Mar 2, 2026 23 MIN READ AI ANALYSIS

NHL Predictions: The Definitive Guide to Using AI and Data Analytics for Smarter Hockey Betting in 2026

Discover how AI-powered NHL predictions are transforming hockey betting nationwide. Learn data-driven strategies used by sharp bettors across the country in 2026.

NHL Predictions: The Definitive Guide to Using AI and Data Analytics for Smarter Hockey Betting in 2026

Table of Contents

Quick Answer: What Are NHL Predictions?
Frequently Asked Questions About NHL Predictions
What Are NHL Predictions and Why Do They Matter?
How AI-Powered NHL Predictions Actually Work
Types of NHL Predictions: Markets and Bet Categories
Benefits of Using Data-Driven NHL Predictions
How to Choose the Right NHL Prediction Model or Service
Real Examples: NHL Predictions in Action
Getting Started With AI-Driven NHL Predictions
Key Takeaways
Related Articles

Quick Answer: What Are NHL Predictions?

NHL predictions are data-driven forecasts of hockey game outcomes — moneylines, puck lines, totals, and prop bets — generated by analyzing team performance metrics, goaltender stats, situational factors, and historical trends. Modern AI prediction models process thousands of variables per game, including Corsi, expected goals (xG), and real-time lineup changes, to produce probability-weighted picks that consistently outperform gut-feel handicapping.

Frequently Asked Questions About NHL Predictions

How accurate are AI-generated NHL predictions?

Top-tier AI models for NHL predictions typically achieve 56–62% accuracy on moneyline selections across a full season sample. That may sound modest, but in a sport where the home team wins roughly 55% of the time and bookmaker vig erodes margins, consistently hitting above 55% translates to long-term profitability. The key is volume and discipline — a 58% win rate over 500 tracked bets produces meaningful ROI.

What data do NHL prediction models use?

Quality NHL prediction models ingest shot attempt metrics (Corsi, Fenwick), expected goals (xG), power play and penalty kill efficiency, goaltender save percentages and recent workload, travel schedules, back-to-back game indicators, head-to-head records, line combinations, and real-time injury reports. The best models also factor in score-adjusted possession data and venue-specific effects like altitude and ice quality.

Can you trust free NHL predictions?

Free NHL predictions vary wildly in quality. Some are backed by legitimate models and serve as teasers for premium services, while others are little more than guesswork dressed up as analysis. Look for transparency — any trustworthy source will explain their methodology, publish verifiable track records, and show unit-based ROI rather than cherry-picked highlights.

How is hockey different from other sports for betting predictions?

Hockey's lower scoring makes individual game outcomes more volatile than basketball or football. A single unlucky bounce or hot goaltender performance can override statistical advantages. This means NHL predictions must account for variance more aggressively, and bettors need larger sample sizes to validate model accuracy. However, that same volatility creates more mispriced lines, which is where disciplined, data-driven bettors find edge.

What is the best NHL prediction strategy for beginners?

Start with moneyline bets on slight underdogs where your model shows positive expected value. Avoid parlays until you understand single-game variance. Use flat betting (1–2% of your bankroll per wager) and track every bet in a spreadsheet. Focus on games where your model's probability diverges from the implied odds by at least 5 percentage points. This approach — similar to principles we cover in our guide to consensus picks — builds discipline early.

Do NHL predictions work for live betting?

Yes, but live NHL predictions require models that update in real-time. In-game momentum shifts — a pulled goaltender, a five-minute major penalty, or a quick two-goal swing — dramatically alter win probabilities within seconds. AI models that process live shot data and game state can identify mispriced live lines faster than bookmakers adjust them, especially during the second intermission and in late third-period scenarios.

How does goaltender performance affect NHL predictions?

Goaltending is the single most impactful variable in NHL predictions. A starting goaltender's recent save percentage, goals saved above expected (GSAx), and workload (starts in the last seven days) can swing a game's predicted outcome by 8–15 percentage points. Late-breaking goaltender confirmations — often announced just 30–60 minutes before puck drop — are one reason models that update in real-time have a significant edge.

Should I use NHL predictions for parlays?

Parlays amplify both edge and variance. If your NHL prediction model identifies three games with genuine positive expected value, combining them in a parlay increases potential payout but also increases the probability of losing. A two-leg NHL parlay with 58% individual win rates has roughly a 33.6% hit rate — profitable at standard parlay odds, but requiring more patience and bankroll buffer. For a deeper look at multi-leg strategy, check out our breakdown of MLB picks and parlays, which applies the same mathematical framework.

What Are NHL Predictions and Why Do They Matter?

The National Hockey League is one of the most analytically rich — and most misunderstood — sports betting markets in North America. With 1,312 regular-season games, a 16-team playoff field, and nightly action from October through June, the NHL offers a massive volume of betting opportunities. Yet compared to the NFL, NBA, and MLB, hockey betting receives a fraction of the public attention, which is precisely what makes it attractive for data-driven bettors.

NHL predictions encompass any forecasted outcome for a hockey game or related market. At their simplest, they answer "who wins tonight?" At their most sophisticated, they assign probability distributions across every possible scoreline, project individual player performance, and identify misalignments between those projections and the odds offered by sportsbooks.

The modern NHL predictions landscape has been transformed by the same data revolution reshaping every major sport. The NHL's partnership with tracking technology providers now generates granular puck and player tracking data at 200 frames per second, capturing skating speed, shot velocity, passing accuracy, and positioning for every player on the ice. This data feeds a new generation of prediction models that go far beyond traditional box-score statistics.

Why does this matter for bettors? Because the NHL betting market is less efficient than you might think. According to research from the UNLV International Gaming Institute, hockey lines receive roughly one-fifth the handle of NFL and NBA lines, meaning bookmakers devote less resources to sharpening them. The result is more frequent and larger mispricings — exactly the conditions where good NHL predictions create exploitable edge.

At BetCommand, we've built our NHL prediction engine around this principle: the gap between what public bettors believe and what the data actually shows is where profit lives. Understanding that gap starts with understanding how modern prediction models work.

In the 2024-25 NHL season, games decided by one goal accounted for 48.3% of all results — making hockey the most variance-heavy major North American sport and the one where disciplined, model-driven betting creates the widest edge over casual bettors.

How AI-Powered NHL Predictions Actually Work

Building accurate NHL predictions requires layering multiple data sources, weighting them appropriately, and constantly recalibrating as the season progresses. Here's a simplified breakdown of what happens inside a modern AI prediction engine.

Step 1: Data Ingestion and Feature Engineering

The model starts by collecting raw data from multiple sources: official NHL play-by-play feeds, player tracking data, injury reports, lineup confirmations, travel schedules, and historical results. This raw data is then transformed into predictive features — measurable variables that correlate with game outcomes.

Key features for NHL predictions include:

Expected goals (xG): A shot-by-shot model that assigns a scoring probability based on shot location, type, angle, preceding events, and game state. A team's xG differential over their last 10–20 games is one of the strongest predictors of future performance.
Corsi and Fenwick percentages: Shot attempt metrics that measure territorial dominance. A team controlling 53%+ of shot attempts at 5-on-5 is generating more opportunities regardless of whether those shots are going in.
Goaltender GSAx (Goals Saved Above Expected): Measures how many goals a goaltender has prevented relative to the quality of shots faced. This separates goaltending skill from team defense quality.
Special teams efficiency: Power play conversion rates and penalty kill success rates, weighted toward recent performance (last 15–25 games) rather than full-season averages.
Situational factors: Back-to-back games, travel distance, time zone changes, home-ice advantage (worth roughly 2.5–3.5 percentage points in win probability), and rest days.

Step 2: Model Training and Calibration

The AI model — typically an ensemble combining gradient-boosted trees, neural networks, and logistic regression — trains on multiple seasons of historical data. It learns which feature combinations most reliably predict outcomes and assigns dynamic weights that shift throughout the season.

Early in the season (October through mid-November), the model leans heavily on prior-year performance and preseason roster changes because the current-year sample is too small. By December, current-season data starts carrying more weight. By February, the model is almost entirely driven by this season's metrics, with historical data serving only as a stabilizing anchor.

Step 3: Probability Output and Line Comparison

The model outputs a win probability for each team — say, 57.3% for the home team and 42.7% for the visitor. This probability is then converted to implied odds and compared against the actual sportsbook lines. If the book offers the home team at -145 (implying a 59.2% win probability), the model sees no value. But if the book offers them at -120 (implying 54.5%), there's a 2.8 percentage point edge — enough to warrant a bet.

This edge calculation is the foundation of every serious NHL prediction system. It's not about picking winners — it's about finding prices that understate a team's true probability of winning.

For a deeper dive into how AI models identify edge across different sports, read our guide on how AI and data analytics are reshaping football betting, which covers the same probability framework applied to a different sport.

Step 4: Real-Time Updates

The best NHL prediction systems don't lock in their projections the night before. They continuously update as new information arrives: confirmed starting goaltenders (typically announced at morning skate or 30–60 minutes pre-game), late scratches, line combination changes, and even warm-up reports. A late goaltender switch — say, a backup getting the surprise start — can shift a game's win probability by 10+ percentage points, and models that capture this in real-time generate significantly more edge than static morning-line projections.

Types of NHL Predictions: Markets and Bet Categories

NHL predictions span a wide range of markets, each with its own dynamics and analytical requirements. Understanding which markets favor model-driven approaches is critical for allocating your betting capital.

Moneyline Predictions

The most straightforward market: pick the winner. NHL moneylines typically range from -250 (heavy favorite) to +200 (significant underdog), with most games featuring lines between -150 and +130. AI models excel here because they can precisely quantify small edges that casual bettors overlook — like the impact of a third-string goaltender or a team playing its fourth game in six nights.

Puck Line Predictions

The NHL equivalent of a point spread. The standard puck line is -1.5 for the favorite and +1.5 for the underdog. Because 23–25% of NHL games are decided by exactly one goal, the puck line creates sharply different risk/reward profiles than the moneyline. Models that accurately predict blowout probabilities (games decided by 2+ goals) find consistent value in puck line markets — particularly when backing underdogs at +1.5 with enhanced plus-money odds.

Over/Under (Totals) Predictions

Predicting whether the combined score will exceed or fall below a posted total (usually 5.5 or 6.0 goals). Totals predictions require modeling offensive output, defensive structure, goaltending quality, and pace of play. The interaction between two specific goaltenders and two specific offensive systems matters enormously. For context on how AI approaches totals betting in another sport, see our analysis of over/under betting in MLB, which shares the same analytical foundation.

Period Betting

Wagering on the outcome of individual periods. First-period betting is particularly popular because it offers a fresh "mini-game" with its own dynamics. Teams vary significantly in first-period performance — some are notoriously slow starters while others dominate early and fade. NHL prediction models that track period-specific metrics uncover edges invisible to bettors looking only at full-game statistics.

Player Props

Projecting individual player performance: goals, assists, shots on goal, saves, and more. Connor McDavid's shot volume in a given game depends on matchup, game state, power play time, and opponent defensive structure. AI models that simulate game flow and project time-on-ice allocations can generate sharp player prop predictions that outperform sportsbook projections.

Futures and Series Predictions

Longer-term markets like Stanley Cup winner, division champions, and playoff series outcomes. These require projecting roster health, trade deadline acquisitions, and schedule difficulty over weeks or months. NHL predictions for futures markets are inherently uncertain, but models that correctly weight goaltender injury risk and playoff-specific performance patterns find exploitable prices, particularly in the weeks before the trade deadline.

See our complete breakdown of how AI is changing basketball betting for a parallel look at how these same market types are analyzed in the NBA.

Benefits of Using Data-Driven NHL Predictions

1. Eliminating Cognitive Bias

Human bettors consistently overvalue recent results, name-brand players, and national television games. A team that loses 6-1 on Monday night looks terrible, but if they dominated possession 58-42% and their goaltender had a historically bad night, the underlying performance was strong. AI-driven NHL predictions strip away emotional noise and evaluate what actually happened beneath the scoreboard.

2. Processing Volume Humans Cannot Match

A typical NHL game generates over 200 measurable events — shots, passes, zone entries, faceoffs, hits, and more. Across a 32-team league playing 1,312 regular-season games, that's over 260,000 events per season. No human analyst can track, weight, and synthesize that volume. AI models process it in seconds.

3. Identifying Value in Thin Markets

NHL betting lines receive less public action than NFL or NBA lines, which means oddsmakers sometimes set initial lines based on simplified models and adjust primarily in response to sharp money. AI prediction systems that generate their own independent probabilities routinely find 2–4 percentage point edges on opening lines — edges that narrow or disappear as game time approaches.

4. Capturing Goaltender-Driven Edge

As mentioned earlier, goaltending is uniquely impactful in hockey. Casual bettors know who the star goalies are, but they often misjudge the impact of workload, recent performance trends, and matchup-specific data. A goaltender facing a team that generates high-danger chances at an above-average rate requires a different projection than one facing a perimeter-shooting opponent. Models capture this nuance automatically.

5. Adapting to Schedule and Travel

The NHL schedule is grueling: 82 games across roughly 185 days, with frequent cross-continent travel. Teams playing the second game of a back-to-back lose approximately 4–5% more often than rest-adjusted baselines suggest. Road teams crossing two or more time zones perform measurably worse. AI NHL predictions bake these factors in by default — something most casual bettors ignore entirely.

6. Quantifying Playoff Adjustments

Playoff hockey is a different sport. Referees call fewer penalties, physical play intensifies, and goaltending variance increases in lower-scoring games. NHL prediction models trained on regular-season data must recalibrate for postseason dynamics. The best systems apply playoff-specific adjustments — reduced power play opportunities, tighter checking, and increased save percentages — rather than blindly extrapolating regular-season projections.

7. Building Systematic Bankroll Growth

Perhaps the most important benefit: data-driven NHL predictions transform betting from entertainment into a structured process. Instead of chasing losses or doubling down after a win, model-driven bettors place consistent, calculated wagers based on quantified edge. Over a full season, this systematic approach — combined with disciplined bankroll management — is how recreational bettors graduate to consistent profitability.

A bettor placing 400 NHL moneyline wagers per season at a 57% win rate with average odds of -115 generates approximately 8.2% ROI — turning a $5,000 bankroll into $5,410 in pure profit, compounding season over season with zero guesswork.

How to Choose the Right NHL Prediction Model or Service

Not all NHL prediction services are created equal. Here's a decision framework for evaluating them, whether you're considering a paid subscription, a free tipster, or building your own model.

Track Record Transparency

The single most important criterion. Any credible NHL prediction service publishes a full, verifiable betting history — not just highlights. Look for:

Unit-based ROI over at least one full season (500+ tracked bets)
Timestamped picks published before game time (not retroactively edited)
Flat-stake results rather than variable-sizing that can be gamed

If a service only shows win/loss percentage without specifying odds, that's a red flag. A 60% win rate on -200 favorites is losing money.

Methodology Disclosure

You don't need to see the source code, but a trustworthy service explains what data it uses, how it generates projections, and what types of edges it targets. "Our proprietary algorithm" with zero further detail is a marketing tactic, not a methodology. The best NHL prediction providers publish regular transparency reports explaining model performance, strengths, and weaknesses.

Specialization vs. Generalization

Some services cover every sport; others specialize in hockey. Specialization matters in the NHL because hockey analytics are more niche than basketball or football analytics. A service that deeply understands xG models, goaltender evaluation, and schedule effects will generally outperform a generalist covering 10 sports with the same model architecture.

Update Frequency

As discussed earlier, late-breaking information — particularly goaltender confirmations — is critical for NHL predictions. A service that publishes its picks at 9 AM and never updates them is leaving edge on the table. Look for services that provide real-time updates or at minimum refresh their projections after morning skate reports.

Bankroll Management Guidance

Edge identification is only half the equation. A good NHL prediction service also provides guidance on bet sizing, bankroll allocation, and risk management. Models that specify confidence levels (high, medium, low) or Kelly Criterion-derived unit sizes help you extract maximum value without overexposing your bankroll.

At BetCommand, we evaluate prediction quality using all five of these criteria — and we apply the same rigor to our own model that we'd expect from any service we recommend.

Real Examples: NHL Predictions in Action

Abstract concepts become concrete with real-world scenarios. Here are examples illustrating how data-driven NHL predictions identify value that traditional handicapping misses.

Example 1: The Back-to-Back Goaltender Trap

Situation: The Colorado Avalanche host the Chicago Blackhawks. Colorado is a -210 moneyline favorite. Casual bettors pile on the Avs — elite team, home ice, weak opponent.

What the model sees: Colorado played in Dallas the previous night. Their starting goaltender made 34 saves in that game. The backup goaltender, with a .898 save percentage over his last 10 starts, is confirmed for tonight. The model adjusts Colorado's win probability from 65% to 55.8%. At -210 (implying 67.7%), the model identifies strong value on Chicago +180.

Outcome reasoning: This isn't a guarantee Chicago wins — Colorado is still the better team. But the price is wrong by over 10 percentage points, making Chicago a clear positive-expected-value bet. Over hundreds of similar situations, backing the underdog in these spots generates consistent profit.

Example 2: The Mid-Season Regression Candidate

Situation: A team has a 12-3-1 record through October and early November. They're being priced as a top-5 team in the league. The public sees a juggernaut.

What the model sees: Their 5-on-5 expected goals percentage is 48.2% — below average. Their shooting percentage is 11.3%, well above the league average of ~9.8%. Their goaltender is posting a .938 save percentage, far above his career .912 average. The model identifies this team as a regression candidate whose record will normalize over the next 20–30 games.

Application: The model systematically fades this team when they're priced as heavy favorites, knowing their underlying metrics don't support their record. This is the same principle that drives MLB predictions — separating process from results to identify teams due for regression.

Example 3: Playoff Series Pricing

Situation: A President's Trophy-winning team is priced at -280 to win their first-round series against a wild card opponent.

What the model sees: The wild card team has the league's third-ranked goaltender by GSAx, and historical data shows that elite goaltending in the playoffs overperforms regular-season projections by a wider margin than any other single variable. The model prices the series closer to -180, identifying meaningful value on the underdog at +230.

Historical basis: Since the 2005 lockout, teams with a top-5 GSAx goaltender have won first-round series as underdogs at a 39.2% rate — significantly higher than their implied odds in most series prices. This is a well-documented structural edge in NHL predictions.

Example 4: The Public Overreaction Game

Situation: The Toronto Maple Leafs lose 7-2 on Hockey Night in Canada. The next night, they host the Senators. The line opens at -150 and moves to -130 as public money backs Ottawa, influenced by recency bias from the blowout loss.

What the model sees: Toronto's five-game rolling xG differential remains elite at +0.8 per game. Their starting goaltender has a .925 save percentage and is well-rested. The 7-2 loss was a statistical outlier — a .890 team save percentage in a single game when their 20-game rolling average is .918. The model prices Toronto at -175, making -130 a significant value opportunity.

This pattern — fading public overreaction to blowout results — is one of the most reliable edges in NHL predictions and mirrors the public betting dynamics we track across multiple sports.

Example 5: The Trade Deadline Bounce

Situation: A contending team acquires a top-six forward at the trade deadline. Their next game's moneyline barely moves because the market hasn't fully priced in the acquisition.

What the model sees: The acquired player's on-ice xG impact over the prior three seasons projects to a +0.3 goal differential improvement per 60 minutes. Inserted into the team's second line, the model projects a 2.1 percentage point increase in the team's game-by-game win probability — a small but real edge that compounds over the remaining 20+ regular-season games and into the playoffs. Betting this team's moneyline in the 3–5 games following the trade deadline, before the market fully adjusts, has shown positive ROI historically.

Getting Started With AI-Driven NHL Predictions

Ready to move from gut-feel hockey betting to a structured, data-driven approach? Here's your step-by-step roadmap.

Step 1: Build Your Data Foundation

Before placing a single bet, familiarize yourself with the core metrics that drive NHL predictions. Bookmark these free resources:

Hockey-Reference for historical statistics, game logs, and advanced metrics
Natural Stat Trick for 5-on-5 shot metrics, xG, and situational data
The NHL's official statistics page for real-time standings, schedules, and player tracking data

Spend at least two weeks studying these metrics before wagering. Understand what Corsi, Fenwick, xG, and GSAx mean and why they matter.

Step 2: Define Your Bankroll and Staking Plan

Set aside a dedicated bankroll — money you can afford to lose without financial stress. A common starting bankroll for NHL betting is $500–$2,000. Use flat staking at 1–2% per bet (so $5–$20 per wager on a $1,000 bankroll). This protects you through inevitable losing streaks while allowing compound growth during winning runs.

Step 3: Choose Your Markets

Start with moneyline bets on games where your analysis diverges from the posted odds by at least 5 percentage points. Avoid parlays, period bets, and props until you've logged at least 100 single-game bets and verified your edge. Simplicity reduces variance and makes it easier to diagnose what's working.

Step 4: Track Everything

Log every bet with the date, matchup, market, odds, stake, model probability, book implied probability, and result. After 200+ bets, analyze your results by market type, favorite vs. underdog, home vs. away, and edge magnitude. This data tells you where your model — or your interpretation of it — performs best.

Step 5: Use AI Tools to Scale

Once you've validated your process manually, consider AI-powered tools to automate data collection, model execution, and line comparison. BetCommand's NHL prediction engine handles all of this in real-time, letting you focus on bet selection rather than data wrangling. The same analytical framework powers our predictions across hockey, football, basketball, and baseball — check out our NFL picks guide and NBA picks guide to see how the methodology translates across sports.

Step 6: Stay Disciplined Through Variance

Hockey is a high-variance sport. You will experience 10-game losing streaks even with a 58% model. This is mathematically inevitable, not a sign your approach is broken. Trust the process, maintain your staking plan, and evaluate performance over rolling 200+ bet samples, never individual nights. As explored in our consensus picks guide, combining your model's output with crowd wisdom data can provide additional confirmation and confidence during rough patches.

Key Takeaways

NHL predictions powered by AI models process thousands of variables per game — shot metrics, goaltender performance, schedule effects, and more — to generate probability estimates that identify mispriced lines.
Goaltending is the single most impactful variable in hockey predictions. Late-breaking goaltender confirmations can shift win probabilities by 10+ percentage points.
The NHL betting market is less efficient than the NFL or NBA, creating more frequent and larger edges for data-driven bettors willing to do the analytical work.
A 56–60% moneyline win rate generates consistent long-term profit when combined with flat staking and disciplined bankroll management.
Start simple with moneylines, track every bet meticulously, and expand to puck lines, totals, and props only after validating your edge over 100+ bets.
Variance is not your enemy — it's your opportunity. The same game-to-game unpredictability that makes hockey exciting is what creates the mispriced lines that fund profitable betting systems.
Always compare your model's probability against the implied odds — the goal isn't to pick winners, it's to find prices that understate a team's true win probability.
Real-time updates are essential. Static morning predictions leave significant edge uncaptured, especially around goaltender announcements and late lineup changes.

Explore more data-driven betting guides from BetCommand:

The Ultimate Guide to NFL Picks — How AI and data analytics are reshaping football betting, with the same probability framework applied to NFL markets.
The Complete Guide to NBA Picks — AI-driven basketball predictions covering spreads, totals, and player props across the NBA season.
Consensus Picks Explained — How to use crowd wisdom and public betting data to sharpen your picks across all sports, including hockey.
The Complete Guide to MLB Picks — A comprehensive look at AI-powered baseball predictions and how data analytics identify value in MLB betting lines.
MLB Predictions — How AI-powered models are changing the way bettors approach baseball, with lessons that translate directly to hockey.
Over/Under Betting in MLB — How AI models analyze totals markets — the same xG-based framework used in NHL totals predictions.
MLB Public Betting — Using crowd data to identify contrarian opportunities, a strategy equally powerful in NHL markets.
The Definitive Guide to MLB Betting — Markets, metrics, and season-long strategy for data-driven bettors — applicable principles for any sport.
Football Tips for Today — A step-by-step system for picking winners that mirrors the structured approach we recommend for NHL betting.
Free Soccer Tips — A data-driven guide to smarter match predictions, showcasing how AI models evaluate low-scoring sports like hockey and soccer through similar statistical lenses.

Start Making Smarter NHL Predictions Today

The difference between recreational hockey bettors and profitable ones isn't luck — it's process. Every data point, every goaltender confirmation, every schedule quirk that most bettors ignore is an opportunity for those equipped with the right analytical tools.

BetCommand's AI-powered prediction engine processes the full depth of NHL data — from shot-level xG models to real-time lineup updates — and delivers actionable, probability-weighted picks designed to identify value others miss. Whether you're placing your first NHL bet or your five-hundredth, the platform gives you the same analytical edge that sharp bettors have relied on for years.

Stop guessing. Start predicting. Explore BetCommand's NHL predictions and see how data-driven hockey betting actually works.

Written by the BetCommand analytics team. BetCommand is a trusted AI-powered sports predictions and betting analytics platform serving bettors across the United States. Our models are built on peer-reviewed sports analytics research and validated against multi-season historical data.