About BBMI

It started with a family NCAA bracket challenge. I built a quick model to get an edge, the model worked better than expected, and I got nerd-sniped into something more serious. What began as a basketball experiment now covers 2,005+ documented NCAA basketball games, a full WIAA high school season, a walk-forward validated NCAA football model, NCAA baseball, and a walk-forward validated MLB model.

I've spent decades as a risk manager building predictive models for healthcare costs and revenue forecasting. The core disciplines — data quality, variable selection, calibration, and out-of-sample validation — translate surprisingly well to sports. Once I noticed the model's projected lines were consistently closer to actual outcomes than Vegas, the logical next step was to track it rigorously and see if the edge was real.

The goal is the same across every sport: publish picks before games, record every result publicly, and let the cumulative record speak for itself. No cherry-picking. No retroactive adjustments.

The BBMI generates its own predicted point spread for every game — independently of what Vegas has set. The gap between the BBMI line and the Vegas line is what we call the "edge." The bigger the edge, the more strongly the model disagrees with the sportsbooks.

Each sport uses a model tailored to its specific inputs, but the same core framework applies: team strength is evaluated using scoring efficiency, opponent quality, and situational factors, then transformed into a projected spread and win probability for each matchup.

• Basketball: Offensive/defensive efficiency, tempo, RPI, home court
• Football: SP+ efficiency ratings, opponent-adjusted box scores, quality wins, home field, altitude, weather-adjusted totals, early-season spread corrections. Walk-forward validated across 2 independent seasons with 0.0pt overfitting gap
• NCAA Baseball: Run scoring, ERA, pitcher adjustments, dynamic park factors, WHIP
• MLB: Negative Binomial engine, FIP-based pitcher ratings, park-neutral wOBA, asymmetric park factors, Bayesian blending, walk-forward validated
• WIAA: Same basketball framework — more noise due to self-reported stats

The goal isn't perfection on any single game. It's consistent, repeatable accuracy across a large sample — and the public log is the proof.

Every pick BBMI has ever made is logged publicly at ncaa-model-picks-history. Wins, losses, dates, spreads, simulated returns — all of it, from the first pick of the season, unedited.

This approach is borrowed directly from risk management practice: a model that can't be validated against out-of-sample data isn't worth trusting. The public log isn't a marketing tactic — it's the only honest way to evaluate whether the model actually works.

Sports betting is one of the few markets where an informed individual can hold a structural advantage over the house — not because Vegas is bad at math, but because Vegas is solving a different problem than you are.

What Vegas is actually optimizing for

Sportsbooks aren't purely trying to predict the correct outcome — they're managing liability. When 80% of public money lands on one side, the book will shade the line to attract action on the other side, even if their internal model says the original number was right. A model focused purely on accuracy — not risk management — can exploit those gaps.

Where BBMI fits in

BBMI doesn't need to be smarter than the sportsbook's internal model. It needs to be smarter than the posted line — the number that's already been distorted by public money, liability balancing, and market incentives. College sports are one of the best places to clear that bar: thin markets, less sharp money, and hundreds of teams that receive far less analytical attention than the pros.

The sports betting information industry is full of services selling picks with no verifiable track record. BBMI was built specifically to be the opposite of that.

The honest version of our pitch: the numbers are in the cards above — real, verifiable, and not perfect. We'd rather you evaluate the actual record than take our word for it.