Backing the Least Favourite in Betfair Football Match Odds: A New Live Experiment

At 5:30pm on 3 October 2026, I switched on another automated football research collector. I gave it what seemed like a perfectly sensible working name: Football Underdog.

A couple of days and 130 bets later, I've realised the name is wrong.

When we talk about a football underdog, we normally mean the weaker team. Cádiz play Leganés, for example, and the market makes Cádiz the least-fancied of the three Match Odds outcomes. Our collector backs Cádiz. They don't just sneak it either. Cádiz win 4–0. Four goals, clean sheet, job done — and our average matched price was 3.15.

That's exactly the sort of result I imagined this collector finding.

Except that's not really what I've built.

The underdog isn't always a team

A Betfair Match Odds market has three outcomes: Home, Draw and Away. Our collector follows whichever of those three outcomes is least fancied by the market when the bet is captured.

Sometimes that's an unfancied home team. Sometimes it's an away team. But here's the bit I hadn't really appreciated until the data started arriving: the least-fancied outcome can simply be the draw.

And in this opening sample, it usually is. Of the first 130 live bets, 97 have been Draw selections, 18 Home and just 15 Away. So almost three quarters of this opening dataset consists of the market effectively saying: of all three possible outcomes, we think the draw is the least likely.

Then football does what football does.

Take CD Fas v CD Inca. The draw was captured at average matched odds of 11.50.

It landed.

Lysekloster v Jerv: Draw at 9.60.

That landed too.

We've already seen winning draws at 6.60 in Tolima v Boyaca Chico and CD Castellon v AD Ceuta. Morocco v Mali at 5.30 finished level. So did Trinidad & Tobago v Curaçao, again with our draw captured at 5.30.

These aren't matches I've searched for afterwards because they make a good story. They're sitting in the live collector. The machine didn't know that 11.50 looked exciting. It didn't care that Cádiz were about to score four. It simply applied the same research question and recorded what happened.

That's precisely what I want.

Then there are the actual underdogs

There are plenty of those too.

Granada, away to Sociedad B, was captured at 3.10 and won. Belgrano, away to Talleres, was the least-fancied Match Odds outcome at 3.00 and won. London City Lionesses, away to Tottenham Women, won after being captured at 2.76. Toluca Women produced an even bigger away result at 4.40 against Pachuca Women.

We've also had the opposite configuration. Cádiz at 3.15 were the unfancied outcome despite playing at home — and then demolished Leganés 4–0.

That's what makes this dataset more interesting to me than a conventional “back the underdog” experiment. We're not actually defining what an underdog should look like. The market is doing that for us. Home, away or draw: whichever outcome occupies the bottom of the three-way market is what gets recorded.

The working name will eventually change. For now I'm leaving the collector alone. The data matters more than tidying up its label.

130 bets and a frankly ridiculous start

This experiment also started at an interesting time: 5:30pm on Saturday 3 October, during an international-break period. That means its first sample is already a slightly mad mixture of domestic leagues, women's football and internationals rather than something I'd pretend represents a normal football week.

And the headline numbers are equally mad.

After 130 bets, we've recorded 42 winners, a 32.31% strike rate and average matched odds of approximately 4.43. At level one-unit stakes, allowing for 2% commission on winning profit, the opening result is approximately +45.08 units.

That deserves repeating for an entirely different reason:

+45.08 units after only 130 bets tells us almost nothing.

At these prices, variance can be brutal in both directions. A draw landing at 11.50 contributes more than ten units of gross profit by itself. Throw in another at 9.60, a couple around 6.60 and some winning outsiders and suddenly a tiny dataset can look like we've discovered the bloody Philosopher's Stone of football betting.

We haven't.

Interestingly, the early profit isn't evenly distributed either. The 97 Draw selections have produced about +40.67 units after commission. The 18 Home selections are virtually flat at around −0.22 units, while the 15 Away selections are approximately +4.63 units.

That's fascinating.

It's also far too early to do anything with it.

The worst thing I could do now would be to announce that draws are “the edge”, modify the collector to concentrate on them, find a convenient price range that makes the historical graph prettier and start optimising 130 bets.

That's how research turns into curve fitting.

The interesting number isn't +45.08

The interesting number is 130.

This collector is only a couple of days old. I want to see what this market looks like after 1,000 observations. Then 5,000. Potentially 10,000.

What proportion of the least-fancied outcomes will ultimately be draws? Does the strange Home/Draw/Away distribution we've seen during this international break persist during normal domestic football? What happens to the apparent profitability as the sample expands? What do the price distributions look like? How violent are the losing runs when those bigger-priced selections stop landing?

And perhaps most interestingly: what will we think the first 130 bets were telling us when we're eventually looking back from bet number 10,000?

That's the experiment.

I'm deliberately not publishing every collection and execution parameter behind the dataset. The objective of the Research Lab isn't to distribute ready-made bot configurations. It's to build prospective live-market datasets under defined and consistent research conditions, preserve them and see what the evidence eventually says.

This particular dataset started with a slightly misleading name, a Cádiz 4–0 demolition, some unlikely draws at 11.50 and 9.60, and an absurd-looking +45.08 units from its first 130 bets.

Great story.

Now comes the important bit: leave the collector alone and let it collect.

Research Lab note: This is a live research project, not a betting system or betting recommendation. Early profit or loss is not treated as evidence of a sustainable edge. The purpose of the collector is to accumulate a sufficiently large body of live-market observations for later analysis.