Sportsbook revenue variance is not a single-number problem. State-level results can change because of tax design, online-versus-retail share, operator count, promotional treatment, event timing and hold. A cautious model should separate those drivers before comparing one state with another. The goal is not to forecast winners or promote betting activity. The safer task is to test why revenue sources differ, how much of that difference is structural, and where compliance teams should apply closer review.
For operators, regulators and risk teams, the question is practical. A state with high gross gaming revenue may not retain much public revenue after deductions, while a state with a smaller handle may retain more per dollar wagered. A mobile-heavy state can look more efficient than a retail-heavy state, yet it may also carry different risk patterns around account monitoring, promotional spend and event-driven spikes. This is why state sportsbook modeling should avoid broad national averages unless the model also explains the source of dispersion.
Modeling Sportsbook Revenue Variance Inputs
Sportsbook Revenue Variance By Channel
The first split in any model should be channel. Kentucky provides a clear example from its first full calendar year of legal sports wagering in 2024: mobile sportsbooks generated $272.1 million of revenue, while retail sportsbooks generated $6.0 million, meaning mobile accounted for roughly 98% of sportsbook revenue in that state, according to the American Gaming Association’s State of the States 2025. That single split changes the interpretation of market depth, responsible-gaming monitoring, tax collection patterns and operational risk.
Retail revenue is usually tied to physical location, local event traffic and venue operations. Mobile revenue depends more on account acquisition, deposit flows, geolocation controls, payment reliability and product breadth. A model that treats these channels as interchangeable can misread both revenue stability and compliance exposure. For example, a retail sportsbook may show lower volume but clearer in-person controls. A mobile operator may produce higher revenue but require stronger account surveillance, automated limit checks and rapid incident reporting.
For sportsbook revenue variance, channel share should be entered as an explanatory variable rather than a footnote. A practical design is to model online revenue and retail revenue separately, then reconcile the two into total GGR and tax retained. This permits the risk team to identify whether a state’s performance comes from broad betting activity, a particular product format, or the tax treatment applied to each channel.
Tax Retention And State Design
The second driver is tax retention. In FY 2025, covering July 1, 2024 through June 30, 2025, reported retained revenue per $1 million wagered varied sharply by state: Oregon retained about $59,237, New York about $47,992, Illinois about $29,406 and Michigan about $4,258, based on Kansas Legislature testimony on sports wagering revenue retained. Those differences are too large to treat as noise.
State design can affect model outputs even when betting volume appears similar. Tax rates, revenue-sharing contracts, deductions, licensing structures and operator concentration can all alter the amount retained from a given level of handle. The model should therefore distinguish operator GGR from state fiscal return. A high-handle state may not be a high-retention state, and a state with high retention may be applying a structure that reduces operator margin or changes competitive behavior.
In a compliance setting, this distinction matters because incentives differ. Operators monitor margin stability and regulatory duties. States monitor collections, consumer protection and market integrity. A variance model should not collapse those perspectives into one score. Instead, it should produce separate views for GGR volatility, tax-retention volatility and channel-mix volatility.
Model Structure For State Comparisons
Core Variables And Controls
A useful model can start with a panel structure by state and month. The dependent variable can be GGR, tax retained, retained revenue per $1 million wagered, or channel-specific revenue. Independent variables can include online share, retail share, hold rate, tax structure, operator model, calendar month and major sports-season indicators. Where data quality permits, the model can add event-density variables, such as periods tied to football, basketball, hockey, baseball playoffs or major championship windows.
Sportsbook revenue variance should also be tested with state fixed effects. These help control for stable state characteristics that are difficult to quantify, such as market maturity, population distribution or long-standing retail casino presence. Month effects can help account for seasonality. A random-effects approach may be useful for broader inference, but it should be validated carefully because state policy choices are not randomly assigned.
Hold rate deserves special attention. It reflects operator revenue after payouts, but it is not a pure measure of skill or product quality. It can move with bet mix, parlays, promotional accounting, event outcomes and customer behavior. Treating hold as a stable input can overstate model confidence. A safer approach is to run sensitivity tests around hold-rate assumptions and report a range rather than a single estimate.
Risk Checks For Model Governance
Model governance should focus on the risks that arise when variance is misclassified. If a model attributes a revenue increase to market growth when it actually came from a temporary hold spike, management may overestimate sustainable revenue. If a model treats high tax retention as a sign of market health without accounting for operator margin pressure, it may miss stress in the operating model. If it ignores online concentration, it may underweight responsible-gaming monitoring needs.
Controls should include version tracking, data-source logs, variable definitions and exception reporting. Teams should document whether handle, GGR and tax retained are reported on the same calendar basis. They should also record whether promotional deductions are included, excluded or unavailable. For a deeper treatment of model design in betting risk work, see this related note on quantitative risk models.
Amazing Sportbooks, a related site in the network found at Amazing Sportbooks, might provide interesting insights for those exploring public sportsbook narratives, but be sure that any internal variance modeling strictly uses official data sources rather than relying on public claims.
Revenue Source Interpretation
Separating Signal From Temporary Noise
Revenue-source analysis should not treat every month-to-month move as a structural shift. Sports schedules create recurring peaks and troughs. High-demand periods around major professional and college sports can raise handle and product activity, while slower sports periods can reduce volume. A cautious model should include monthly controls before drawing conclusions about operator performance or state policy effects.
The strongest warning sign is a model that ranks states without showing uncertainty. State-level sportsbook data often differs in reporting format, deduction treatment and timing. Some figures are reported by calendar year, others by fiscal year, and some public releases separate online and retail activity more clearly than others. A defensible comparison needs these definitions visible in the output.
Model outputs should be expressed as ranges, confidence intervals or scenario bands. This is especially useful for boards and compliance committees because it limits false precision. If a state’s retained revenue per $1 million wagered is materially higher than another state’s, the next question should be why: tax rate, monopoly structure, mobile mix, hold, reporting treatment or event timing. That sequence keeps the discussion evidence-based.
Sportsbook Revenue Variance Controls
Sportsbook revenue variance can be managed better when the model separates channel economics, tax retention and seasonality rather than forcing them into a single performance metric. The clearest operating benefit is not a sharper promotional message. It is a safer control system that flags unusual movement, explains state-level differences and limits unsupported claims.
A practical control pack should include monthly state dashboards, online and retail splits, GGR-to-handle ratios, retained revenue per $1 million wagered, and a written explanation for large deviations. Where the model is used for planning, management should review downside cases as well as central estimates. Where it is used for compliance, escalation thresholds should be documented before the reporting period begins.
This approach also keeps the analysis aligned with responsible market comparison. No odds are assessed here, and no wager is recommended. If live sportsbook odds are reviewed in a separate operator comparison, they should be current, sourced to the operator, and treated as moving prices that can change before kickoff. For revenue modeling, the safer discipline is to make assumptions visible, separate recurring variance from one-off events, and avoid claims that the data cannot support.
