Baseball is the most analytically intense sport on the planet. Since the moneyball revolution of the early 2000s, MLB teams have built entire decision frameworks around advanced metrics — roster construction, in-game tactics, pitcher management. Yet most bettors still analyze games with blunt instruments: the final score, the bookmaker's line, and not much else.
Our MLB AI model works through each game in a structured 5-step sequence, pulling Statcast data (exit velocity, launch angle, spin rate), advanced pitcher metrics (xERA, FIP, SIERA), bullpen workload, park factors, and hourly weather forecasts. The results: 63% win rate on F5 totals (1,101 official settled picks) and 65.2% on Run Lines (342 picks) — built on a reproducible methodology, not luck.
This guide walks through the full method our AI uses to analyze every MLB game. Whether you're new to baseball betting or looking to build a more structured approach, these five steps give you a framework you can apply to any game on the schedule.
Step 1 — Analyze the starting pitcher duel
The starting pitcher is the single most influential individual variable in a baseball game. Unlike the NFL (17 games for a quarterback) or the NBA (82 games for a star), a starting pitcher takes the mound roughly every five days. Each start is its own event — quality shifts based on current form, his track record against the opposing lineup, and that night's conditions.
Key metrics to examine
| Metric | What it measures | Why it matters |
|---|---|---|
| ERA | Earned runs allowed per 9 innings | Classic indicator, skewed by the defense behind the pitcher |
| xERA | Expected ERA based on contact quality (Statcast) | Catches pitchers who are outperforming or due for regression |
| FIP | Expected ERA stripping out defensive influence | Isolates the pitcher's true performance |
| WHIP | Walkers + hitters reached per inning | Measures base control and pressure on the opposing team |
| K/9 & BB/9 | Strikeouts / walks per 9 innings | Whether dominance comes from swing-and-miss or contact management |
| SIERA | Context-weighted expected ERA | The most predictive metric in sabermetric research |
The core rule: use xERA and FIP instead of ERA. A pitcher with a 3.50 ERA and a 4.80 xERA is overperforming — his recent results won't hold. Conversely, a pitcher at 5.10 ERA / 3.70 xERA is getting hurt by bad luck or poor defense and is undervalued by the market.
Concrete example — 2026 season:
Pitcher A: ERA 3.12 / xERA 4.95 / FIP 4.70
→ Overperforming, regression likely → avoid the F5 Under
Pitcher B: ERA 5.40 / xERA 3.85 / FIP 3.90
→ Underperforming, potential value → market is underestimating his real form
If these two face off, the bookmaker sets the line on visible ERA.
Our model calculates from xERA → detectable edge.
Platoon splits: the detail that changes everything
A right-handed pitcher statistically dominates right-handed hitters, and vice versa. Our model integrates the platoon splits of the opposing lineup: when a right-hander faces a team where 70% of hitters are right-handed (his weaker matchup), total projections rise by 0.3 to 0.5 additional runs. Most casual bettors ignore this — and it explains a meaningful share of our edge in totals markets.
The same logic applies to recent performance: a pitcher coming off three rough outings with altered mechanics (rising WHIP + falling K/9) carries risk that his season-to-date ERA doesn't yet reflect.
For a deeper look at pitcher metrics, our guide ERA vs xERA: why traditional ERA misleads your MLB bets covers the specific regression patterns worth tracking.
Join ProbWin for free
Access our AI predictions for NBA, NHL, Soccer & Tennis. Sign up in 30 seconds.
Step 2 — Assess bullpen condition
One of the most expensive mistakes in MLB betting is tunnel-visioning on the starter while ignoring the bullpen. In 2026, MLB has fully entered the opener/multi-pitcher era. Starters average just 5.2 innings — the bullpen handles everything else.
Key indicators of bullpen health
| Indicator | Alert threshold | Impact on bets |
|---|---|---|
| Pitchers used over 3 days | > 4 relievers | Fatigued bullpen → potential total increase |
| Innings logged over 7 days | > 20 innings | Collective exhaustion → likely breakdowns |
| High-leverage ERA | > 4.50 | Poor in critical situations → line at risk |
| Reliever WHIP (30 days) | > 1.45 | Lack of control → walks and runs |
| Saves converted/opportunities | < 75% | Unreliable closer → extreme late-game variance |
Our model pulls real-time data from mlb_bullpen_fatigue to quantify reliever workload over 3, 7, and 14 days. If a favorite's bullpen has absorbed 18 innings over 5 days and that night's starter is an inexperienced spot starter, that's a strong signal on the full-game total (FT).
Classic scenario — bullpen overuse:
Context: Team A played 4 consecutive games in 4 days
Bullpen usage: 5 different relievers last night (18 pitches each)
Tonight's starter: spot starter, projected 4 innings max
Innings exposed to bullpen: 5, 6, 7, 8, 9
→ OVER signal (Full Game Total) if the line hasn't moved from yesterday
→ Possible F5 Under if the spot starter is solid through 4 innings
→ Avoid the Run Line: too much variance from bullpen uncertainty
An exhausted bullpen is one of the most exploitable situations in MLB because bookmakers don't always adjust total lines in real time. Our model tracks lineup submissions up to 90 minutes before first pitch.
Step 3 — Factor in park factors and weather
Baseball is unique among major North American sports: the playing field is radically different at each venue. Coors Field in Denver (altitude 5,200 feet, thinner air = ball traveling 8-10% farther) and Oracle Park in San Francisco (marine fog + regular headwind) produce diametrically opposite offensive profiles. Ignoring the ballpark when analyzing a total means accepting a permanent structural disadvantage.
Profile of the main MLB stadiums in 2026
| Stadium | Team | Park Factor (total) | Profile |
|---|---|---|---|
| Coors Field | Rockies | 1.18 | Altitude + dry air → +15-18% more runs vs neutral park |
| Globe Life Field | Rangers | 1.09 | Dry Texas air → ball carries well |
| Fenway Park | Red Sox | 1.07 | Green Monster → easier home runs to left field |
| Dodger Stadium | Dodgers | 0.94 | Slight pitcher-friendly, wide foul territory |
| Oracle Park | Giants | 0.88 | Fog + sea wind → suppresses offense |
| T-Mobile Park | Mariners | 0.91 | Humidity → less carry, pitchers advantaged |
A park factor of 1.10 means the venue produces 10% more runs than a neutral park. Our model applies these adjustments directly to every total projection — it's baked in, not a footnote.
Weather: the most underestimated factor
A wind blowing in from the outfield can cut the distance of batted balls by 15-20 feet, wiping out several potential home runs. A wind blowing out does the opposite. Temperature matters too: a ball hit at 95°F travels roughly 10 feet farther than the same hit at 50°F.
Our model pulls hourly weather forecasts for every game — a capability few amateur bettors develop. A game at Coors Field with a 15 mph headwind can see its projected total drop from 10-12 to 8.5-9, even when the bookmaker's line hasn't moved because it was set hours earlier.
Step 4 — Choose the right market (Moneyline, F5, or Run Line)
One of baseball's great advantages for analytical bettors is market variety. Unlike most sports where the final result is the primary option, MLB offers three main markets with very different logic and risk profiles.
Comparing the three main markets
| Market | What it covers | ProbWin results (all-time) | Best use case |
|---|---|---|---|
| First 5 Innings (F5) | Innings 1-5 | 63.0% WR / +238.5u across 1,101 picks | Ace duel, risky bullpen situation |
| Full Game Totals (FT) | 9 complete innings | 61.4% WR / +214.4u across 1,167 picks | Fatigued bullpen + offense-friendly weather |
| Run Line / Spreads | Win by ≥2 runs | 65.2% WR / +45.2u across 342 picks | Clear favorite with a dominant starter |
When to focus on the F5 market?
The First 5 Innings is our best historical market for a straightforward reason: by removing bullpen uncertainty (innings 6-9), you're only betting on what can be analyzed precisely — the two confirmed starters. When the matchup features two dominant aces (xERA < 3.50, K/9 > 9.5), the F5 Under is often underpriced because bookmakers reason from full-game statistics.
Conversely, when one starter is shaky but facing a strong bullpen on the other side, the F5 Over offers edge on the first five innings without taking on the risk of a late game managed by an elite relief corps.
When to prefer full-game totals?
FT totals become attractive in three specific situations: - The starter is unlikely to complete five clean innings (recent poor outings + WHIP > 1.5) - The opposing bullpen is clearly gassed (signal for a higher full-game total) - Weather favors offense (wind blowing out + high temperature at Coors or Globe Life)
The Run Line: save it for clear spots
The Run Line (-1.5 for the favorite) amplifies variance but offers a better price. Our model reserves it for clear situations: a solid favorite with an ace on the mound against a team struggling offensively. At 65.2% across 342 picks, it's our most precise market — but also our most selective. We only play the Run Line when every signal points the same direction.
Decision tree — which market to choose:
1. Confirmed ace duel + both bullpens intact
→ Analyze F5 (Under or Over based on platoon splits)
2. Shaky starter (xERA > 4.5) or clearly fatigued bullpen
→ Consider Full Game Total / Over if matchup supports it
3. Clear favorite with a dominant starter + opposing team in offensive trouble
→ Run Line (-1.5) if model edge > 7 points
4. Uncertain conditions (starter unconfirmed, shifting weather)
→ Wait or pass (no bet)
Step 5 — AI model confirmation and expected value calculation
The first four steps form the qualitative analysis of a game. Step 5 is where our model synthesizes everything to calculate an objective expected value (EV) and determine whether the bet is worth placing.
ProbWin MLB model architecture
Our technical pipeline runs through several automated phases:
- Scout: collects live odds, pitcher stats (MLB Stats API + Statcast), bullpen workload, lineup confirmations, park factors, and hourly weather
- ML model: trained on seasonal history, generates a probability for each side/total in each game
- Executor: compares model probability against bookmaker implied probability, calculates edge and EV, applies safeguards (confirmed starter, minimum edge threshold)
Sample edge calculation:
Offered line: OVER 8.5 runs @ -115
Bookmaker implied probability: 53.5%
Model probability: 62.8%
Edge = 62.8% - 53.5% = +9.3 points
EV = (0.628 x 100/115) - (0.372 x 1) = +17.5% (before vig)
→ Official pick if edge > threshold (typically +5%)
→ Stake sized using fractional Kelly formula
Model safeguards
Our system never sends a pick on a "probable" unconfirmed starter for F5 markets — a last-minute spot starter invalidates the entire analysis. Similarly, if odds shift significantly after our initial projection (sharp money detected), the model recalculates before confirming.
These safeguards explain why our daily pick count is sometimes low: across a 15-game slate, the model might find only 2 or 3 games with a genuine edge. Selectivity is the foundation of long-term profitability.
Verified official results
All official settled picks in the ProbWin MLB database:
| Market | Settled picks | Wins | Win rate | Units won |
|---|---|---|---|---|
| F5 Totals (Over/Under) | 1,101 | 694 | 63.0% | +238.5u |
| Full Game Totals (Over/Under) | 1,167 | 716 | 61.4% | +214.4u |
| Run Line / Spreads | 342 | 223 | 65.2% | +45.2u |
A 63% win rate across more than 1,000 picks is statistically significant. For reference, the break-even threshold at an average line of -110 is 52.4% — our model beats it by more than 10 percentage points across every market.
For a broader look at how our AI works across all sports, see our guide to AI sports betting and machine learning.
5 MLB analysis mistakes to stop making
Even when following the method above, these recurring traps degrade analysis quality.
x "ERA of 2.80 → this team will dominate"
v Check xERA and FIP — a low ERA can hide an overperformer heading for regression
x "Team is hot (7-3 over last 10) → automatic confidence"
v Analyze tonight's specific conditions: opponent quality, bullpen state
x "Coors Field = automatic Over"
v Even Coors can produce an Under: strong headwind + two dominant aces
x Betting before starter confirmation
v Always wait for the official confirmation (up to 90 min before first pitch)
x Playing the Run Line by default for the better odds
v Run Line amplifies variance — reserve it for clear favorites with a confirmed ace
Frequently asked questions about MLB analysis
What's the difference between ERA and xERA?
ERA (Earned Run Average) measures runs actually allowed per 9 innings. xERA (Expected ERA) calculates the runs expected based on contact quality — measured by Statcast (exit velocity, launch angle, spin rate). A pitcher with a 3.00 ERA and a 4.50 xERA is outperforming through luck or defensive support; the regression is predictable. Our model uses xERA throughout to avoid short-term bias.
Do you always need to wait for starter confirmation?
Yes, especially for F5 markets. An unexpected spot starter completely invalidates the pitcher duel analysis. Our Scout monitors confirmations in real time and can adjust or cancel a pick if the pitcher changes after publication. That discipline is why our daily volume is sometimes low: we'd rather pass than bet on stale analysis.
How much can weather move a total projection?
A 15+ mph tailwind can push a game's projected total up by 1 to 1.5 runs. An equivalent headwind does the reverse. At Coors Field, the effect can be larger due to altitude. Bookmakers set lines hours before first pitch — if weather deteriorates after the line is posted, the edge can disappear or flip entirely.
When does Moneyline make more sense than Run Line?
The Moneyline is better when the favorite is solid but not certain to win by 2+ runs — for example, a favorite facing a quality opposing starter backed by a deep bullpen. Taking the Moneyline at -180 costs less variance than the Run Line at -1.5, even if the price looks less attractive on paper.
Next step
The 5-step method — pitcher duel analysis (xERA/FIP/SIERA + platoon splits), bullpen condition, park factors and weather, market selection (F5/FT/Run Line), and AI model confirmation — is the framework we apply to every MLB game. No magic formula: just a systematic process that, run consistently across a sufficient volume of picks, produces a measurable and repeatable edge.
To compare with structured analysis approaches in other sports, our 7-step NHL game analysis method and our complete 5-step tennis analysis method cover analogous frameworks adapted to their respective sports.
To measure your betting performance over time and understand how bookmakers identify winning bettors, our guide on Closing Line Value (CLV) is required reading for anyone building a serious, sustainable approach.
For a current look at which teams and pitchers are in form right now, check our MLB predictions hub where we track live model data throughout the season.
