Guest-worth analytics for casino resorts

ResortWorth

Sabermetrics for the resort: a documented library of guest-worth metrics that goes past ADT, behavior-first personas your hosts can actually remember, and an AI analyst that shows its SQL and refuses to guess.

100% synthetic data. Every guest, rating, and offer is generated by a seeded simulator. No real patron data appears anywhere in this project.

The demo is the real dashboard, running on the synthetic warehouse.

Overview · trailing 90 dayssynthetic
Theo win
$4.21M▲ 3.2%
Rated visits
41,882▼ 1.1%
ADT
$58.40▲ 0.9%
RE
0.86×▲ 4.4%
Midnight Felt Warriors
overnight regulars, metronomic rhythm
6.4% of theo
WI+118
VAW$66
Visits / mo4.6
LTV24$9.4k
Theo win by day3 ms
Win-back queue
RFH 0.9118d out · rhythm 6d
RFH 0.7831d out · rhythm 14d
RFH 0.7244d out · rhythm 21d
Illustration of the demo dashboard. Every figure shown is synthetic.
The metrics library

Eighteen documented metrics, not one opaque score.

The industry still ranks guests on ADT — average daily theoretical — the way baseball once ranked hitters on batting average. It is a single number with no peer context, no sample-size discipline, and no way to tell luck from worth.

Every metric here is documented like a baseball card: a definition, the formula, why it beats the ADT view, and a line on how to act on it. The same registry that computes a metric writes its glossary entry, its tooltips, and the analyst’s description of it — so a CFO can audit any number down to the arithmetic.

Also in the library: LCK, a luck index in game-math standard deviations, which separates a guest who is running bad from a guest who is worth less (and flags the persistent negative tail as possible advantage play). And RE / PEL, which ask the question reinvestment percentage cannot: did the offer buy incremental play, or did it pay someone to do what they were going to do anyway?

WI+Worth Index Plus

Worth indexed to comparable guests. 100 is exactly peer par; 150 is 50% above.

100 × worth-per-active-day ÷ peer average

Why it beats ADT — The OPS+ move. A $60/day local and a $60/day fly-in are not the same asset — peer context is what raw ADT never gives you.

VAWVolatility-Adjusted Worth

Worth per day, shrunk toward the peer mean in proportion to how little we have actually observed.

(n · worth + k · peer_mean) ÷ (n + k), k = 8 active days

Why it beats ADT — Empirical-Bayes shrinkage. Two visits at $400 theo is weak evidence — one hot weekend can’t mint a fake VIP.

RFHRecency-Frequency Hazard

Probability that the current absence already exceeds this guest’s own visit rhythm.

P(gap ≤ days_since_last) under the guest’s own Weibull fit

Why it beats ADT — 21 quiet days is an emergency for a three-times-a-week local and noise for a quarterly visitor. Fixed churn windows miss both.

The persona engine

Clusters your hosts can actually remember.

Guests are clustered on eighteen behavioral features — worth, cadence, game mix, bet level, time fingerprint, promo response, hotel stays, dining, and how much of their worth comes from outside the casino — with a Gaussian mixture and soft membership, so a guest can be 60/40 across two personas instead of being forced into one bucket.

Each cluster comes out as a character with a memorable name: Midnight Felt Warriors, Weekday Video Poker Regulars, Floor Resort Guests. Each ships a stat line as percentiles of the whole book, psychographic traits that are always labeled with the behavior they were inferred from, and a playbook of trigger → action → the metric it should move.

Demographics are profiled, never clustered on. Age and distance describe a persona after the fact; they never define one. Clusters are gated on bootstrap stability before they earn a card, and guests with too little observed play are told so rather than being quietly assigned.

Weekday Video Poker Regulars

mid-week grinders, low bet, long sessions

17% of total worth
ADT$44
WI+127
Visits / mo6.2
PEL0.12

Bars are percentiles of the whole book; the tick is the median.

Psychographics — inferred from behavior
  • high routine-seeking — from inter-visit regularity, TCR 0.84
  • low novelty-seeking — from game concentration, GDE 0.14
  • low promo elasticity — from visit odds on offer vs non-offer days
Playbook · RFH crosses 0.7 → mid-week free-play tuned to their daypart, not a weekend mailer → should move visits / mo and RE, not just response rate.
A persona card from the demo. Synthetic guests, real method.
The AI analyst

An analyst that shows its SQL — and admits what it can’t answer.

Ask in plain language. The analyst matches the question against a curated catalog over a semantic layer, extracts the slots it needs — dates, metrics, personas, top-N — runs parameterized SQL, and narrates the result deterministically.

The query is always on screen. Numbers come from the warehouse, never from a language model’s memory of what a number like that usually looks like.

Off-catalog questions get a straight refusal plus the nearest questions it can answer. A model that says “I don’t have that data” is worth more than one that produces a confident, unfalsifiable paragraph — especially when someone is about to spend reinvestment budget on the answer.

Which personas have the best reinvestment efficiency this quarter?

Weekday Video Poker Regulars lead at RE 1.42 — $1.42 of incremental worth per comp dollar. Midnight Felt Warriors trail at 0.31: their offers are largely paying for visits that would have happened anyway.

Weekday Video Poker Regulars1.42
Reel Regulars0.94
Midnight Felt Warriors0.31
▾ SQL — always shown
select persona_name,
       round(sum(incr_worth) / nullif(sum(comp_cost), 0), 2) as re
from marts.patron_360
where last_active between $start and $end
group by persona_name
order by re desc
deterministic engine · confidence 0.94 · 4 ms

Which guests are most likely to book a spa package?

I can’t answer that. This warehouse has no spa data, and I don’t guess at numbers. Here is what I can answer:

Who is most movable by offers?Which personas are lapsing fastest?How is theo trending?
Theo Hunter

The theo the rating system never sees — and who it belongs to.

Slot meters and table drop see every bet. Player-card ratings see only the carded play. The gap between them is already in the meters; what has been missing is who it belongs to. Theo Hunter produces estimates of that uncarded theo, attributed to guests.

Every estimation method is scored against planted truth. The simulator knows what each guest actually played, so each method’s error can be measured exactly, and the methods are ranked on a method leaderboard inside the demo. That is only possible because the data is synthetic; real play has no answer key. The methods are also stress-tested on a harder property where some visits involve no play at all.

A responsible-gaming screen runs before any recommendation. Guests on the RG watchlist, and flagged guests with a large estimated share of play off the card, are never marketed to.

What each system sees
Slot meters + table dropsee every bet
Player-card ratingssee carded play only

The dashed span is theo the meters recorded but no rating ties to a guest. Theo Hunter estimates who it belongs to.

Method leaderboard — error vs planted truth (shorter is better)
Method A
Method B
Method C
Responsible-gaming screen · runs first

Watchlist guests, and flagged guests with a large estimated share of play off the card, are never marketed to.

Illustration. Shapes show the idea, not real results; method names are placeholders.
Labor

Staff to the demand the whole guest creates.

Hour-by-hour demand by department is turned into required staff under published standards. Those standards are synthetic staffing assumptions, shown on the page rather than buried in a model. The Labor page also shows the share of the floor that is off the player card, so the staffing view does not shrink to the play the ratings happen to see.

The Labor plan goes a step further: weekly demand forecasts and shift schedules, backtested over a year and scored on what actually happened, not on how reasonable the plan looks.

One department, one day: demand by hour → required staff
0006121824

Each hour’s demand sets the staff every role needs, under a published standard. The standards are synthetic assumptions, shown beside the number.

Labor plan — backtested, scored on what happened
Forecast
What happened
Off the player card

The share of the floor that never shows in ratings is shown beside the demand, so staffing is not sized to carded play alone.

Illustration. Shapes show the idea, not real results.
Responsible gaming

Protection wired into the marketing engine, not bolted onto it.

Five behavioral flags — session escalation, loss chasing, frequency spikes, late-night concentration, stake drift — each carrying the evidence that triggered it, composited into a risk tier a human can review.

Guests at the Elevated tier are automatically excluded from frequency-stimulating offers. That exclusion lives inside the persona playbooks, in the pipeline, not in the memory of whoever pulls the list this month.

Flags are evidence for a conversation, not a verdict. Every one of them shows the behavior and the window that produced it, so the person acting on it can see exactly what the model saw.

Guest 4482-1179
Midnight Felt Warriors · 14 months tenure
RG Elevated
risk 0.78
  • session escalation
    median session 41 → 96 min over 8 weeks
  • loss chasing
    re-buy within 20 min in 3 of the last 4 losing trips
  • late-night concentration
    63% of play 12am–6am, up from 28%
Guardrail

Auto-excluded from frequency-stimulating offers. The exclusion is enforced in the persona playbook itself — it is not left to whoever pulls the mailing list.

Illustrative flag card. The guest, the flags, and the evidence are synthetic.
Where this goes

The floor, the room, and the restaurant. Built for the whole property.

Gaming ratings, hotel stays, and card-linked dining now resolve to one Total Folio Worth per guest. The hotel runs as a real house: occupancy, cash ADR, RevPAR, and what comp rooms actually cost on sold-out nights. Personas are clustered on the whole stay, not just the floor.

Entertainment, spa, and retail plug into the same guest key next. The same demand view already drives staffing, so new departments inherit it.

Gaming, hotel, dining, Theo Hunter, and labor are shipped. Entertainment, spa, and retail are the direction, and it is honest to call it a direction.

Under the hood
DuckDB
columnar warehouse — the analytics live in SQL, never in the API layer
25,000
synthetic guests, drawn from eight latent behavioral archetypes
2 years
of simulated play, stays, and dining: ratings, comps, offers, folios
1 seed
one command rebuilds the identical warehouse, every time
1–5 ms
typical query time — and every response carries its own

Python, polars, scikit-learn, and DuckDB for the pipeline; FastAPI over a read-only warehouse; Next.js for the dashboard. Determinism is a test, not a promise: the generator’s math invariants, the metric golden values, and the persona stability gates all run in CI.

See it running

The demo is the whole thing: the glossary, the persona gallery, guest 360s, the floor and promotion views, hotel and dining, Theo Hunter, labor and the labor plan, the responsible-gaming queue, and the analyst — all over 25,000 synthetic guests.

Signed in? Take the guided tour. It walks through the story, and the dashboard works on a phone.