How we use AI
AI writes one short note on some startup pages and powers nothing else on the site. This page says what it is given, what it must not do and where it is not used.
- Only verified figures. The weekly outlook is written from verified figures and the statistical projection, and nothing else.
- Checked after writing. Any number that is not in those figures causes the note to be rejected.
- Bands stay private. A startup that shows only a band is never sent to a model.
- Not used to decide anything. Verification and ranking involve no AI.
Given to the model
- Startup name and category
- Verified tier and rank
- Monthly revenue in US dollars
- 30 and 90 day growth
- The projection, low, base and high
- Milestones reached
Never given to the model
- Email addresses
- Provider keys and provider names
- Founder names
- Customer names or payment details
- Anything about a band-only startup
On this page
The weekly outlook
Startup pages with at least 30 days of verified history can carry a short outlook below the projection: a headline, a few sentences, the drivers that could push the result up, the risks that could push it down, and a confidence level. It is refreshed weekly, and skipped when a startup's figures have not changed.
The projection itself is statistical. It comes from an ensemble of trend and smoothing models checked against the startup's own history, and gives a low, a base and a high case. The model's only job is to explain that projection in plain words. It does not make a forecast of its own.
What the model is given is a short list of public, verified facts: the startup's name and category, its tier and rank, monthly revenue in US dollars, 30 and 90 day growth, the projection and the milestones reached. It is not given emails, keys, founder names, provider names or anything a founder keeps private.
Its answer is checked before it is saved. Any number that does not appear in those facts is rejected, and so is an answer that is not in the expected shape. A note that fails is not shown, and the page simply has no outlook that week.
Every outlook is labelled as AI-generated and as a projection, never a promise, and shows when it was written.
The Ask AI buttons
Pages have an Ask AI button that opens ChatGPT, Claude, Perplexity or Grok with a question already written, or copies it for Gemini. The question holds the address of the page and the public facts the page already shows, such as the figure a startup displays, or only its band when it shows a band.
MRRSignal sends nothing to those services itself. Your browser opens the assistant you pick, and what you do there is between you and that service. Nothing from your account goes into the question.
What AI is not used for
AI plays no part in the things that make a figure trustworthy.
- Verification. A figure is read from the payment provider by code, then converted and summed by code.
- Ranking. The leaderboard orders startups by verified revenue and nothing else. Sponsor cards, boosts and AI never change a rank.
- Decisions about people. No account is suspended, listed or removed by a model.
- Inventing figures. A model is not allowed to calculate, round or add a number, and the check enforces it.
Your controls
A founder can turn the outlook off for their startup from the startup editor. When it is off none is shown and none is generated, so no model is called for that startup.
The whole feature can be switched off by an administrator without a deploy, and it stops by itself when the monthly spend limit is reached.
Questions about any of this? Write to [email protected].