Where the arguments actually come from
A meeting is convened to discuss why active player numbers differ between two reports. An hour later the conclusion is that one report counts anyone who placed a bet and the other requires a deposit in the period.
That is not an analytical problem. It is a definitional one, and it accounts for a large proportion of the time operators spend arguing about numbers.
This lesson sets out the core measures precisely and identifies where the divergences occur, on the basis that agreeing definitions once is considerably cheaper than reconciling their consequences repeatedly.
Volume and revenue measures
Established in the iGaming Basics course and restated here with the definitional traps.
Turnover, also called handle or amount staked, is the total value wagered. The trap is that casino turnover includes restaked winnings, which makes it enormous relative to money entering the business and makes casino and sportsbook turnover figures non-comparable.
Gross gaming revenue is turnover less winnings paid. The traps are the treatment of bonus-funded play, whether jackpot contributions are included, and the timing of unsettled bets.
Net gaming revenue deducts bonus cost and, depending on the definition, gaming duty, payment fees and platform costs. There is no universal definition, and this is the single most commonly divergent metric in the industry. Any NGR figure should carry a statement of what was deducted.
Contribution deducts the directly variable costs of serving customers. It answers whether the relationship is worthwhile, which revenue does not, and it requires cost attribution at customer level that many operators lack.
Hold is GGR as a proportion of turnover, meaningful for sportsbook and largely a description of blended house edge for casino.
The practical guidance is that any revenue figure should be accompanied by its level in this stack. A number described simply as revenue is ambiguous.
Player measures
Registrations counts accounts created. Traps include whether unverified registrations count and whether duplicates are removed.
First time depositors counts customers making an initial deposit. The trap is the treatment of customers who deposit, fail verification and are refunded.
Active players counts customers meeting an activity threshold in a period. The threshold is a choice, and it is where divergence most commonly occurs. One bet? A deposit? Activity on more than one day? A spend threshold? Each is defensible and they produce different numbers.
The related trap is the period. Monthly actives, weekly actives and daily actives measure different things, and an operator quoting actives without specifying the period is quoting nothing.
Average revenue per user divides revenue by actives. Covered extensively elsewhere in these courses and worth restating: in a skewed distribution this describes nobody, and it should be accompanied by the median and the distribution.
Retention measures the proportion of a group still active after a period. The traps are which group, measured from what starting point, active by which definition, and over what horizon. Retention figures without all four specified are not comparable to anything.
Churn is the inverse and carries the same traps, plus the definitional question covered in the CRM material: whether churn is a fixed threshold or is relative to the individual's pattern.
Lifetime value estimates total future contribution. The traps are whether it is measured on revenue or contribution, over what horizon, and whether it is observed or predicted, since operators frequently present a model output as though it were a measurement.
Cost and efficiency measures
Customer acquisition cost divides acquisition spend by customers acquired. The traps are extensive: whether bonus cost is included, whether it is blended across channels or calculated per channel, which spend counts as acquisition, and whether the denominator is registrations or depositors.
CAC excluding bonus cost is a flattering figure that describes nothing useful, and it appears more often than it should.
Bonus cost should be net of the wagering generated rather than the awarded value, as the CRM material established. Operators reporting awarded value are reporting a figure several times the real one.
Payback period measures how long a cohort takes to repay its acquisition cost. The traps are whether it is measured on revenue or contribution and whether the acquisition cost includes everything.
Cost per acquisition by channel requires attribution, which as established is the least reliable data layer, and channel-level figures should carry that caveat.
Product and operational measures
Session requires a definition, since a session is a construct rather than a recorded event. The usual approach is activity separated by a gap exceeding a threshold, and the threshold is a choice.
Session length inherits that, plus the question of whether idle time within a session counts.
Conversion rate requires specifying from what to what, since registration to deposit, visit to registration and deposit attempt to success are all called conversion.
Acceptance rate requires the definitions set out in the Payment Operations course, particularly whether operator-side rejections are included alongside issuer declines.
Contact rate divides support contacts by active customers and requires both terms defined.
Building a metric dictionary
The infrastructure that resolves all of this, and the highest-return analytical investment available.
A metric dictionary records, for each measure: the name, the plain-language definition, the calculation logic, the authoritative source, the owner, the known limitations, and the variants where more than one legitimately exists.
The last item matters. Some metrics genuinely need more than one version. Finance may require a revenue figure on an accounting basis while marketing needs one on an activity basis. The dictionary's job is to name them distinctly rather than to pretend one exists.
Making it work requires a few things.
A single owner for the dictionary, empowered to arbitrate.
Implementation in the semantic layer, so that the definition is enforced in the tooling rather than depending on each analyst applying it.
Change control, since definitions do change and an undocumented change makes historical comparison invalid.
Discoverability, meaning anyone can find what a metric means without asking.
Enforcement in reporting, so that a figure appearing in a report is calculated from the agreed definition.
The argument for the investment is straightforward: an operator with agreed definitions spends its analytical capacity answering questions, and one without spends a substantial proportion of it reconciling numbers and resolving disputes about which figure is correct.
Choosing the right measure
A closing point, since precision about definitions does not by itself produce good analysis.
The right metric depends on the question.
For assessing whether acquisition is working, cohort contribution against acquisition cost, not registrations.
For assessing product changes, retention and contribution over a meaningful horizon, not immediate conversion.
For assessing customer base health, cohort retention curves and value distribution, not active player counts.
For assessing promotional efficiency, incremental contribution against net bonus cost, not uptake.
For assessing operational performance, resolution and repeat contact, not handling time.
For assessing exposure, concentration measures, which appear in no standard performance report.
The recurring pattern is that the readily available measure describes activity and the useful measure describes value, and that the second requires more work and a longer horizon.
An analyst's contribution is frequently to redirect a question from the first to the second, which is more useful than answering the question as asked and is less welcome, since it takes longer and produces a less flattering answer.
A worked definitional dispute
To demonstrate how these arise, a realistic example.
An operator's board pack shows monthly active players of 84,000. The marketing team's dashboard shows 112,000 for the same month. Both are produced from the same warehouse.
The investigation finds the following.
Finance's definition requires a real-money transaction in the period, excludes customers whose accounts were subsequently closed, excludes activity later voided, and counts only verified accounts.
Marketing's definition counts any customer who logged in and took any action including free play, includes accounts closed after the period, and includes unverified accounts that were active.
Both are defensible. Finance wants a figure consistent with the revenue it reports. Marketing wants a measure of the reachable audience.
The dispute is not resolvable by deciding which is correct, because they are answering different questions. It is resolved by naming them differently: transacting players and engaged players, defined explicitly, both available, neither called active players.
That resolution takes an afternoon. The dispute, left unresolved, recurs monthly, consumes management time, and undermines confidence in every other figure from the same source, because if the actives number is contested then everything might be.
The general lesson is that most definitional conflicts arise because two teams need different things from the same word, and the remedy is naming rather than adjudication.
Metrics that should exist and usually do not
A closing list of measures that answer important questions and are absent from most operators' standard reporting.
Contribution by cohort over time, which is the fundamental measure of whether acquisition creates value and requires cost attribution most operators lack.
Value concentration, meaning what proportion of revenue comes from what proportion of customers, and its trend. This is both a commercial and a protective indicator.
Cohort quality trend, showing whether successive acquisition cohorts are worth more or less than their predecessors.
Contact rate, meaning support contacts per active customer, which measures the friction the product generates.
Journey completion rates by stage, which locate where prospects are lost.
Net bonus cost, calculated from actual play, rather than awarded value.
Incremental effect of promotional and lifecycle programmes, from holdouts.
Concentration exposure by market, supplier, channel and customer segment, which appears in no performance report and describes the risks that materialise suddenly.
Protective coverage, meaning verified evidence that suppressions and controls operate.
Each of these answers a question that matters and none is technically difficult. Their absence is a choice about what the organisation looks at, which is the point the reporting lesson develops.
Vertical-specific measures
The core metrics apply across an operator and each vertical has measures specific to it, which are worth knowing because they are frequently reported without definition.
Sportsbook. Turnover, hold and margin as described in the Sportsbook Trading course, plus expected against actual margin, bet count, average stake, accumulator share and cash-out rate. The trap is that hold varies enormously with results over short periods, and a monthly hold figure says more about sporting outcomes than about pricing.
Casino. Turnover, house edge, game-level performance, session metrics and bonus contribution. The trap is that casino turnover includes restaked winnings, making it non-comparable to sportsbook turnover and making revenue as a proportion of deposits the more meaningful ratio.
Poker. Rake, liquidity by hour and format, player counts by stake, and the ecology measures described in the poker course. The trap is that revenue is a function of volume rather than of outcome, which makes poker's economics behave unlike anything else in the operator.
Bingo. Attendance by room and session, ticket sales, prize fund percentages, and the community measures described in the bingo course. The trap is that bingo revenue in isolation understates the vertical, since side game revenue is attributed elsewhere.
Live casino. Table utilisation, which is the operational measure that determines whether the supplier economics work.
The general point is that an operator reporting a single set of metrics across all verticals is applying definitions that fit some and distort others, and that vertical-appropriate measures should sit alongside the group-level ones.
Maintaining definitions over time
A final practical consideration, since definitions decay.
Changes happen. A platform migration changes how something is recorded. A new payment method requires a category. A regulatory change alters what must be captured. Each can silently change what a metric measures.
Historical comparability breaks. A definition changed without versioning makes every comparison across the change point invalid, and the invalidity is invisible.
Ownership lapses. The person who defined a measure leaves, and the reasoning behind it goes with them.
Drift occurs. A report is rebuilt, slightly differently, and the two versions diverge without anyone noticing.
The maintenance practices that address this are modest. Version definitions with effective dates. Record the reason for each change. Flag comparison across change points in reporting. Review the dictionary periodically rather than only when a dispute arises. And assign ownership for each measure to someone who will notice if it stops meaning what it meant.
None of this is interesting work, and an operator that does it has an analytical environment that can be trusted over multi-year horizons, which is where the questions that matter most are answered.
Presenting metrics honestly
A closing point that belongs with definitions because it concerns how they are communicated.
State the basis. A revenue figure without its level in the stack, a retention figure without its window, or an active count without its threshold is incomplete and invites misinterpretation.
Pair volume with quality. Registrations with deposit conversion, conversion with downstream value, acceptance with fraud rate. A volume measure alone can be improved by degrading what it is meant to represent.
Show the distribution. Any per-customer average should carry the median and, where it matters, the deciles.
Indicate stability. Figures subject to retrospective adjustment should say so, since a reader comparing to last month's report needs to know why they differ.
Give the comparison. A number without a reference point conveys little. Against the previous period, against plan, or against a comparable group.
Be precise about precision. Reporting to a granularity the underlying measurement cannot support conveys false confidence and invites decisions the evidence does not carry.
Name the caveat once, clearly. Analysis buried in qualifications is ignored; analysis with no qualifications is misused. One clear statement of what the figure does not tell you is the useful middle.
These are presentational disciplines rather than analytical ones, and they determine whether a correct analysis produces a correct decision, which is the only thing that ultimately matters.
The metrics that carry protective weight
A final category, because several measures in this lesson serve two purposes at once and the second is easily overlooked.
Value concentration indicates commercial risk and indicates how dependent the operator is on a small number of very heavy spenders, which is the overlap described in the iGaming Basics course.
Spend distribution and its trend shows whether growth is broad or is coming from a narrowing group.
Session length distribution, particularly the tail, indicates whether a small number of customers are playing for periods that warrant attention.
Deposit frequency within sessions is a recognised harm indicator and is also a payment and product measure.
Reversed withdrawals is among the more reliable individual indicators and appears in payment reporting.
Tool adoption, meaning what proportion of customers set limits, indicates whether the tooling is findable and is a compliance-relevant measure.
Intervention outcomes, meaning what changed after an interaction, which the Law and Compliance course identifies as the measure regulators actually examine.
The point for an analyst is that these figures are frequently already being produced for commercial reasons, and presenting them with the protective interpretation alongside costs nothing and puts information in front of people who would otherwise not see it.
An analytics function that reports value concentration purely as a commercial concentration risk is reporting half of what the number means.