Skip the demo. Try it free.

If you sell a product people need to see before they believe it, you probably know this. A visitor fills in a contact form. Someone from the team gets back to them, books a call and shows a demo. Every lead costs a meeting before anyone knows whether it is worth having.

This company sells analysis built on large datasets. Its website was one page on a stock theme: a long headline, a line saying the platform was available by invite only, and a form. After that, the team would be in touch. The copyright in the footer was two years out of date. There was nothing to try and no example of the output.

Behind the site, each study was a run of ten numbered scripts, started by hand in the right order. The result was a long written report, so there was no small piece of it to show anyone.

Here is what a curious visitor went through, before and after.

We left the analysis itself alone. On top of it we built a free version of the product: the front door and the delivery layer, so a stranger can feel what it does without a demo or a person in the loop.

• A language model reads the question and matches it to a topic, but it may only choose from a strict list of what the dataset actually covers. It cannot invent a match. If the model is unavailable, keyword matching takes over and the visitor never sees an error.
• The site shows what it found before it asks for anything: the match, and how much real data sits behind it.
• Below a set minimum, it says there isn’t enough data and offers the nearest match that has enough.
• Only then does it ask for a first name and a work email. One report per address. The private link appears on screen and goes out by email, so nothing depends on an inbox.
• The report shows the real signal: how much evidence there is, the main finding and what it means.

That refusal is the best part of the product. Any tool can draw a confident chart. One that tells you, in public, when your question is outside what it can answer is making a much stronger claim.

The owner’s side is a small internal tool inside the CMS.

• The dataset comes in as one CSV through an admin screen, with a template to fill in. Everything the public site shows comes from that import. Nothing is typed twice.
• The written parts of each report are drafted by a model and saved as drafts. Someone approves them, one by one or in bulk, before a visitor sees them.
• Every report issued and whether it was opened, every question anyone typed, and coverage per topic, on one screen.
• Every piece of site copy is editable, with the original shown underneath. Clear a field and the original comes back.
• The handbook lives inside the admin rather than in a file, where it cannot drift away from the product.

The site hands out private links, so this got real attention.

• Links are unguessable, expire after a set time and are kept out of search engines, caches and referrers. An expired link gets a page of its own.
• One report per email address. Plus-tags and dotted variants count as the same address, and throwaway domains are refused.
• Hourly limits per visitor, a bot challenge on both public forms, and hidden traps for bots that screen readers skip.
• Limits are only as good as the address they count, so the site trusts forwarded addresses only from its known network edge, and a health check shows which address it actually sees.
• Analytics load only after consent. Author pages, the public users list and comments are closed, and the administrator’s login name no longer leaks through the site’s embed data.

It is early, and we don’t publish numbers we don’t have yet. What is measured today:

• PageSpeed: 98 on desktop and 97 on mobile, 98 for accessibility, 100 for best practices.
• Zero blocking time and almost no layout shift.
• Zero PHP notices across every public page and admin screen.
• One CSV import puts a dataset live on the site. No script run needed.

Free reports issued, how many get opened and how many turn into a request for the full version come after the first month. The database already records all three.

First, those numbers. Then more coverage, because every topic added makes the honest “not enough data” answer a little rarer. And a faster way to approve drafts, for when there are many more of them.