Model Win % vs Betfair Tennis Odds

Run it yourself: the Tennis Trader Board is open source — source on GitHub. It needs a Live Tennis API Ultra key (code botblog) for real-time push.

If you have opened the Tennis Trader Board and noticed the model win % column, you may have wondered exactly what it is measuring — and whether a gap between that number and the Betfair price means anything you can act on.

The short answer: sometimes yes, often no, and understanding the difference is more useful than the number itself.

What the model is

The model win % on the board comes from Live Tennis API's built-in win-probability model. It updates on every point and uses:

• Current set score
• Current game score within the set
• Current points within the game
• Who is serving
• Match format (best of 3 or 5, final-set tiebreak rules)

It is a score-state model. Given those inputs, it calculates the probability of each player winning the match from this exact moment, using baseline assumptions about how often servers hold serve at a given level.

What it does not know:

• Recent form or ranking trajectory
• Surface-specific serving and return stats for these two players
• Physical condition, injury risk, or fatigue across a long match
• Head-to-head patterns between the specific players
• The fact that Djokovic typically holds at 95%+ and an ITF qualifier holds at 55%

This matters because Betfair prices in much of the above. The model uses a generic serve-hold assumption; the market uses everything.

How to calculate the gap

The board does this for you when you paste a Betfair decimal into the P1 odds field. But the maths is simple:

• Betfair implied probability = 1 ÷ decimal odds (e.g. 2.20 → 45.5%)
• Model win % = the number shown on the board for that player
• Gap = model % − implied % (in percentage points)

Example: model says 58%, Betfair decimal is 2.20 (implied 45.5%). Gap = +12.5 pp. The model rates P1 12.5 percentage points higher than the market does.

When the gap is meaningful

A large gap often reflects a recent score event that moves the model faster than the market catches up. The most common situations:

• First set just won — the model updates immediately to reflect the new win probability from 1-0 up; the market may lag by a few seconds, especially in lower-liquidity matches
• Break of serve just landed — model shifts; match odds may not have fully repriced yet
• Early points in a tiebreak — small leads in a tiebreak have outsized model impact; the market sometimes underreacts at the start of a TB

In each case, the gap closes within a few seconds on liquid markets. On ATP top-tier events, algorithmic traders reprice almost instantly. The window is narrow.

When to ignore the gap

A gap is not informative — and can be actively misleading — in these situations:

• Big serve matchup — against two top servers, the model's hold-rate assumption is too low. It will consistently overestimate the trailing player's chance of breaking back, showing a false positive gap favouring them.
• Dominant player running away — if Alcaraz is 6-0, 4-0 up and 15-40 down on serve, the model may not fully discount how good he is at holding in that position. The market already has.
• Medical timeout just called — the model has no idea. The market may be repricing significantly based on what the crowd and TV are showing. The gap here is noise.
• Low-liquidity match — the Betfair price itself is stale or wide-spread. The implied probability is unreliable, so the gap calculation is unreliable.

How to use it in practice

Think of the model gap as a consistency check, not a signal generator.

Before you trade into a position, a quick glance at the gap can answer: "is the market pricing this the way a baseline model would expect, or is it pricing in something extra I should understand?"

If the gap is near zero, market and model agree on the score state. If there is a large positive gap in your favour, it may mean the market sees something the model does not — and that is often a reason to pause, not to pile in.

The most disciplined use: run the board on a second monitor while you trade manually, note when the gap is consistently large or small in certain score states on a particular surface, and use that pattern to refine your own trading rules over time. That is a proper edge-building exercise. The gap on a single point is not.

Connecting this to a bot

If you are building a Betfair tennis trading bot, the model gap can be an additional filter in your entry rules:

• Trigger: break point (score trigger in BF Bot Manager or raw WebSocket in a custom bot)
• Filter: model win % for the server is above X% (i.e. the model still rates the server a favourite despite the break point)
• Additional filter: gap vs Betfair odds is within a range (the market also rates them above even)

Stacking score state + model + price range gives a tighter entry condition than any single filter alone. It does not guarantee a positive edge — you still need to test it over a meaningful sample in Simulation mode — but it reduces the set of trades you are taking.

Related guides

• How to use the Tennis Trader Board
• Break-point trading playbook
• 15–40 tennis trading strategy
• BF Bot Manager tennis automation
• Live Tennis API for exchange bots
• All tennis trading guides

Disclaimer: Model win probability and board data are informational only. Not betting advice. This post contains affiliate links to Live Tennis API and BF Bot Manager. Full disclosure.