Every Machine Must Earn Its Place
Casino floors run on capital that is authorised long before it is understood. Slot replacement is the largest recurring capital line most operators sign off, and it is routinely decided from a performance ranking rather than an analysis.
“The bottom 10% is not a replacement list.”
Raw performance is a diagnosis problem before it is a procurement decision. This report sets out how to tell a bad game from a bad location, what evidence is required before capital moves, and how to authorise on incremental floor contribution rather than gross machine uplift.
Every Machine Must Earn Its Place
How casino operators should decide what to keep, move, expand, reconfigure or replace.
Figures are labelled by evidence class. Supplier average selling price is a supplier-reported unit price, not an operator's all-in capital cost — installation, conversion, licensing, systems and floor disruption sit outside it. Study effect sizes describe the sample studied and are not universal effects.
Five conclusions that change how a floor should be judged.
Ranking machines by raw performance and cutting the tail is a procurement habit, not an analysis. A weak number is a diagnosis problem first: it may be the title, the location, the configuration, the denomination, the period, or simply noise. Replace before you diagnose and you buy the same result in a new cabinet.
Placement demonstrably changes outcomes — but the direction is not portable. A 2025 paired study and a 2026 follow-up found opposite best pairings between title and floor position. Any operator who imports someone else's placement rule is importing someone else's floor.
Actual win carries volatility and luck. Theoretical win measures the math the machine was configured to produce. Economic value nets out space, capital, lease and incentive costs. Ranking on the wrong one of the three produces confident decisions about the wrong thing.
A relocation or reconfiguration is cheap, quick and reversible; a replacement is none of those. When the diagnosis is ambiguous, the move preserves the option to learn. Replacement spends the option to buy certainty you have not yet earned.
Attribute title and location effects before optimizing anything. Optimizing adjusted contribution on an unattributed floor simply re-encodes the existing placement bias into the plan and calls it a model.
Bad game or bad location?
Two signals, four states, four different decisions. The question is not which machines are weakest — it is which cause the weakness can be attributed to.
The title performs elsewhere or against comparables; the position is the constraint. Relocate or trial a better position before spending capital.
Both signals agree. Protect the placement and evaluate a measured expansion of the title or math family.
Confounded. A weak title in a weak position cannot separate the two causes, so this state is not high-confidence evidence for replacement. Run a low-cost move or reconfiguration test first.
The position is proven by its neighbours, so the title is the most likely cause. Reconfigure denomination or math first; replace if reconfiguration does not close the gap.
A low-performing game in a low-performing location is confounded: the two causes cannot be separated from the observation itself. That state is not high-confidence evidence for replacement. It is evidence that a cheap test is required before any capital is committed.
A proof ladder, climbed in order.
Each rung is cheaper than the decision it protects. Skipping rungs does not save money; it moves the cost from analysis into capital.
Is the underperformance stable across periods, or an artifact of a short window and volatility?
How does the unit perform against like-for-like titles, denominations and positions on the same floor?
Separate the title effect from the location effect before assigning blame to either.
Move, reconfigure or re-denominate. Cheap, quick, reversible evidence beats an expensive assumption.
Build the capital case on incremental floor contribution net of displacement, not gross machine uplift.
Measure the realised outcome against the authorised case, and feed the variance back into the next decision.
A successful machine can still be a failed capital allocation.
A new cabinet that outperforms the unit it replaced looks like a win on every machine report. But if most of that gain was displaced from neighbouring machines, the floor earned little or nothing — the play simply moved. Measured at the machine, the decision succeeded. Measured at the floor, the capital did not.
Displacement is not an edge case on a dense floor; it is the default. Adjacent banks compete for the same session, the same visit and the same wallet. Any capital case built on gross machine uplift systematically overstates return, and the overstatement compounds across a replacement programme.
Authorise capital on incremental floor contribution — the change in contribution across the affected zone net of displacement — never on gross machine uplift. Then post-audit the realised zone effect against the authorised case.
Eight steps, always in this order.
The order is the method. Every failure mode in slot capital allocation is a step taken out of sequence — usually deciding before attributing, or optimizing before diagnosing.
Sixteen chapters across thirty-six pages.
Where unit demand actually comes from, and why replacement rather than expansion sets the cycle.
What is feasible to establish from disclosed data, and where only operator-side data can answer the question.
What the published placement studies do and do not support, including their disagreements.
Actual win, theoretical win, adjusted contribution and economic value — what each one is fit to decide.
Separating title effects from location effects with paired designs, fixed effects and pre/post controls.
Bank position, sightlines, traffic paths and adjacency effects at sub-floor resolution.
Purchase, lease, participation and conversion economics on a comparable basis.
Measuring displacement so a machine-level win is not mistaken for a floor-level gain.
How title performance decays, and when decay is the signal versus when it is the season.
The keep / move / expand / reconfigure / replace decision surface and its evidence thresholds.
Optimizing adjusted contribution across the floor subject to space, capital and mix constraints.
Capital scenarios with the assumptions and sensitivities each one depends on stated explicitly.
Concrete sequences for slot operations, finance and executive sponsors.
The measurement layer required to make these decisions repeatable rather than heroic.
The operator-side studies that would resolve what public data cannot.
What this report cannot claim, and every figure traced to its original source.
Built for the people who sign the capital.
A defensible sequence for keep, move, reconfigure and replace decisions.
Authorisation on incremental floor contribution with a post-audit built in.
Understanding how operators should evaluate a placement, and how to be evaluated well.
Whether capital on the floor is compounding or quietly churning.
Every number carries its evidence class.
The report separates what the data proves from what judgement adds. Where a study establishes association rather than cause, it is described that way. Where we model, the range and its inputs are disclosed.
Published by gambling regulators and state agencies. Cited exactly as reported, with the period and jurisdiction attached.
Audited filings, investor decks and public statements from operators and suppliers.
Participating-operator panels and industry trackers. Always labelled as a sample, never as a total market.
Our own build-up, presented as a range with the assumptions and inputs disclosed. Never blended silently into reported figures.
Get the full Every Machine Must Earn Its Place.
The complete publication — replacement-market context, metric hierarchy, attribution methods, spatial analysis, capital economics, cannibalization, the allocation matrix, scenarios, playbooks and the full source library.
The research frames the market. Your own data decides the machine.
No published study can tell you whether a particular unit on your floor is a title problem or a position problem. That answer lives in your own performance, placement and player data — attributed, normalised and read against a single governed definition of contribution. Nucleus is where that layer sits.
External intelligence frames the market. Internal intelligence provides the operator-specific decision evidence.
Explore Nucleus
Shemal authors the Dr.D Intelligence series and builds the analytics environments behind it. He works on governed metric layers, executive reporting and applied AI — which is why the research is written the way an operator would need it: definitions first, evidence labelled, conclusions stated plainly.
shemal@dr-danalytics.com