The pilot: method and results

Two properties. No names.

The pilot ran at two sister-property pilot casinos in Colorado. We do not disclose client identities, in this dossier or anywhere else. In this industry floor data and player data are guarded for good reasons, and a vendor that trades client names for marketing is telling you how it would treat yours. The anonymity here is policy, not absence of results, and the same policy would protect your property.

OptiFloor reads a nightly copy of a casino management system's data and turns it into decision support for slot and marketing teams. It is advisory by construction: it never moves a machine, never writes to casino systems, and never sends an offer. The pilot tested whether its answers can be trusted.

Do floor moves do what the reports say they do?

Every slot team moves and converts machines, and every CMS produces reports that appear to grade those decisions. The pilot asked the uncomfortable version of the question: if you replay decades of actual relocations and score each one honestly, how well do naive readings of the data hold up, and how well does OptiFloor's corrected model do against them?

Replay history. Grade the model. Show the error.

• The dataset. 760 historical machine relocations across twenty-two years of the two properties' own CMS history.
• Blind scoring. Each relocation was scored using only information available before the move. The model never sees the outcome it is asked to predict.
• Held-out error. Accuracy is measured on held-out data and compared against a no-change baseline, so a number only counts as skill if it beats doing nothing.
• Selection-bias correction. Good machines sit in good zones. A naive zone-difference estimate reads that placement advantage as lift. OptiFloor measures the bias per casino and shrinks its estimates accordingly.

The same machinery ships with the product: every forecast OptiFloor makes reports its own tested accuracy, measured by replaying the model against the machine's actual history. A pilot at your property runs this same replay on your own data before you are asked to trust anything.

The numbers, exactly as measured.

The overstatement finding deserves the emphasis: before any model does anything clever, the pilot showed that ordinary readings of relocation results can overstate real effects by 2 to 4x. If your floor decisions are graded by naive before-and-after comparisons, some of what looks like lift is selection bias.

What these numbers do not say.

• No revenue claim. The pilot measured measurement quality: direction accuracy and estimate bias. It did not measure, and we do not claim, a revenue outcome from acting on the recommendations.
• Two properties. The figures are specific to the two pilot casinos' histories. Your floor gets its own replay, and its own numbers, before any conclusion applies to you.
• Known blind spots. Uncarded play, thin data windows, and uncontrolled seasonality limit what any floor model can see. OptiFloor discloses these on the page wherever they affect a number, and says "insufficient evidence" rather than bluffing when the data cannot support a verdict.