bet365 occasionally offers a "winnings boost" on Bet Builder selections — most commonly 25%, occasionally 50%. A 50% boost on a multi-leg combination is a rare enough offer that it's worth asking a simple question before taking it at face value: does it actually overturn the book's built-in margin, or does it just make a bad bet less bad? This piece walks through a real example, checks the coverage is genuinely complete, calculates the aggregate margin (overround) with and without the boost, and lays out what can — and can't — be concluded from the arithmetic alone.
The market in question combines three independent football card-related props into a single 3-leg Bet Builder:
Each of these three markets is binary and exhaustive — for card counts specifically, a push at exactly 3 cards isn't possible since 3.5 is a half-line. That gives 2 × 2 × 2 = 8 mutually exclusive, collectively exhaustive combinations. Building all eight as separate Bet Builder selections and pulling their boosted odds gives a genuine picture of the whole outcome space — not just a favourable slice of it.
Eight bet slips were built, one per combination, and every one returned a valid non-zero price. That confirms the coverage is complete: exactly one of the eight will settle as a winner regardless of how the match plays out. This matters because the overround calculation below is only meaningful if every possible outcome is represented — leave one combination out and the "missing" probability mass distorts the sum.
| Market | Selections & Odds | Implied Probabilities | Overround |
|---|---|---|---|
| Cards O/U 3.5 | Over 11/10, Under 4/6 | 47.62% + 60.00% = 107.62% | 7.62% |
| Both Teams Carded | Yes 1/2, No 6/4 | 66.67% + 40.00% = 106.67% | 6.67% |
| Time of First Card | Before 5/6, No-Before 5/6 | 54.55% + 54.55% = 109.09% | 9.09% |
Each leg on its own carries a standard 6-9% bookmaker margin. The question is what happens once they're combined into a Bet Builder, since bet365 doesn't simply multiply the three independent prices together — it applies its own correlation adjustment.
| # | Combination | Odds | Implied Probability |
|---|---|---|---|
| 1 | Over, Yes, Before | 12/5 | 29.41% |
| 2 | Over, Yes, No-Before | 11/2 | 15.38% |
| 3 | Over, No, Before | 20/1 | 4.76% |
| 4 | Over, No, No-Before | 33/1 | 2.94% |
| 5 | Under, Yes, Before | 13/2 | 13.33% |
| 6 | Under, Yes, No-Before | 11/2 | 15.38% |
| 7 | Under, No, Before | 15/2 | 11.76% |
| 8 | Under, No, No-Before | 3/1 | 25.00% |
| Sum | 117.98% | ||
The unboosted 8-selection book carries an aggregate overround of roughly 18%. If you could stake proportionally across all eight to guarantee an equal return whatever the outcome, total stake required would be about 118% of the guaranteed payout — an unrecoverable ~15% loss on turnover. That's the baseline the boost has to beat.
It's also worth noting the combos aren't priced as a simple product of the three leg prices. Comparing actual odds to what independence would imply shows bet365 shortens the "correlated" combinations (e.g. Over cards + Both Carded + early first card, which naturally go together) and lengthens the combinations that fight each other (e.g. Over 3.5 cards but no card before 35 minutes, which requires a late flurry of cards). The 18% isn't three stacked margins multiplying up — it's the bookmaker's actual priced margin after correlation adjustment.
The boost applies to winnings, not stake (bet365's standard format):
| # | Combination | Original | Boosted | Boosted Implied Prob. |
|---|---|---|---|---|
| 1 | Over, Yes, Before | 12/5 | 18/5 | 21.74% |
| 2 | Over, Yes, No-Before | 11/2 | 33/4 | 10.81% |
| 3 | Over, No, Before | 20/1 | 30/1 | 3.23% |
| 4 | Over, No, No-Before | 33/1 | 49.5/1 | 1.98% |
| 5 | Under, Yes, Before | 13/2 | 39/4 | 9.30% |
| 6 | Under, Yes, No-Before | 11/2 | 33/4 | 10.81% |
| 7 | Under, No, Before | 15/2 | 45/4 | 8.16% |
| 8 | Under, No, No-Before | 3/1 | 9/2 | 18.18% |
| Sum | 84.21% | |||
Aggregate implied probability drops to ~84.2% — an underround of about 15.8 percentage points. That's not a smaller margin, it's a negative one: theoretically enough slack that somewhere in the eight outcomes, the true probability must exceed the boosted implied probability.
| # | Combination | Original | Boosted | Boosted Implied Prob. |
|---|---|---|---|---|
| 1 | Over, Yes, Before | 12/5 | 3/1 | 25.00% |
| 2 | Over, Yes, No-Before | 11/2 | 55/8 | 12.70% |
| 3 | Over, No, Before | 20/1 | 25/1 | 3.85% |
| 4 | Over, No, No-Before | 33/1 | 41.25/1 | 2.37% |
| 5 | Under, Yes, Before | 13/2 | 65/8 | 10.96% |
| 6 | Under, Yes, No-Before | 11/2 | 55/8 | 12.70% |
| 7 | Under, No, Before | 15/2 | 75/8 | 9.64% |
| 8 | Under, No, No-Before | 3/1 | 15/4 | 21.05% |
| Sum | 98.26% | |||
Still technically underround — but only by about 1.74 percentage points in aggregate, roughly a tenth of the buffer the 50% offer produces. That's the difference between a book that's clearly beatable and one that's marginal-to-fair once estimation error is accounted for.
Whether the boosted underround translates into a specific, exploitable edge on any one leg is a separate question from whether one must exist. That second part can be shown directly, without needing to estimate any true probabilities.
Let p₁...p₈ be the true probabilities of the eight outcomes and q₁...q₈ be the
boosted implied probabilities. Two things are certain:
Σp = 1 — true probabilities of exhaustive, mutually exclusive outcomes must sum to exactly 1Σq = 0.8421 (at 50% boost) or 0.9826 (at 25%) — calculated directly from the boosted odds
Suppose, for contradiction, that every outcome had pᵢ ≤ qᵢ. Summing across all
eight would give Σp ≤ Σq, i.e. 1 ≤ 0.8421 (or 0.9826) — which is false. So at least
one outcome must have pᵢ > qᵢ strictly, meaning pᵢ × Oᵢ > 1: genuine positive
expected value on that leg in isolation.
This is airtight at both boost levels: a positive-EV leg exists among the eight, full stop, regardless of what the true card-market probabilities actually are. What it does not tell you is which leg, or by how much. The proof is non-constructive. At 50% boost, the ~15.8-point buffer makes that uncertainty easy to tolerate. At 25%, the ~1.74-point buffer is thin enough that ordinary estimation error could plausibly erase it on any specific leg you pick.
Favourite-longshot (F/L) bias is a well-documented pattern across bookmaker markets — most rigorously in horse racing, but observed elsewhere too — where margin isn't spread evenly across selections. Favourites tend to be priced close to fair value; longshots carry disproportionately inflated implied probability relative to their true chance.
Applied here, that predicts the slack is more likely concentrated in the shortest-priced leg (Over, Yes, Before at 12/5) than in the longest (Over, No, No-Before at 33/1), because the shorter price started closer to fair before any boost was applied. It's a genuine reason to lean toward the favourite leg — but it's a heuristic, not a guarantee, and its magnitude hasn't been measured for football card markets specifically. It's well-quantified for horse racing; for in-play football booking markets it's a plausible extrapolation, not a validated number.
A separate dataset — 13,631 matches across 13 European leagues from football.data.co.uk — showed that blindly backing Over 2.5 goals at bet365 loses -3.15% ROI on average, and that the bias isn't uniform: 10 of 13 leagues show the goals market underpricing the over (actual over-rate higher than implied), while 2 leagues (Serie A, Belgium) show the opposite. The driver appears to be the market anchoring on a generic ~52-55% over-rate prior and under-adjusting for league-specific scoring cultures (Bundesliga at 3.18 goals/game vs Segunda at 2.34).
That -3.15% figure is a real, well-sampled result — but it's measured on goals markets, not cards markets. No equivalent dataset exists for cards at the level of granularity needed (football.data.co.uk's free tier doesn't carry booking-points/cards fields), and the mechanism behind the goals bias — stable, well-documented league-level scoring cultures — has no established equivalent for cards, which depend far more on referee strictness, match tension and in-play flashpoints than on a fixed league "carding culture." Applying the -3.15% figure, or the asymmetric over/under split it implies, to a cards market would be borrowing a number from the wrong market and presenting it as measured when it's actually an untested assumption.
In the absence of cards-specific data, the honest fallback is a uniform-split assumption across the Cards O/U 3.5 leg (each side carrying roughly half of the 7.62% market margin) — which gives no directional lean between the Over-based and Under-based halves of the eight combinations. Closing that gap would need match-level cards data with attached bookmaker O/U prices — a genuinely separate data-sourcing project from the one behind the goals analysis, not a re-run of the same query on a different column.
A rare 50% boost on a fully-covered, correctly-priced 8-leg Bet Builder is the case where the arithmetic alone gives real confidence. A routine 25% boost is a much closer call, and treating it the same way risks mistaking "provably exists somewhere" for "safely bettable here" — the gap between those two is exactly the F/L-bias magnitude nobody has measured yet for this specific market type.