Our Approach
A hybrid analytical system
WeBetSmart relies on a hybrid analytical system that pairs classical statistical methods with modern AI. We don't invent data — every figure on this page has a clear source and a transparent formula.
The goal isn't just to hand you a prediction — it's to show how and why we reached it. Every number in our analysis can be checked and reproduced.
Market Data — Prices from every bookmaker we track, including ones we don’t list, with each bookmaker’s margin removed
Fair Price — The median of those prices is the fair price; the probability of an outcome is the fair price turned back into a percentage
Value — The best price at a bookmaker available to you, against the fair price — no bet unless the price is 5–30% better
Statistics — Expected goals (xG) feed a Poisson model that we show for reference; it doesn’t change the probability or the pick
AI Analysis — Reads the data and writes the analysis; the pick itself is chosen by code, not by the AI
Don't take our word for it
Every qualifying pick is recorded before kick-off and settled automatically against the final result. Wins and losses are published together — including the losing markets.
See the track record4-Stage Analytical Pipeline
From raw data collection to final recommendation
Data Collection
DeterministicWe collect every objective statistical input from verified sources. Nothing is fabricated — just facts and figures.
Team statistical profiles
- Formations
- Discipline (cards)
- Goal patterns
- Form & results
xG metrics
- Average team xG
- Recent-match trends
- Sample size per team
- Goals-based fallback where xG is missing
Lineups & injuries
- Key players
- Injuries
- Suspensions
- Lineup impact index
Market data
- Best odds
- Line movement history
- Implied probabilities
- Bookmaker margin
AI Analysis
Multi-model AIAn AI model reads the collected data and writes expert conclusions. It shapes the written analysis; it does not set the probability or choose the pick.
Probability & Verdict
DeterministicCode, not the AI, produces every number: the fair price from the market median, checked against the best current price at a bookmaker available to you.
Written Analysis
Quality controlThe AI writes the article from the verdict it has been given. If the prose and the numbers ever disagree, the AI Verdict block — computed by code — is the one that counts.
How We Estimate Probability
The exact formulas behind every percentage in AI Verdict
Margin-free prices
Margin removal (power method): find k so that Σ (1 / price)^k = 1 — fair probability = (1 / price)^kEach bookmaker's prices add up to more than 100% — the difference is its margin. We remove it from every bookmaker separately, so the outcome probabilities of that bookmaker add up to 100%.
Fair price = the median
Fair probability = margin-free median price of all tracked bookmakers; fair price = 1 ÷ fair probabilityWe take the median across at least 4 bookmakers. The median includes bookmakers we don't list on the site; a bet is only offered at bookmakers available to you.
Worked example
Illustrative example — not a real match · Made-up prices, shown to explain the arithmeticMedian prices (median of 7 bookmakers), match winner. Margin removal: 1/1.88 + 1/3.60 + 1/4.30 = 104.2% → margin removed → 51.8% + 26.3% + 21.9% = 100%.
| Outcome | Median price | Fair prob. | Fair price | Best price | EV | Verdict |
|---|---|---|---|---|---|---|
| Home win | 1.88 | 51.8% | 1.93 | 2.05 | +6.2% | Value |
| Draw | 3.60 | 26.3% | 3.80 | 3.75 | -1.3% | No value |
| Away win | 4.30 | 21.9% | 4.57 | 4.40 | -3.7% | No value |
Home win: Fair probability 51.8% is above 50% and the best price 2.05 beats the fair price 1.93 by 6.2%: inside the +5% to +30% window, so it qualifies as an official pick.
Draw: The best price 3.75 is below the fair price 3.80, so there is no edge.
Away win: The best price 4.40 is below the fair price 4.57, so there is no edge.
xG model — for reference only
Next to the fair price we show a Poisson model on expected goals. It does not change the probability, the verdict or the pick. Goals are rare, roughly independent events, so each team's goals are modelled as a Poisson count whose average is that team's expected goals (xG).
P(h:a) = Poisson(h | xG home) × Poisson(a | xG away), h, a = 0…10Home win adds up every scoreline with h > a, away win every scoreline with a > h, the draw is what remains.
P(Over 2.5) = 1 − P(total ≤ 2), total ~ Poisson(xG home + xG away)Treats the match total as one Poisson count with the combined expected goals.
P(BTTS Yes) = (1 − e^(−xG home)) × (1 − e^(−xG away))The chance that each side scores at least once, assuming the two scoring processes are independent.
- The Poisson model is shown for reference as “xG model”. It does not feed the probability, the verdict or the pick. Our backtest on 2,758 matches found that blending it with the market made the estimate worse than the market alone.
- xG input: each team’s average expected goals per match over its recent finished matches (API-Football data). Where a league has no xG, actual goals stand in and the input is marked as less reliable.
- The fair price needs prices from at least 4 bookmakers. With fewer, the probability falls back to the best prices we have with the margin removed, and the analysis says so.
- No match-winner prices yet: no analysis is published until the market opens.
We also score how reliable the xG inputs are (sample size, source, stability). That score is a diagnostic only: it does not change the probability, the verdict or the pick.
A verdict for every market
EV% = (fair probability × best price − 1) × 100Fair price = 1 ÷ fair probabilityValue from = (1 + 5%) ÷ fair probabilityChecks run in this order; the first one that applies is the verdict you see.
- 1No dataNo fair price or no bookmaker price for this market. We don’t calculate EV from a price that doesn’t exist.
- 2No valueThe best price is less than 5% above the fair price. The most common verdict — most prices are fair.
- 3Edge too large — likely model errorThe best price is more than 30% above the fair price. Almost always a stale line, a data error or news the market has already priced. Flagged — “price too good, double-check” — never recommended.
- 4ValueThe best price is 5–30% above the fair price. The strongest such market becomes Our Pick; it is official when the fair probability is also at least 50%.
Official picks
Only official picks count in the track record. A pick is official when all of these hold at publication:
- It is the primary pick of the analysis
- The fair probability is at least 50%
- The best price at an available bookmaker is 5–30% above the fair price (EV)
- Market: match winner, Over/Under 2.5 or both teams to score
- The price is real and currently available — never a placeholder
- No model confirmation and no confidence level are required
Before 2 October 2026 the probability was a blend of 53% market and 47% Poisson model, and an official pick required High confidence, a probability of at least 50% and EV of 0% or more. Those picks stay in the track record under that rule; results are kept for history and do not describe the current method.
Value Betting Engine
Our in-house engine for finding +EV bets
What is the Value Betting Engine?
Automatic detection of market disagreementsThe Value Betting Engine is our in-house analytical engine. It continuously scans the odds of 7+ bookmakers and compares the best price at a bookmaker available to you with the fair price (see above). When that price is higher than the fair one — that is a value bet, i.e. a bet with positive expected value.
Value% = (fair probability × best price − 1) × 100Value% > 0 means the bet is +EV. The rule: fair probability ≥ 50% and Value% between 5% and 30%.
A price more than 30% above the fair price almost always means a stale line, missing news or a data error — not that the bookmaker mispriced the match by that much. We flag it as «price too good, double-check» rather than recommend it; see the signal rules on the predictions terminal.
Transparency of results and system status
The numbers as they actually played out — and an honest note that the service is still maturing
Track Record
How our recommendations actually performed
In Track Record we don't tell stories about «how good we are» — we show what happened after the final whistle. Once a match is over, the bet gets a status, the history accumulates, and you can inspect ROI and hit rate by market over the chosen period. Not slogans — a table that lets you decide whether to trust our signals right now.
View track recordWe're still in shake-down mode
Take any single prediction with a level head
Models learn, data updates, logic gets refined — like any living product. So we say it directly: do not stake everything on a single article. Cross-check the track record, read the verdict for each market, hold your own limits — and treat any prediction as one brick in your picture, not a copy-paste solution.
An honest caveat: the fair price is an estimate of what the market thinks, not a guarantee. Prices and limits change, the settled sample is still small, and early backtest results (positive on favourites) are not yet statistically proven. We don’t publish stake-size advice.
Model Version History
How the methodology above has actually changed, and what each version's real record looks like
WeBetSmart Analytical Terminals
Three «workspaces» — matches, odds and predictions — when you want a denser view of the market
A terminal isn't about a «cinematic» dark theme — it's a single screen where a lot of information fits into one feed: a ticker, filters, fast answers to «what's happening in the market today and what we think about a given match.» Useful once you're comfortable with the Football section and want to work faster — without extra scrolling.
Matches terminal
Your day on one screen: what’s live, how the line is moving, where the best price beats the fair price, and where our AI analysis is already in. Filter by date and league — perfect for a morning scan or an evening session.
OpenOdds terminal
Every quotation and view mode — from a familiar table to a more «technical» one — collected with a market ticker. Filters for arbitrage, steam moves, and bets where the odds beat the fair price. A coupon helps assemble outcomes without jumping between tabs.
OpenPredictions terminal
Predictions in a dense table: pick, fair probability, line behavior and EV margin at a glance. Expand a row to see the reasoning step by step. After the match the outcome shows up beside the recommendation; a single click takes you to the full record.
OpenAuthored & maintained by
The person responsible for the methodology and its evolution
Our Transparency Guarantees
The principles we follow
100% verifiability
Every figure has a clear source and can be cross-checked
Open methodology
The fair-price method, verdict thresholds and pick rules are published on this page
Reproducibility
Given the same bookmaker prices, anyone can rebuild our fair price — see the worked example
Accuracy tracking
A factual record is kept: ROI, hit rate and history — no after-the-fact tuning to make a slide look good
Frequently Asked Questions
Quick answers to the common ones