Every sharp bettor you’ve heard of started somewhere, and for most of them, “somewhere” was a spreadsheet with team ratings and a formula for projecting spreads. You don’t need a PhD in statistics or a custom Python pipeline to build a functional power rating model. You need a spreadsheet application, a few publicly available data sources, and the willingness to spend a few hours each week updating your numbers. The model won’t be perfect — no model is — but it will give you something infinitely more valuable than gut instinct: a structured, repeatable framework for evaluating every game on the board.
The Mechanics of NFL Power Ratings
A power rating is a single number assigned to each team that represents their overall strength on a neutral field. The concept is simple: if Team A has a power rating of 85 and Team B has a rating of 78, Team A is projected to win by 7 points on a neutral field. Add a home-field adjustment for whichever team is hosting, and you have a projected spread that you can compare to the sportsbook’s line.
Power ratings work because they force you to evaluate teams relative to each other on a consistent scale. Without them, your analysis is ad hoc — you look at a matchup, consider a few factors, and arrive at a subjective opinion. With power ratings, every opinion is anchored to a system. Your system might be simple, but it applies the same logic to every game, which eliminates many of the cognitive biases that plague subjective handicapping.
The elegance of power ratings lies in their modularity. You can start with a crude version based on point differential, then refine it over time by incorporating efficiency metrics, strength of schedule adjustments, and situational factors. Each improvement adds a layer of sophistication without requiring you to rebuild from scratch. The spreadsheet grows with your understanding of the game, and after a few seasons of iteration, you’ll have a personalized tool that reflects your unique analytical perspective.
Choosing Your Inputs
The foundation of any power rating model is the data you feed it. For a beginner-friendly model, start with three core inputs: points scored per game, points allowed per game, and strength of schedule. These are available for free from sites like Sports Reference, ESPN, and the NCAA’s own statistical database.
Points scored and allowed form the basis of your point differential, which is the single best predictor of future team performance available in simple box score data. A team that scores 28 and allows 17 per game has a point differential of +11, and that number correlates more strongly with future wins than win-loss record alone. Why? Because win-loss record is contaminated by luck — close games that could have gone either way, special teams flukes, and other high-variance events. Point differential smooths out that noise and gives you a cleaner picture of underlying quality.
Strength of schedule matters because a +11 point differential against SEC opponents is not the same as a +11 differential against the Sun Belt. To adjust for this, you can use a simple iterative method: calculate each team’s initial rating based on raw point differential, then adjust each team’s rating based on the average rating of their opponents. Repeat this process three or four times, and the ratings will converge on values that account for the difficulty of each team’s schedule. It sounds complex, but in a spreadsheet, it’s just a few columns of formulas that reference each other.
For bettors who want to go a step further without dramatically increasing complexity, replace raw points with yards per play on offense and defense. Yards per play is less influenced by field position, turnovers, and garbage time than raw scoring, making it a more stable predictor of future performance. A team that averages 6.2 yards per play on offense and allows 4.8 on defense is performing at a high level regardless of whether the scoreboard reflects it on any given Saturday.
Setting Up the Spreadsheet
Create a spreadsheet with one row per team and the following columns: team name, games played, total points scored, total points allowed, point differential, point differential per game, raw power rating, schedule strength adjustment, and adjusted power rating. For the first few weeks of the season, you’ll have limited data, so many bettors seed their model with preseason projections and gradually phase in actual results as the season progresses.
The raw power rating is simply the point differential per game. If a team has outscored opponents by an average of 8.5 points per game, their raw rating is 8.5. The schedule adjustment modifies this based on opponent quality. A straightforward method is to calculate the average raw rating of each team’s opponents, then add a fraction of that average — typically 50% to 70% — to the team’s own rating. This rewards teams that have performed well against strong competition and penalizes teams that have padded their numbers against weak opponents.
Once you have adjusted ratings for every team, projecting a spread for any matchup is straightforward: subtract the away team’s rating from the home team’s rating, then add your home-field advantage value. For the NFL in 2026, home-field advantage has settled around 1.5 points, down from the historical average of roughly 3 points. For college football, home-field values vary significantly by venue — Death Valley in Baton Rouge is not the same as a half-empty stadium in the MAC — but a reasonable default is 3 points, with adjustments for specific stadiums if your data supports it.
From Ratings to Projected Spreads
With your adjusted power ratings in place, projecting the spread for any game becomes a one-line formula. Suppose your model gives the Bills an adjusted rating of +6.2 and the Dolphins a rating of +2.8, and the game is in Buffalo. The projected spread is: (6.2 – 2.8) + 1.5 (home-field) = 4.9, or Bills -4.9. If the sportsbook line is Bills -3, your model sees value on Buffalo. If the line is Bills -7, your model says the market has overpriced them and Miami is the value side.
The important caveat is that your projected spread is not gospel — it’s a starting point for analysis. A model built on season-long averages doesn’t know that the Bills’ starting left guard was ruled out this morning, or that the Dolphins just traded for a new defensive end who hasn’t been integrated into the scheme yet. Your model gives you a baseline, and your own research adjusts it. The combination of systematic ratings and informed judgment is far more powerful than either one alone.
Track the difference between your projected spread and the sportsbook’s line for every game. This creates what bettors call a “look” — a quantified disagreement between your model and the market. Not every look is a bet. Most experienced modelers set a threshold — typically 1.5 to 2 points of disagreement — before considering a wager. Below that threshold, the difference is within the margin of error for a simple model and isn’t worth the risk. Above it, you’ve found a spot where your analysis diverges meaningfully from the market, and the game deserves deeper investigation.
Calibrating and Improving Over Time
The first version of your model will be imperfect, and that’s fine. Perfection isn’t the goal — improvement is. After every week of games, compare your projected spreads to the actual results. Calculate the mean absolute error (MAE) — the average difference between your projected margin and the actual margin — across all games. For a simple model, an MAE of 12 to 14 points in the NFL is typical. Professional-grade models get this down to 10 to 11 points. Your goal over the course of a season is to reduce your MAE incrementally by refining your inputs and adjustments.
One of the most effective improvements is weighting recent games more heavily than early-season games. A team’s performance in Week 12 is more reflective of their current quality than their performance in Week 1, when the roster and scheme may have been different. A simple recency weight — giving the most recent four games twice the weight of earlier games, for example — can improve your model’s accuracy meaningfully without adding much complexity to the spreadsheet.
Another improvement is splitting your ratings into offensive and defensive components. Instead of a single power rating, assign each team an offensive rating (based on points or yards generated) and a defensive rating (based on points or yards allowed). This allows your model to capture matchup dynamics — a strong offense against a weak defense should produce a wider spread than two average units meeting — and adds a dimension of analysis that a single composite number misses.
As you accumulate data across multiple seasons, you can also begin to evaluate your model’s performance against the closing line rather than against the final score. Closing line performance is a better measure of model quality because the closing line already accounts for the most efficient price the market produced. If your model consistently disagrees with the market in profitable directions — your projected underdogs outperform the closing spread more often than chance would predict — your model is capturing real information. If it doesn’t, you need to revisit your inputs or your methodology.
Where the Spreadsheet Ends and Judgment Begins
A power rating model is a tool, not an oracle. It processes the data you give it and produces a number. That number reflects average performance under average conditions, and it says nothing about the specific circumstances of this week’s game — the weather, the motivation, the injury nobody reported yet, the coaching chess match between two staffs that know each other intimately from years in the same division.
The best use of a spreadsheet model is as a filter. It tells you which games are worth your deeper attention by flagging the largest disagreements between your numbers and the market’s numbers. Instead of trying to handicap all sixteen NFL games each week, your model narrows your focus to the three or four games where your analysis diverges most from the line. You then apply your judgment, research, and situational analysis to those games specifically, making informed decisions about whether the disagreement reflects a genuine edge or a limitation of your model.
Over time, your model becomes an extension of your thinking. You’ll start to notice patterns — your model consistently undervalues teams after a bye, or overvalues teams with new starting quarterbacks — and you’ll adjust accordingly. Each adjustment is a lesson encoded into a formula, and the formula remembers the lesson even when your memory gets clouded by last week’s results. That’s the real power of the spreadsheet: not that it knows more than you, but that it doesn’t forget what you’ve already learned.
