Current Overnight Guests
2.91 avg nights stayed
The only segment actually booking rooms today. Balanced taste — rides, shows, food. A "jack of all trades" guest already paying to stay.
Priority: High · Retention
A data-driven blueprint for Lobster Land's expansion into on-site lodging — from guest segmentation and conjoint design to classification of premium-package buyers.
Lobster Land's gates have been busy — but most guests come, ride, eat, and leave. Management wants to know: could a hotel keep them around?
Borrowing pages from Disney and Universal's "stay and play" playbook, the park is weighing three lodging concepts: a modest family hotel, a premium coastal resort, or a tiered hybrid. Our team was hired as analytical consultants to figure out which guests would actually book, what package features they want, and how much it should cost.
This site is a tour of what we found.
A K-Means model on ten numeric guest variables (income, party size, distance traveled, experience preferences) surfaced four distinct personas. We chose k = 4 on business grounds — the silhouette and elbow plots didn't disagree, and four clusters give management a manageable, distinguishable set of guest types.
The only segment actually booking rooms today. Balanced taste — rides, shows, food. A "jack of all trades" guest already paying to stay.
Older, small parties, culture-focused. They prefer shows over rides and spend big — but currently come and go in a single day.
Largest parties, longest drives, highest household incomes — but tight per-day budgets. Natural overnight candidates from a logistics standpoint.
Youngest, most frequent visitors, food-obsessed. They say they'd stay overnight — but rarely do. The biggest gap between intent and behavior.
We ran a ratings-based conjoint on stay_options.csv, joined the part-worth utilities
to vendor costs, and brute-forced every possible bundle. Goal: maximize utility under
the $150 per-guest cost cap finance handed us.
Trade-offs: we left fitness at the basic tier (utility lift wasn't worth the $40+ cost) and skipped the Luxury Suite (Premium View beat it on utility per dollar). Every cent saved went into perks guests actually rated highly.
We trained Logistic Regression and Random Forest on stay_play_guests.csv to
predict premium-package purchase. The Random Forest hit 94% on training data but cratered
to 67% on test — textbook overfitting. Logistic Regression won.
Generalizes cleanly — test accuracy slightly above train, no overfitting, interpretable coefficients.
Memorized the training set. Even after GridSearchCV tuning, the gap stayed too wide to trust.
Lobster Land licensed four hero images for an email campaign. We used appropriate statistical tests to compare engagement and recommend a winner.
Couple, ocean view from room
Daytime crowd, balconies
Lounge chairs, family beach
Resort exterior, golden hour
p-value > 0.05 — no statistically significant difference across images.
A graduate student consulting team out of Boston University's MS in Applied Business Analytics program.
Segmentation lead
Conjoint & recommendation
Classification
A/B testing & viz