NFL Playoff Odds API — Monte Carlo Simulator for Football Postseason Probabilities

Get NFL playoff odds as structured data, not as a web page. This Actor replays every remaining game of the NFL regular season thousands of times on the real schedule and returns, for all 32 teams, the probability of winning the division, taking a wild card, reaching the playoffs, and landing the No. 1 seed — plus projected records, market edges and Kelly-style bet sizing.

Playoff odds tables are published all over the web, but almost never as an API you can pull into a spreadsheet, a model or an AI agent. This Actor computes them from scratch and hands you clean JSON, CSV, Excel or an API endpoint.

No API key. No account. No configuration. Standings and the remaining schedule come from the public ESPN feed. Run it with an empty input and it projects the current season.

---

What you get

• Playoff probabilities for all 32 NFL teams — division, wild card, overall playoff berth, and No. 1 seed
• Projected final records — mean wins, losses and ties across every simulated season
• The strength estimates behind the numbers — actual win %, Pythagorean win % from points scored and allowed, and the regressed true-talent estimate the simulation actually uses
• Market edge — supply bookmaker implied probabilities and get the gap between model and market, per team
• Bet sizing — quarter-Kelly stake suggestions with per-position and total-portfolio caps
• Three ready-made dataset views — playoff overview, market edge table, and the records / point differential the model was built from

Export to JSON, CSV, Excel or XML, call it from the Apify API, schedule it weekly, or plug it into n8n, Make, Zapier or an MCP-enabled AI agent.

---

Who uses this

Use case	What you do with it
Sports betting models	Compare fair probabilities against sportsbook futures prices and size positions with the built-in Kelly caps
Fantasy football	Know which teams are still playing meaningful games in December before you plan your playoff roster
Sports media and blogs	A defensible playoff-odds table you generated yourself, refreshed every Tuesday morning
Data science and backtesting	Set a past season and project it from where it stood, then score the model against what happened
Dashboards	A weekly scheduled run feeding Google Sheets, a database, or a BI tool
AI agents	A tool that returns real computed probabilities instead of hallucinated ones

---

Quick start

Run it with an empty input:

{}

That projects the current season with 20,000 simulated seasons. A full 32-team run finishes in a few seconds.

To compare against the market, add bookmaker implied probabilities:

{
  "iterations": 20000,
  "marketProbabilities": [
    { "team": "Kansas City Chiefs", "market": "playoff",  "impliedProbability": 0.82 },
    { "team": "Detroit Lions",      "market": "division", "impliedProbability": 0.55 }
  ]
}

Team names are matched loosely — "chiefs" and "Kansas City Chiefs" both work. market accepts playoff or division.

---

Why simulate instead of scraping a playoff odds page

Most football tools copy numbers out of a website. This one computes something that is not published as data anywhere: a full distribution of season outcomes built from the actual games each team still has to play. Four design choices make the difference.

It simulates the real schedule, not an average one. NFL schedules are unbalanced by design — place-based scheduling means a first-place team draws other first-place teams. A 6-4 club with three divisional road games left has a harder path than its record suggests. Every remaining fixture is loaded individually, with the correct home and away side, and played out.

It leans on point differential, not just the record. Points scored and allowed predict future wins better than the win-loss column does, and over a 17-game season that gap is enormous. A 7-3 team outscored by 20 points is not a 7-3 team. The default weight on Pythagorean expectation is 0.65 and you can move it.

It treats a 17-game sample as the small sample it is. Every simulated season redraws each team's true talent from a distribution rather than assuming this year's record is the truth. Without that, a 17-game season produces absurdly confident probabilities. This is why the regression setting here (6 games) is deliberately much smaller than you would use in baseball, while the strength uncertainty is larger.

It reproduces the real playoff format. Four division winners per conference, seeded 1 through 4; the three best non-winners take the wild cards; the top division winner in each conference gets the first-round bye. Ties count as half a win in the standings, exactly as the league treats them.

---

How it works

1. Load the standings — wins, losses, ties, points for and points against for all 32 teams, from the public ESPN feed.
2. Estimate true talent — blend the actual winning percentage with the Pythagorean expectation from point differential (exponent 2.37), then regress the result toward .500 by a configurable number of games.
3. Load the remaining schedule — every unplayed game for every team, deduplicated, with home and away sides.
4. Simulate — for each iteration, redraw every team's strength from a normal distribution around its estimate, then play out every remaining game as a logistic function of the log-odds difference plus home-field advantage, with a small chance of a tie.
5. Seed the field — sort each division, take the four winners, then the three best remaining teams per conference as wild cards, and record the top seed.
6. Aggregate — count how often each team wins its division, takes a wild card, makes the field and lands the No. 1 seed, and average the projected record.

---

Value bets and position sizing

If you supply marketProbabilities, the Actor does more than show you the gap. It sizes the bet.

• playoffEdge / divisionEdge — model probability minus market probability, in percentage points.
• hasValue — true when the best edge for that team clears edgeThreshold (default 0.05, i.e. five points).
• kellyFractionUncapped — quarter-Kelly stake as a fraction of bankroll, before any limits.
• suggestedFractionOfBankroll — the same number after clipping to maxPerPositionPct (default 2% of bankroll).
• suggestedStake — that fraction times your bankroll, in dollars.
• cappedByPortfolioLimit — true when the bet qualified but the total exposure ceiling (maxTotalExposurePct, default 6%) was already used up by better edges. Those rows come back with a zero stake so you can see what was dropped and why.

Positions are filled best-edge-first. At the defaults, 6% total over a 2% per-position cap means roughly three bets fit.

Quarter-Kelly is a deliberate choice: full Kelly is optimal only if your probabilities are exactly right, and nobody's are.

---

Input

Every field is optional. An empty input gives you the current season with sensible defaults.

Field	Type	Default	What it does
iterations	integer	20000	Simulated seasons. More iterations, smoother probabilities.
season	integer	current	Set a past year to project that season from where it stood — useful for backtesting.
archiveToNamedDataset	string	empty	Name of a dataset in your account to also append this run's rows to. Named datasets are kept indefinitely, so scheduled runs accumulate a permanent history. See “Build your own history” below.
regressionGames	integer	6	How hard a partial season is pulled toward .500. Small on purpose: an NFL season is only 17 games.
pythagoreanWeight	number	0.65	How much of the strength estimate comes from point differential rather than the raw record.
homeFieldAdvantage	number	0.2	Added to the home team's strength in log-odds. 0.2 ≈ a 55% home win rate between equal teams.
strengthUncertainty	number	0.16	How unsure the model is about each team's true talent. Setting it to zero makes the output far too confident.
tieProbability	number	0.004	Chance any single game ends in a tie — roughly one per season across 272 games.
edgeThreshold	number	0.05	Minimum model-minus-market gap to flag a team as value.
bankroll	integer	1000	Your capital, in dollars. Scales suggestedStake only.
maxPerPositionPct	number	0.02	Hard ceiling on any single bet, as a fraction of bankroll.
maxTotalExposurePct	number	0.06	Ceiling on the sum of all open positions.
marketProbabilities	array	[]	Bookmaker implied probabilities to compare against. Each entry needs team, market (playoff or division) and impliedProbability.

---

Output

One row per team, sorted by playoff probability. Three prebuilt views let you jump straight to what you need.

Playoff overview — division, team, projected wins, and the four probabilities.

Field	Meaning
fairDivisionProbability	Chance of winning the division
fairWildCardProbability	Chance of making it as a wild card
fairPlayoffProbability	Chance of reaching the playoffs by either route
fairTopSeedProbability	Chance of the No. 1 seed and the first-round bye
projectedWins / projectedLosses / projectedTies	Mean final record across all simulations

Market edge — model vs market, the gap, and the stake sizing described above.

Records and point differential — wins, losses, ties, gamesRemaining, pointsFor, pointsAgainst, pointDifferential, actualWinPct, pythagoreanWinPct and trueTalentWinPct, so you can see exactly what the model thought each team was before it started simulating.

Every row also carries simulations, season and retrievedAt, so a stored dataset stays interpretable months later.

---

Run it every week

NFL games are mostly weekly, so the numbers move in one big step. Schedule a run for Tuesday morning and the odds are fresh for the whole week:

1. Open the Schedules tab in Apify Console
2. Add this Actor
3. Set the cron to 0 13 * * 2 (Tuesdays, 13:00 UTC)

Point the dataset at Google Sheets, a webhook or a database and you have a self-updating playoff odds table.

---

Build your own history

A run's own dataset is temporary — Apify deletes it 31 days after the run. A named dataset is kept indefinitely.

Set archiveToNamedDataset to a name such as nfl-playoff-odds-history and every run appends its rows to that dataset as well as writing its normal one. Point a daily or weekly schedule at it and you accumulate your own record of how each team's playoff probability moved across the season — a time series you cannot reconstruct afterwards, because the standings and remaining schedule that produced each snapshot no longer exist.

Every row already carries retrievedAt, season and simulations, so stacked snapshots are never ambiguous: you always know when a projection was taken, for which season, and how many simulated seasons it came from.

The named dataset is created the first time it is used and lives in your account under Storage → Datasets. Leave the field empty and nothing changes: you get the normal run dataset only.

---

Pricing

Pay per result. You are charged for each team projection returned — a full league run is 32 of them. No platform usage fees on top, no subscription, no minimum.

A full 32-team league run costs about $0.10. Running it once a week for a whole NFL season is roughly $2 in total.

---

FAQ

Where does the data come from? The public ESPN standings and schedule feeds. No API key and no account are required.

Are these the same numbers as FiveThirtyEight or ESPN's FPI? No. Those are different models with different assumptions, and several of them are no longer published as data at all. This one is transparent about its inputs and every knob is exposed, so you can tune it and see exactly what changed.

How many iterations do I need? 20,000 is plenty. Probabilities are typically stable to well under a percentage point at that level. Going to 100,000 costs more time for a difference you will not act on.

Can I project a past season? Yes. Set season to a past year and the Actor projects it from the point the standings stood at, which is how you backtest the model.

Does it handle ties? Yes. Ties count as half a win in the standings, exactly as the NFL does it, and the simulation produces them at a realistic rate.

Does it model the exact NFL tiebreakers? No, and this is worth being clear about. Head-to-head, common games and strength of victory are not reproduced; ties in simulated win totals are broken randomly. Over 20,000 seasons this washes out to a small effect on team probabilities, but it means a specific late-season tiebreaker scenario is not what this tool answers.

Can an AI agent call this? Yes. It works through the Apify MCP server like any other Actor, and it returns computed numbers instead of guessed ones.

Can I get the output in a spreadsheet? Yes. Every run's dataset exports to CSV, Excel, JSON or XML, or you can pull it from the API.

What does hasValue actually mean? That the model's probability beats the market probability you supplied by at least edgeThreshold. It is a flag on your own inputs, not a recommendation.

Is there a version for other sports? There is an MLB version — MLB Playoff Odds API — built on the same engine, and a soccer model for 17 leagues, Football (Soccer) Monte Carlo Season Predictor.

---

Limitations and honest caveats

• Injuries, weather, benched starters and quarterback changes are not modeled. The strength estimate is based only on results and points. A team that just lost its starting quarterback will look better here than it should.
• Tiebreakers are simplified, as described above.
• Early in the season the numbers are soft. With three games played there is very little signal; the regression toward .500 is doing most of the work, and it should be.
• This is a model, not a prediction. A 78% playoff probability means the team misses roughly one season in five. That is the point of a distribution.
• Nothing here is betting advice. The Kelly outputs are arithmetic on numbers you supplied, and the edge is only as good as your market prices.

---

Not affiliated with, endorsed by or sponsored by the National Football League or ESPN. Team names and league marks belong to their respective owners. Data comes from publicly available feeds.