NGR Driver Decomposition

The bridge before the tree

Decomposition is worthless if the top-line number is ambiguous. Write the bridge from stakes to net revenue explicitly, and make it the only one:

Line	Definition	Where it goes wrong
Turnover	Total stakes placed	Recycled winnings inflate it; not comparable across product types
GGR	Stakes minus wins	Period cut-off and late settlement re-state closed periods
− Bonus cost	Consumed bonus value, not granted — see bonus economics	Granted value used as a proxy; overstates the deduction
− Gaming taxes and duties	Per market, on the locally defined base	Applied on a group base that no regulator recognises
− Payment costs	Processing, chargebacks, FX on player money flows	Sits in a different team's P&L and never reaches the revenue view
= NGR	What the business actually keeps from gaming	Two internal definitions coexisting for years

The driver tree

Net revenue for a period reduces to a small number of multiplicative drivers plus the cost lines above:

NGR ≈ active players × bets per active player × average stake × hold − bonus cost − taxes − payment costs

Active players — the volume driver, itself the sum of retained, reactivated and newly converted players. Splitting it that way is usually where the real answer lives.
Bets per active player — engagement frequency, sensitive to product mix, session structure and promotional cadence.
Average stake — player-mix sensitive; a handful of high-stake players can move it without anything changing for anyone else.
Hold — GGR over turnover. A rate, not a decision, and noisy over short periods.
Cost lines — the ones most likely to be measured by a different team on a different definition.

Diagnosing in the right order

The order matters because the wrong entry point produces a plausible story that survives the meeting and fixes nothing.

Step	Question	If yes
1	Did active players fall?	Split retained / reactivated / new, then by market, channel and device. This is a funnel or churn problem, not a margin one.
2	Did frequency fall at stable players?	Look at product mix, session entry points, promotional cadence and any recent lobby change.
3	Did average stake fall?	Check player mix first — usually a shift in who is active, not a change in behaviour.
4	Did hold fall?	Test variance before narrative: game mix shift, configuration change, or a few large outcomes in a short window.
5	Did a cost line rise?	Bonus cost, tax base, payment mix. Reconcile against the owning team's number before concluding.

Run the steps in sequence and stop at the first one that explains the magnitude. If none of them does, the answer is usually a definition change or a data-pipeline event — check what shipped, not what players did.

Segments that hide inside the average

Group-level NGR is an average of populations that behave nothing alike. Before concluding anything, cut by:

Market — tax base, payment mix and competitive pressure differ enough to reverse a conclusion.
Product — casino, sportsbook and prediction markets have structurally different hold and volatility.
Player tenure — a new-player cohort and a two-year cohort have different economics in every line of the tree.
Acquisition source — see media buy; a decline is often one channel, not the business.

Make it reproducible

The decomposition is only useful if two analysts produce the same numbers. That requires the definitions to live in one dictionary, the tree to be built on one dataset with a stated grain, and late-arriving data to be handled by an explicit re-statement rule rather than by whoever refreshed the dashboard last. Everything else is negotiation dressed as analysis.

Continue reading: iGaming glossary — the metric definitions this tree assumes. The Turbo Stars platform — reporting built on one set of definitions.

NGR drivers — common questions

What is the difference between GGR and NGR?

GGR is stakes minus wins — what the games produced. NGR subtracts the costs booked against that revenue: consumed bonus value, gaming taxes and duties, payment costs, and depending on house policy, platform and content fees. The definitions differ between operators, which is why cross-company NGR comparisons are usually meaningless without the bridge. See GGR and NGR for the definitions.

Why does the same NGR number differ between two internal dashboards?

Almost always one of four causes: a different cost line included in the bridge, a different timezone or accounting cut-off, late-arriving settlements re-stating a closed period, or currency conversion applied at a different point. Fix it in the metric dictionary, not in the dashboards — otherwise the correction lasts until the next dashboard is built.

What is the right order to diagnose an NGR decline?

Volume before rate, and structure before behaviour. Check active players first, then frequency per player, then stake per bet, then hold, then the cost lines. Most declines that get blamed on 'margin' turn out to be fewer active players in one segment or one market, which is a completely different problem with a completely different fix.

Is hold a driver or an outcome?

Both, which is why it is the most misread line in the tree. Hold moves because the game mix moved, because a payout configuration changed, because a small number of large outcomes landed in the period, or because the player mix shifted. Before treating a hold change as a decision, check whether it is simply variance in a period too short to read.

How granular should the driver tree be?

Deep enough that every branch has an owner and shallow enough that the whole tree fits on one screen. A tree nobody can hold in their head does not get used in the meeting where the decision is made. Start with five levels and add depth only where a recurring question keeps hitting a leaf.

Decomposing NGR into drivers.