How validation scores are calculated
How scoring works
Below is the formula, the inputs each check reads, and how confident you should be in each part of a report.
Validate an idea free
What is a validation report?
StartupDeckAI turns a short idea description into a structured memo: validation score, competitor landscape notes, market sizing sketches (TAM / SAM / SOM), revenue scenarios, and risk prompts.
The output is a draft for founder diligence, not investment advice, legal guidance, or a guarantee of market success. Each layer has different reliability, combining rules-based scoring with LLM synthesis.
The Scoring Engine
Every score comes from a fixed formula in code. The weights are the same for every idea and are not adjusted to flatter one. What goes into the formula depends on which check you run.
Quick check
No AI model is involved. Rules read your industry, buyer and description, and score nine dimensions with these weights.
- Market potential 18%
- Demand signal 16%
- Monetization fit 14%
- Defensibility (moat) 12%
- Buyer clarity 10%
- Pricing power 10%
- Competition (inverse) 8%
- Distribution (inverse) 6%
- Execution (inverse) 6%
Full report
An AI model reads your idea in five passes, and a fixed formula scores eight dimensions from what it returns. The formula is ours; most of what it is given is the model’s judgement. Every figure in the report carries a label saying where it came from.
The model's reading of your idea
Problem severity, distribution, timing
The model classifies facts from your description (how urgent the problem is, how hard the sale is). A fixed formula turns those into scores.
The model's estimates
Market size, business model
Market size, customer acquisition cost, lifetime value and payback are the model’s figures. When it gives none, a formula or an industry benchmark fills in, and the report labels which.
The model's own 0–100 scores
Founder advantage, defensibility, technical execution
Used as the model gives them. Treat these three as an informed opinion, not a measurement.
Penalties reduce the final score when the engine finds inconsistent inputs, missing figures, or unit economics that rest on industry benchmarks instead of your numbers.
Structured LLM pass
Extracts likely direct and indirect competitors from your idea text and industry.
Industry templates
When evidence is thin, known players for the selected sector may seed the list.
Web research
The places listed under "Where a report looks" are searched before the model is asked. A competitor is marked web-verified only when its name appears on a page that search found.
Similarity ranking
Embedding-style matching ranks competitors by semantic overlap with your idea.
Competitor Identification
Competitor names come from a multi-step research pipeline, not a single chat completion. We gather, check and rank the candidates.
You should verify every competitor name, website, and positioning statement before citing them to investors. We surface candidates, not certified market share data.
Where a report looks
Before the model is asked about your idea, the report searches these places. The list is printed from the code that runs the search, so a place that is not searched cannot appear here.
The open web
Found with a web search.
- What it looks for
- Companies already doing what you describe, and pages that compare them.
- What it does not get
- A search returns a handful of pages, not every company. A competitor it misses is still a competitor.
Product Hunt
Found with a web search limited to this site.
- What it looks for
- Products launched in the same space.
- What it does not get
- The launch page and its description. Not vote counts or rankings.
Crunchbase
Found with a web search limited to this site.
- What it looks for
- Company profiles in the same space.
- What it does not get
- The public profile page a search finds. Not funding rounds, revenue or investor data, which sit behind a Crunchbase licence.
GitHub
Read through its public API.
- What it looks for
- Open source projects that already solve the problem.
- What it does not get
- Public repositories only. A star count shows interest from developers, not paying customers.
Reddit
Found with a web search limited to this site.
- What it looks for
- Threads where people describe the problem or the tools they use for it.
- What it does not get
- The thread title, link and an excerpt. Not scores or comment counts, and not private or removed posts.
Hacker News
Read through its public API.
- What it looks for
- Stories and launches on the same subject, with their points and comment counts.
- What it does not get
- Story titles only, not the comments under them. The audience is mostly people who build software.
X
Found with a web search limited to this site.
- What it looks for
- Public posts on the subject.
- What it does not get
- Only the few posts a web search has indexed, so this often finds nothing. Not likes, reposts or reply counts.
Every page found is listed in the report's Sources section with a link, under the place it came from. A place that returns nothing adds nothing: the report does not fill the gap with a figure. The model reads the company and product pages when it lists competitors. It does not read the threads and stories, which are there for you.
Confidence Levels
How far to trust each part of a report. While market and financial figures are model estimates, a full report is rated Medium at best.
High Confidence
Reserved for figures that come from outside the model: your own numbers, or sources we have checked. No report section is rated High today.
Medium Confidence
The analysis behind it ran and returned what was asked. The content is still the model’s reading of your idea, so check it before you rely on it.
Low Confidence
A pass failed or came back incomplete and a fallback filled in, or the inputs were too thin. Treat it as a prompt to investigate.
Human Responsibility
Founders remain responsible for customer discovery, legal compliance, financial decisions, and investor communications. StartupDeckAI does not replace accountants, lawyers, or licensed financial advisers.
Start Validating
Explore validation tools
Tools & workflows
Free idea validator Validation tool overview Startup metrics dashboard AI chief of staff MVP roadmap generator Business model generator
Industry validators
ai ml validator marketplace validator proptech validator agtech validator
Comparisons
All tool comparisons Compare vs founderpal Compare vs pitchbob Compare vs hyper Compare vs cerenovus
Guides
ai vs traditional startup validation create investor pitch deck 2026 understanding tam sam som startup validation india guide
Trust & docs