CASE STUDY / DATA DON'T LOSE (AS MUCH)
It Started with Football:
Turning Data into Decisions
How Team d²l combined data engineering, Alkemy, and analytics to build a data and machine learning (ML) pipeline that predicts college football point spreads and totals
80% reduction in research time
62% prediction accuracy, yielding an 18% ROI
100% automated pipelines
Extensible to other sports or domains
The Client
Our Fearless Founder, a.k.a Don
When he's not running Team d²l, Don enjoys a variety of sports and the brief surge of adrenaline that comes with wagering.
As a stats-minded gambler, he's been known to spend fall Saturday mornings sitting down with his warm chai, a dozen spreadsheets, and too many opinions. He'll spend hours pulling stats from various APIs and sites, checking injuries, and recalculating odds using messy Excel tabs and formulas.
Each week is a fresh grind with updating data, ignoring noise, and prior experience doing most of the heavy lifting. As a data scientist with access to some powerful technology, Don knows there's a way to sharpen his edge and personally realize the value of his team's work - a win-win.
Company size
1 lone wolf
Industry
Personal vices
Tech stack
Team d²l engineering tooling, Alkemy, Zenlytic
“If the models beat me, great. If I beat the models, even better. Either way, we learn.”
Don
Gambler
Team d²l
This wasn’t just about winning bets; it was about proving that a repetitive, data-heavy decision process could be transformed into a scalable, automated system.
Project Overview
Game On
Team d²l ran with Don's request. It was an experiment in bridging data and intuition, anchored in the key question:
Could Team d²l turn sportsbook data into a reliable, automated pipeline for producing and evaluating predictive strategies?
Don already knew some of the hurdles:
Data Sourcing
CollegeFootballData and SportsDataIO API's needed to be synced and standardized.
Feature Selection
With hundreds of possible stats, deciding what actually influenced game outcomes mattered most.
Model Tuning
Markets adapt quickly, models must stay ahead.
Turning Data into Decisions
Don and the team combined Team d²l’s internal data engineering tooling, Alkemy, and Zenlytic, a Business Intelligence (BI) tool, to automate the path from raw data to actionable predictions as quickly as possible.
Using data pipeline capabilities developed by the team, engineering stood up the underlying infrastructure in days.
Don then worked closely with the data science team, treating himself like any other Alkemy client. He knew which stats mattered most, and that a model could handle the math. He reviewed model outputs, gave feedback, and helped shape what “good” predictions looked like. Together, they found that balance between instinct and automation that’s hard to replicate in betting… or business.
Team d²l’s internal data engineering tooling handled the data plumbing, regularly pulling live and historical stats through Airbyte connectors, transforming and standardizing them with dbt, and writing everything into BigQuery.
Airflow orchestration ensured data consistency, scheduling, and monitoring to keep the pipeline running reliably, so Don never had to worry about missing a feed before kickoff.
Alkemy replaced the spreadsheet chaos with structure and reproducibility. It organized model training into clearly versioned experiments and trials, complete with input tracking, output storage, and easy retraining.
For this project, Alkemy ran two models: a Point Spread Model to predict the margin between teams, and a Totals Model to predict the combined points scored by both teams.
"The models react quicker than I can to emerging trends."
Don
Zenlytic: The Eyes
Zenlytic took the model outputs and turned them into a readable, interactive dashboard, which is perfect for Don, who’d rather watch football than write SQL.
The dashboard tracks:
-
Weekly model accuracy vs. actual results
-
Spread vs. total prediction variance
-
Upcoming wager recommendations
There's no more digging through spreadsheet tabs, just clean, visual performance metrics at a glance.
Zenlytic dashboard showing weekly wagers
"I don't need to see every number anymore. I just need to see the ones that matter."
Don
Before Team d²l rebuilt the workflow, Don’s “pipeline” was a combination of tabbed spreadsheets and intuition. Afterward, it became a fully orchestrated system—consistent, measurable, and reproducible.
Before: Don spent hours researching stats and matchups manually.
After: He opens a dashboard and reviews predictions with his morning chai.
Don now gets updated predictions for morning, afternoon and evening games. Inference is as easy as calling the next play.
Results
Gambling Wins Through Data
He still brings his gut to every game, but now it’s backed by a cloud of data.
Efficiency Gains
Don spends 80% less time prepping each week, allowing more time for analyzing outcomes.
Performance
Models show a consistent prediction accuracy of 62%, well above the profitability threshold.
Reliability
The data pipeline runs automatically with zero manual pulls.
Scalability
The underlying infrastructure can expand to any sport or domain.
Human Impact
Don now spends more time watching football and less time debugging spreadsheets.
The point isn’t football, it’s proof. If Team d²l can turn Don’s Saturday betting routine into a data-driven decision engine, imagine what the same approach could do for your organization’s toughest predictive problems.
What's Next
Beyond Sports Betting
The success of this project has sparked new ideas for applying the same approach beyond the gridiron:
Fantasy Sports
Player-level predictions and lineup optimizations
Real Estate Investing
Forecasting property performance using market and demographic data
Content Analysis
Predicting engagement and virality trends
Financial Analytics
Building similar ML pipelines for portfolio prediction