EV Scanning Isn't the Edge You Think It Is

Retail pricing has caught up to sharp pricing

The whole premise rests on retail books being meaningfully less efficient than sharp books. The traditional story: a book like Pinnacle builds its edge by welcoming sharp, high-volume bettors at thin margins, which forces its lines to stay accurate. A book like bet365 does the opposite, courting casual volume through marketing and pricing wider margins around it.

That gap has been closing for a while. Real-time data access means odds propagate near-instantly across books, retail lines increasingly track sharp lines with almost no lag, and the pool of people building these models professionally has grown across the industry, narrowing the skill gap in pure pricing accuracy.

I tested this directly, comparing closing 1X2 odds from bet365 and Pinnacle across 2,280 Premier League matches from the 19/20 through 24/25 seasons (data: football-data.co.uk; full workings in this spreadsheet).

	bet365	Pinnacle
Accuracy	55.83%	55.57%
Log-loss	0.9558	0.9556
Matches predicted better	1,030	1,250

Log-loss measures how well-calibrated a probability forecast is, penalising confident wrong predictions more heavily; lower is better. The two books come out essentially tied, and bet365 edges ahead on raw accuracy. Match-by-match, Pinnacle does win more often on log-loss, which points to Pinnacle pricing more aggressively while bet365 prices more conservatively, rather than either being simply “more accurate.”

Even a genuine pricing gap needs volume most bettors can’t run

Suppose the pricing gap is real. Turning it into actual profit still requires a lot of bets. The measured differences above are small on average, genuine positive-EV opportunities only appear intermittently, and a statistical edge doesn’t convert to profit deterministically; it needs a large sample size before it shows up reliably in results.

That creates a practical problem: retail books, the exact target of EV scanning, are also the books most likely to limit or restrict bettors who show a pattern of consistent, well-priced action. Retail books depend on marketing spend to bring in casual volume, which makes them quick to flag and act on accounts that look like a loss risk.

There’s a second constraint on top of that. Most retail books respond to genuine uncertainty by restricting markets rather than mispricing them. The moments where the retail-sharp gap is widest tend to be exactly the moments something unexpected has happened close to kickoff. In that window, retail books typically pull the market rather than leave it open, then reopen once the sharp-side price has moved enough to confirm the new risk level.

The real edge is the philosophy behind value betting, not EV scanning itself

Its popularity is itself a signal that it isn’t pure nonsense. But the part worth keeping isn’t the scanning method, it’s the value betting philosophy underneath it.

That philosophy is straightforward: if the true probability of an outcome sits above what the bookmaker’s price implies, the bet has positive expected value. The entire question is how you get to a better estimate of that true probability than the market’s. EV scanning’s specific failure is substituting one sharp book’s fair odds for that true probability, treating a market price as if it were ground truth.

Actually answering how to estimate that true probability better is a separate, longer conversation, one we work on directly through OddsLine’s model and AI service. But laying out where EV scanning falls short is a necessary first step before that conversation is worth having.