Inspiration
A fictional casino planned to replace its ten lowest-earning machines. We wanted to investigate whether low earnings reflected machine problems, downtime, location, or ordinary payout variation—and help management make better decisions before spending money.
What it does
DataLarp presents four interactive Tableau dashboards covering demand forecasting, machine performance, tracked guests and loyalty, and budget decisions.
The July 1–30 outlook estimates $40.71 million in wagers and $3.60 million in expected contribution after supplied costs, rather than full net profit. Interactive scenarios compare staffing and promotion break-even requirements. A Gemini-generated briefing explains the supporting evidence and limitations.
How we built it
We used Python, pandas, NumPy, and PyArrow to validate and transform the supplied synthetic data into Tableau-ready CSVs.
We compared constant-mean, weekday-average, and weekday-plus-trend forecasts. Model selection used pre-May validation, followed by frozen predictions for the May–June holdout.
We built the dashboards in Tableau Public and connected Gemini through a local Python API server. Successful explanations export to Tableau with evidence identifiers and timestamps.
Challenges we ran into
Different datasets required careful handling to avoid duplicated money and player counts. Machine comparisons needed to account for exposure, availability, and luck.
Learning Tableau, managing memory limits, and resolving Gemini billing configuration also required troubleshooting. We protected credentials through local key storage, ignore rules, and reviewed source packaging.
Accomplishments that we're proud of
We published four dashboards and integrated a genuine Gemini explanation.
Our weekday forecast achieved $68,699 daily holdout MAE versus $218,258 for the constant baseline—a 68.5% reduction.
Our proposed plan funds one $1,200 inspection of M-1158 and preserves $98,800, without assuming guaranteed cash recovery.
What we learned
More complex models are not automatically better. Clear definitions, honest validation, and explicit assumptions make analysis useful.
We also learned to distinguish player outcomes, casino win, and contribution—and that a break-even requirement does not establish a causal benefit.
What's next for DataLarp - Dashboard Royale
We would test promotions or staffing interventions using credible comparison groups, measure incremental contribution after additional costs, and monitor inspected machines using new data before expanding spending.
