What data do you need for an AI concert ticket-sales forecast?

An AI ticket-sales forecast starts with a defined question and a usable dataset. A folder of final event totals cannot explain how demand developed before each show. Before choosing a model, decide what the forecast must estimate, when the team will use it and which information will actually be available at that moment.

This is a suggested data brief for a promoter commissioning or evaluating a forecasting workflow. It does not describe a required Lookport upload format or promise that the platform implements every modelling step below.

Write the target in one sentence

For example: estimate final net paid tickets, using information available 28 days before the concert, for shows that are already on sale. Define net paid tickets as a consistent count after recorded refunds, while tracking complimentary tickets separately. A revenue forecast requires additional assumptions about ticket mix and future prices.

Keep the forecast horizon fixed when comparing results. Predicting tomorrow's sales and predicting final attendance six weeks away are different tasks. Also specify whether the model predicts actual sales under inventory constraints or unconstrained demand; sold-out events cannot reveal all unfulfilled demand from purchase records alone.

Prepare a daily event-level dataset

Suggested fields for a concert forecasting brief

Group	Examples
Event identity	Stable event ID, market, onsale date, show date, time zone
Sales observations	Observation date, paid tickets, refunded tickets, consistent revenue measure
Inventory	Saleable capacity, available categories, holds and release dates
Commercial context	Prices, offer changes, recorded campaign spend
Known interventions	Artist announcements, support additions, campaign launch dates

Agree whether a row contains a daily amount or a cumulative total. Do not add cumulative totals together. Preserve zero-sales days, and mark missing data as missing instead of silently turning it into zero. Check that the daily records reconcile to the final event report using the same definitions.

Keep an honest history of what was known

Suppose a support act was confirmed ten days before the concert. A forecast reconstructed for day minus 28 must not use that later announcement as though it were known. The same restriction applies to eventual ad spend, final ticket mix and a capacity change agreed later.

Save dated snapshots or change histories where practical. Distinguish planned future actions known at the forecast date from what eventually happened. Testing with information from the future creates misleading accuracy. Chronological evaluation is explained in Forecasting: Principles and Practice.

Run checks before training

• Can every row be assigned to one event and one observation date?
• Do currencies and time zones remain explicit?
• Are cancellations, refunds and rescheduled shows distinguishable?
• Does an inventory restriction explain an apparent sales pause?
• Are there duplicate exports or missing reporting periods?
• Can the training data be reproduced from a documented source?

A rescheduled show may keep earlier purchases while changing the remaining sales window. Flag it rather than treating it as an ordinary new event. A new market with little comparable history also needs a visible uncertainty flag; an AI model does not manufacture reliable evidence simply by accepting more columns.

Use only the detail the task needs

This event-level brief does not need buyer names, email addresses, payment details or individual messages. Start with aggregate observations in an approved workspace. Any use of more detailed customer information needs a specific purpose and the organisation's approved permissions and handling process. Avoid pasting raw order exports into an unapproved chatbot.

Ask for a reproducible handover

Request a data dictionary, known limitations, forecast timestamp, model version and comparison with a simple baseline. The report should show which events could not be scored and why. Read how to test AI forecast accuracy before accepting a single headline percentage.

Use the forecast as an input to the event budget, not as a replacement for cost modelling. Lookport's documented ticketing features provide the product context; the dataset and evaluation process should match the actual tools and evidence available to your team.