage — idea, MVP, or paying customers?
A ten-year-old agency needs different analysis than a napkin-sketch startup. Same pipeline, different lens.
Stage 2: Research
The system generates 40 to 55 search queries specific to your business. These run through 242 search engines plus deep web scraping tools — collecting market data, competitor pricing, legal requirements, financial benchmarks, and public records about the company and its founders.
If data isn't found on the web right now, it's marked as unverified. Not invented. Not estimated from the model's training data. Marked honestly and moved on.
Stage 3: Verification
This is where Cortex AIF breaks from every other tool in this category.
Six layers, every report, every tier:
Domain. WHOIS for age and registrar. Wayback Machine for the earliest archived version. A domain from 2014 tells a different story than one from last Tuesday.
Business registry. Companies House, KRS, OpenCorporates. Legally registered? When? Who are the directors? Filing status?
Founder. Social profiles found on the website are scraped directly. LinkedIn connections, Instagram followers, activity — real numbers from real pages, not guesses.
Reviews. Trustpilot, Google Reviews, Clutch. External sources only. Reviews on the company's own website don't count as independent.
Claims crosscheck. The website says "100+ clients." Portfolio shows 12 case studies. LinkedIn has 3 recommendations. Adjusted range: 15 to 40 actual clients. Not a guess — arithmetic from evidence.
Synthesis. All verification data becomes structured context for every analysis block. Status assignment is done by code, not by the AI model. This distinction matters.
Stage 4: Scoring and Verdict
Sixteen analysis blocks, each producing a score from 1 to 10. Market, finance, risk, legal, team, technology, marketing, operations, channels, execution, and more.
Scores are combined by a weighted formula. Market and financial blocks carry more weight than legal or technology for most business types. For established businesses over five years old, certain weights adjust — because surviving a decade is evidence that deserves recognition.
The final score produces the verdict:
Startups: GO, CONDITIONAL GO, or NO_GO.
Existing businesses: GROWING, STABLE, or AT_RISK.
Run the same input twice. Get the same score. This is arithmetic, not creative writing.
The Marking System
Every data point carries a tag. No exceptions.
No single signal is a verdict. A young domain doesn't mean a scam. Missing LinkedIn doesn't mean a bad business. The system builds a mosaic, not a prosecution.
What the System Does Not Do
Does not guarantee third-party accuracy. If Companies House has incorrect data, the report reflects it with a verified tag. The tag means "independently sourced" — not "absolutely true."
Does not replace professional advice. The verdict is informational. One input in your decision, not the decision.
Does not predict the future. Markets shift. Competitors emerge. The report captures evidence available right now.
Does not fabricate. If a profile can't be scraped: "could not retrieve data." Never an invented number.
Does not flatter. Evidence points to NO_GO — that's the verdict. The Killer Question hits the weakest point in your plan. Not to discourage. To make sure you've thought about it before the market does it for you.
The Difference
Every AI tool gives you a confident wall of text. It reads like someone who knows your industry wrote it. Nobody did. A model that's good at sounding knowledgeable wrote it.
Cortex AIF makes a different trade: less polish, more proof. Every claim tagged. Every source traceable. Every score calculated. You can disagree with the verdict — but you can see exactly how it got there.