What Actually Runs Behind Our Daily Picks

Most days there is a new prediction on our picks page. Somebody asked me last month where those come from, and whether a human writes them or a machine does. Fair question, and the honest answer is: a bit of both, and the machine part has a name — Erudil.

I want to explain that properly rather than bury it in a footer disclaimer, because “AI picks” has become the kind of phrase that means nothing. Every tipster site on the internet now claims an algorithm. Most of them mean a spreadsheet, or a man in a rented flat who types confident sentences.

Why we bothered with an engine at all

The old version of this site linked out to a prediction tool and left it at that. Send the reader away, hope they come back. They do not come back.

The other option was to write picks ourselves. I have done that before, at a previous site, and I will tell you what happens: you produce four good analyses a week, and then you produce filler, because a publishing schedule does not care that you have nothing useful to say about a Tuesday night KHL game. Filler is where the confident nonsense creeps in, and confident nonsense about betting is how people lose money they did not plan to lose.

So we took the third route. Erudil produces the analysis, we publish it under our own rules, and neither side pretends the other does not exist.

What it actually does

It is not one model answering a question. It is a chain, and the chain matters more than the model.

First it validates that the event exists and is what you think it is — an obvious-sounding step that a surprising number of tools skip, which is how you end up with a confident prediction for a fixture that was postponed two days ago. Then two things run in parallel: one collects statistics, the other aggregates existing market data and published opinion. A final stage merges those into an output, and a last pass strips anything that reads as a promise.

That last pass is the part I care about. The output is a probability with reasoning attached, not a tip. It shows uncertainty instead of hiding it. There is a difference between “the home side has the edge here and here is why” and “lock of the day”, and the second one is always, without exception, a sales pitch.

The people behind it are explicit that the goal is reducing bad decisions rather than encouraging more betting. You can read that in how the thing behaves: ask it for a certainty and it will not give you one.

How the picks reach this page

We pull one free prediction a day from their feed. Each one is handed out to exactly one site, so what you read here is not syndicated across forty domains with different logos on it.

Then we rewrite it. The raw output is technical — readable if you already follow the sport, dry if you do not. The rewritten version goes through a banned-phrase filter before it publishes: no “guaranteed”, no “risk-free”, no percentages of your bankroll, no money amounts, no “bet now”. If the text trips any of those, it does not get published and the pick is skipped for the day. That has happened. We would rather post nothing.

Every pick carries the source credit and the same line at the bottom: opinion, not a guarantee, 21+.

What this does not mean

It does not mean the picks win. Some lose. A prediction engine that is right 55% of the time on properly priced markets is doing well, and 55% still means you are watching four in nine go against you, often in a row, often in the ways that make people chase.

It also does not mean you should bet every pick we publish. We have no idea what your bankroll is, what your staking looks like, or whether you can afford the variance. Nobody who publishes a tip for free knows that about you, and anyone who talks as though they do is selling something.

Judge the picks the way you would judge a weather forecast — over a season, on calibration, not on whether yesterday was right. That is the standard Erudil holds itself to and it is a harder one than accuracy alone.

If you want to use it directly

You can. It is their own product, not something we resell — you request an analysis on a match you care about instead of reading the one we happened to be given that day. They run a token model for access with the current rates on their pricing page, and they write up their own methodology on their blog in more depth than makes sense here.

We are not paid to say any of this. There is no affiliate link to Erudil in this article, and the ones elsewhere on this site go to sportsbooks, which we disclose on our ranking page. We use the engine because it is the only one we found that will tell you it does not know.

That, honestly, is the whole reason it is on this site.