AI NBA Picks: Spreads, Totals and Player Props

In the NBA, the edge is who plays

Basketball is the sport where a single player moves the line most. Five players share the floor, stars play heavy minutes, and teams rest players more aggressively than in any other league. A spread can swing several points between the morning line and tip-off because one name went from questionable to out. That makes the NBA a timing market as much as a handicapping one.

The other thing that makes the NBA distinct is volume. Most nights bring a full slate, every team plays 82 games, and those games are dense enough that rest, travel and schedule fatigue are measurable. The model is built around those two realities: price availability precisely, and price the schedule honestly.

The factors the model weighs

Efficiency per 100 possessions. Points per game rewards fast teams and punishes slow ones. The model works in offensive and defensive rating per 100 possessions, adjusted for opponent strength, which tells you how good a team is independent of how fast it plays.

Lineups, not just rosters. A team's season numbers include games its best player missed. The model estimates efficiency for the players who will actually be on the floor, so a star's absence is priced as the specific rotation that replaces him, not as a generic "minus a few points."

Minutes projections. Every prop and most spreads come down to minutes. Who starts, who closes, who's on a minutes restriction after injury, and who absorbs the shots and touches when someone sits. When a player is ruled out late, the model reallocates his minutes and usage across the rotation.

Pace. A total is roughly possessions times points per possession. Two fast teams create extra possessions; a fast team against a slow, half-court defense creates a tug-of-war. The model projects possessions for each matchup rather than averaging two teams' paces.

Rest and the schedule. The second night of a back-to-back, a fourth game in six nights, the end of a long road trip, or a game in Denver after a late flight all register. So does the opposite: a rested team at home against a tired one.

Three-point variance. Opponent three-point percentage is mostly outside a defense's control over short stretches. A team that "held opponents to 30% from three" for two weeks probably got lucky. The model leans on shot volume and shot quality, and lets hot and cold shooting regress.

Turnovers, rebounding and free throws. The other levers of possession value. A team that forces turnovers and wins the offensive glass gets extra shots; a team that sends opponents to the line gives away cheap points and slows the game down.

Motivation, carefully. Late-season tanking, play-in races and seeding battles are real, but they're hard to measure. The model applies small, conservative adjustments and leans on confirmed rotations instead.

Market by market

Spreads are the heart of the NBA card, and most of the edge comes from availability. When a starter is ruled out, the market adjusts, but it doesn't always adjust correctly for who replaces him and how the rotation changes.

Totals depend on possessions. The model also accounts for blowout risk: lopsided games often slow down late as benches empty, which can drag a total under even when both offenses are efficient.

First-half lines isolate the starters. Teams with strong starting units and weak benches can look very different over 24 minutes than over 48.

Player props (points, rebounds, assists, threes made, and combinations like points plus rebounds plus assists) start from minutes and usage. When a teammate sits, someone's shot volume goes up. The model projects how that volume is redistributed rather than assuming the next man up simply inherits it.

Situational spots the model watches

• Back-to-backs. Especially the road team on the second night, and especially when the first game went to overtime.
• Long road trips. The last games of a five- or six-game trip tend to carry fatigue.
• Altitude. Games in Denver and Utah ask more of visiting teams' conditioning.
• Rest disadvantage. A rested team at home against a team on its third game in four nights.
• Post-trade-deadline rotations. New roles take time, and the model widens its uncertainty until rotations settle.
• Late-season rest. Teams that have clinched their seed, or out of the race, play their regulars less. The model is cautious until lineups are confirmed.

NBA season calendar

What happens when a star is ruled out late

Late scratches are where NBA prices are most often wrong, so the model treats them as a process rather than a reflex.

1. Remove the player's minutes. The model takes the minutes he was projected to play and redistributes them across the rotation the coach is most likely to use, including who moves into the starting lineup.
2. Re-price the lineup. The team's efficiency is recalculated for the players who will actually be on the floor, on both offense and defense. Losing a great defender can matter as much as losing a scorer.
3. Re-project usage. The shots, drives and assists he would have produced don't vanish. The model estimates who absorbs them, which is where player props often become mispriced.
4. Compare to the moved line. The market will have moved too. The question is whether it moved the right amount. Sometimes the market overreacts to a big name; sometimes it underreacts to a role player whose absence breaks the rotation.

Only if the new fair price and the new market price disagree by enough does a pick go out.

Mistakes the model is built to avoid

• Trusting a hot shooting week. Three-point shooting regresses, for and against.
• Treating points per game as quality. Fast teams score more and allow more. Efficiency per possession is the real measure.
• Pricing names instead of minutes. A star on a minutes restriction isn't the same player.
• Ignoring blowout risk. Big spreads and totals are shaped by what happens when benches empty.
• Betting every game on the slate. Most NBA lines are close to right. The model passes on most of them.

How many NBA picks, and when

Pro members get about five picks a day across all the sports we cover. During the winter, a good share of those are usually NBA, but the number on any night depends on how many prices the model likes. A 12-game Wednesday might produce two NBA picks; a quiet Monday might produce none.

Timing follows the injury report. Some picks go out early when an opening number is clearly off. Many wait until afternoon availability updates, because a pick released before a star is ruled out isn't a pick worth selling. Every pick goes out before tip-off and is timestamped.

How NBA picks are graded

• Spreads, totals and moneylines grade on the final score, including overtime.
• First-half picks grade at halftime.
• Player props grade from the official box score, including overtime stats.
• A result that lands exactly on the number is a push.
• Every pick is graded against the line it was released at, and losses stay on the record.

What Pro members get during basketball season

• Every NBA pick with the line, price, confidence rating, suggested stake size and a short write-up
• Picks across every other sport we cover, about five a day in total
• Live scores and play-by-play for the games you're following
• The bet tracker and dashboard to log your own bets alongside ours
• Live score alerts on your active bets

Playoff basketball is a different sport

In the playoffs, rotations tighten to eight or nine players, stars play heavier minutes, and coaches adjust game to game. The model gives more weight to playoff-style lineups and to the information from earlier games in the same series, and it re-prices each game individually rather than assuming the series will follow the regular season.

NBA betting glossary

• Offensive / defensive rating: Points scored or allowed per 100 possessions.
• Net rating: Offensive rating minus defensive rating.
• Pace: Possessions per 48 minutes.
• Usage rate: The share of a team's possessions a player ends while on the floor.
• Back-to-back: Games on consecutive nights.
• PRA: A prop combining points, rebounds and assists.
• Questionable / doubtful / out: NBA injury designations, updated through game day.
• Garbage time: Late minutes of a blowout when starters sit, which distorts stats and totals.