Superintelligence for Theaters and Venues: A Ticketing Agent With Limits

Superintelligence for Theaters and Venues: A Ticketing Agent With Limits

An AI agent for theater and venue ticketing and production questions: approved answers only, fees and contracts out of scope, human handoff.

•

Want to apply this in your company?

Book a call with Juan →

Superintelligence, in the technical sense, describes systems that would outperform people on most cognitive tasks, and that is not what a box office or production team needs to put to work. What theaters and concert venues can run safely is an AI agent that answers ticketing and production questions from approved information, logs every conversation, and passes anything involving fees, contracts or exceptions to the human team. The value comes from well-defined limits, not from how much the agent can answer.

Key takeaways

  • The agent answers only what is in a knowledge base approved by the venue: showtimes, exchanges, accessibility, discount rules and cancellation policy.
  • Artist fees, contract terms, technical riders and date negotiations stay outside the agent's scope, without exceptions.
  • Buyer and supplier data follows LGPD: minimal collection, a stated purpose and no exposure of third-party data.
  • Any question outside the knowledge base becomes a handoff with a summary for the responsible person, never a guess.
  • A pilot with one support queue, real questions and readable logs shows where the agent performs well and where it must stop.

Which questions the agent can answer on its own

The audience of a venue asks repetitive, predictable questions: where the box office is, whether a show's age rating suits children, how to exchange a ticket bought online, whether there is wheelchair access, what time the doors open. These answers exist in some document, but they are usually scattered across the website, the box office email and the memory of whoever works the counter. When information is spread out, each staff member answers differently, and the patron receives different versions of the same rule.

For production, the common questions are different: arrival time for the load-in crew, documents required to bring equipment in, the loading dock area, the check-in window for technicians and suppliers. An agent can answer these if the information is written, dated and approved by the production manager. When the answer depends on something that changes every week, such as a tour schedule, the agent should say it will confirm with the team and should not invent a date.

What stays outside the agent's reach

Artist fees are the question most likely to push an agent into improvising. The question sounds simple, but the figure depends on the contract, the negotiation, the terms specific to each production and financial rules the venue should not expose in an open channel. A wrong number said on WhatsApp becomes a commitment, a screenshot and a complaint. For that reason, the agent does not state a fee, does not estimate one, does not compare amounts between artists and does not comment on ongoing negotiations.

Contracts follow the same logic. The agent can say who is responsible for production and how to request a copy through a formal channel, but it does not interpret clauses, does not promise cancellation, postponement or refunds outside the published policy and does not accept changes to terms through a chat message.

A written instruction in the prompt is not enough on its own. The agent also needs limited technical permissions, with no access to fee spreadsheets or contract folders. People try to make agents reveal internal information with clever questions, a practice known as prompt injection, and the OWASP guidance on the topic describes ways to reduce that risk. When the access simply does not exist, the leak is no longer a possibility. A useful parallel is SAC with AI, which automates what is repetitive and keeps decisions with people.

How to build the knowledge base

The knowledge base needs an owner, one document per subject and a date of last review. Avoid putting everything into one huge text, because nobody can review or maintain that kind of file. Separate tickets, production, accessibility and policies. Each item should have a question and an answer approved by the responsible area, plus a clear instruction on what to do when the question is not covered.

A simple check before publishing the knowledge base:

  • Each subject has a named owner at the venue and a date of last review.
  • Answers are short and state their conditions, such as specific performances, deadlines and exceptions.
  • There is an explicit list of forbidden topics, including fees, contracts and negotiation.
  • There is a standard handoff message for questions outside the knowledge base, with a promise of a human reply.
  • Any policy change updates the knowledge base before the agent answers about that topic again.

That last rule is the one most often broken in practice. If the box office changes its exchange policy and the knowledge base still shows the old one, the agent confidently repeats something that no longer applies. For that reason, the review has to be tied to the venue's process, not to someone's memory.

Plan for a support pilot

A well-designed pilot fits in a few steps, and the order matters:

  • Choose a single queue, for example ticketing questions on WhatsApp, and leave production for a second phase.
  • Gather the real questions received by the box office and the message inbox, removing personal data before any analysis.
  • Write the approved answers and the list of forbidden topics together with the owner of each area.
  • Test with edge cases: a fee question disguised as a ticket question, a request for a copy of a contract, a threatening complaint, a request for another buyer's data. Run these in a test environment before exposing the number to the public.
  • Define the success measure before starting, combining the share of questions resolved without handoff with the share of handoffs that the team judges correct on review.
  • Review the logs every week: what was answered, what was handed off and where the agent made mistakes.
  • At the end of the agreed period, decide whether to expand, adjust or stop. Stopping is also a valid result when the pilot shows the flow does not fit the chosen channel.

Illustrative scenario: a ticket question that turns into a fee question

Illustrative scenario: a patron writes on WhatsApp asking whether they can exchange a ticket for a canceled show and, in the same message, asks how much the artist was paid for the performance. The agent answers the first part with the exchange policy published in the knowledge base and the deadline written there. On the fee, it states that the venue does not disclose that information and that it cannot confirm amounts. It then logs the case for the production team with a summary: the show name, the performance date, the fee question marked as out of scope and a note that the patron asked for a reply. A team member responds later through the human channel. The log shows exactly what was said and why, and the agent did not invent any information.

Buyer and producer data

The agent handles names, phone numbers, emails and, in some cases, data needed to locate a purchase for an exchange. Apply the logic of LGPD: collect only what is needed to resolve the request, explain the purpose in the first message, do not pass data to third parties outside the workflow and never expose one buyer's information to another. If the service requires a document or sensitive data, the request should be made through the box office's official channel, not through the chat. The National Data Protection Authority publishes guidance, and the official material is available on the authority's own website.

Active messages, such as performance reminders or notices of schedule changes, require prior permission and WhatsApp-approved templates. This point deserves its own attention, and the subject is covered in WhatsApp outbound messages, with templates and opt-in. Replies to people who write first follow different timing rules, which are worth knowing before you design the flow: the WhatsApp 24-hour customer service window defines how the agent should act after a message from the public.

How we apply the method: map, limit, pilot and hand off

In practice, we start by mapping the current box office and production flow: question volume, types of questions, response time and where the information is stored. This survey is the baseline against which the pilot will be compared. Then we limit data and actions, so the agent reads the approved knowledge base and does not access fee spreadsheets or contract folders. Next we pilot a real flow, with logs and a measure agreed in advance. Finally we hand the operation over to the venue's team and decide, using the pilot's own data, whether the design should grow, be adjusted or be closed.

If the leadership team still needs to align on vocabulary before deciding, it is worth reading the difference between AGI and superintelligence, and what your company actually operates. For a performance venue, the useful question is not what kind of intelligence sits behind the tool, but which answers it can give on its own and which ones require a person.

Frequently asked questions

Can the agent tell a patron an artist's fee if they ask? No. The fee depends on the contract and the negotiation, and a wrong figure becomes a commitment for the venue. The agent should say the information is not available in that channel and log the case for the responsible team.

Does an agent replace the box office team? No. It absorbs repetitive questions and records the rest with context, so the team can respond with less rework. Exceptions, refunds outside the policy and negotiations remain with people.

How long does it take to launch a pilot? The timeline depends less on the technology and more on the team's availability to write and approve the answers. For that reason, the pilot should start with a single queue, with defined owners and a set date for review and decision.

Do I need to give the agent access to all buyers' data for it to work? No. The agent only needs the data required for each type of request, such as locating a purchase for an exchange. Collection should have a purpose explained to the data subject, and data that is not used should not be kept without need, following LGPD.

Sources