About and method

What this site is, and what it is not

Most writing about AI implementation quotes one alarming statistic and then sells you something. This site does the opposite. It quotes the research, shows you the sample size and the method, links the original, and sells you nothing.

Who writes this

Sunny Patel is a UK SEO consultant and independent developer who builds and ships AI-assisted software. He does not sell AI transformation projects and has never run one. That is deliberate, and it is the point: this site has no consultancy to steer you toward and no tool it needs you to buy, so the evidence can point wherever it points.

His role here is narrow and checkable. He reads the primary research behind the headline failure statistics, quotes it with its methodology and sample size, and builds the free diagnostics on this site. Where a figure is contested, he shows the range and names each survey.

What this site may never claim

These are hard rules, not aspirations. If you find the site breaking one, it is a bug.

• It has run no surveys and owns no proprietary dataset.
• It has tested no tools and delivered no client AI projects.
• It never invents a sample size. When a study does not publish one, the citation says not stated in accessible sources, which is the honest answer.
• It never repeats a vendor's cost estimate as though it were research. Agency blog figures such as the widely quoted ten-to-fifty-thousand ranges are marketing, and are labelled as such.
• The risk diagnostic is not a validated statistical model and cannot predict whether a project will fail.

How a statistic gets onto this site

1. Find the primary. Not a blog quoting a blog. The report, the press release, the national statistics bulletin.
2. Read what it actually says. Numbers get more precise and more dramatic as they travel. A source saying "over 80%" becomes "80.3%" three articles later. We quote the source's own wording.
3. Record the scaffolding. Sample size, method, fieldwork window, publication date. A figure without these is an anecdote wearing a suit.
4. Note whether it is a measurement or a prediction. Several of the most-quoted AI failure figures are analyst forecasts about the future, not observations of the past. We mark those.
5. Record whether you can actually read it. Several primary sources return an error to ordinary readers. Where that happens we say so and name the credible secondary we read instead.

The citation chip

Every statistic on this site carries a chip in the margin. Its dot tells you how much to trust the chain of custody:

• Green. We fetched and read the primary source.
• Amber. The figure is contested, ranged, or a prediction rather than a measurement. Read the range, not the headline.
• Oxblood. The primary source blocks ordinary readers. The figure comes from a named credible secondary, and we link both.

The current corpus

This site cites 15 sources and no others. Of those, 9 were read directly, 4 are contested or predicted, and 2 block ordinary readers and are quoted via a named secondary. 3 do not publish a sample size we can obtain, and we say so rather than guess.

Adding a source means verifying it first. If a page you want does not exist, that is usually because the research to write it honestly does not exist either.

How this site pays for itself

Some links to software go to affiliate programmes, which means this site may earn a commission if you subscribe. Those links appear only inside advice we would give anyway, never as a list of tools to browse. Nothing on this site is gated behind an email address, there is no sales call, and no vendor has paid for a mention or seen a page before it went live.

Corrections

If a figure here misrepresents its source, that is the most serious error this site can make. Say so and it gets fixed, with the change noted. Every page shows the date its figures were last checked.