The data problem is the number of events
A venue running two hundred events a year has two hundred rows a year. Each is different in artist, format, pricing and audience. This is a small-data problem however large the ticket volumes look.
That shapes everything. Complex models overfit immediately; the useful approaches lean on structure - booking curves, comparable events, and explicit assumptions - rather than on learning freely from attributes.
Booking curves are the workhorse
How sales accumulate over the weeks before an event is far more informative than the event's description. An event tracking 20% behind comparable events at four weeks out is a signal you can act on.
- Group historical events into comparable types - genre, size, day of week, price band.
- Build the typical sales curve for each group as a share of final attendance.
- For a live event, compare actual sales against the curve to project the final figure.
- Update as more data arrives; the projection narrows as the date approaches.
- Record the projection at each point so you learn how reliable it is at each distance.
This gives something genuinely useful: an early warning that an event is underselling, while there is still time to add marketing or adjust pricing.
Walk-ups, no-shows and the difference between them
Tickets sold is not attendance. For catering, staffing and safety, what matters is people through the door.
| Factor | Effect on attendance |
|---|---|
| No-show rate by ticket type | Comps and early-bird typically show lower attendance |
| Weather on the day | Significant for outdoor and marginal events |
| Day of week and time | Weeknight events see more late drop-off |
| Advance window | Tickets bought months ahead show up less reliably |
No-show rates are usually stable enough per ticket type to predict reasonably, and getting them right matters more for bar and catering planning than small errors in total sales.
Dynamic pricing, carefully
Adjusting price by demand is well established in events and carries a real reputational risk if handled badly. Customers who see a price drop after buying, or a surge on a popular event, react strongly.
Practical constraints that keep it defensible: publish a maximum, never reduce below what earlier buyers paid without a goodwill gesture, and hold a fixed allocation at the original price. The revenue upside is smaller with these constraints and the downside is much smaller too.
Staffing and catering
These are where attendance forecasts convert to money. Over-staffing a bar is a direct cost; under-staffing it loses sales at the interval and produces complaints.
Forecast spend per head as well as attendance - they vary independently, and the same audience size spends very differently across event types. A model of attendance alone will systematically mis-plan the bar.
Two hundred events a year is two hundred rows. Plan the analysis around that, not around the ticket count.