Superintelligence for Shopping Malls: Tenant Support, Contracts and Tickets

Superintelligence for shopping malls, in practice, is not a system that runs the property on its own. What mall management can operate safely are AI agents with a defined scope. They help the team answer tenants, organize service tickets, track contract deadlines and prepare notices, while decisions about billing, exceptions and clauses stay with a responsible person. The real gain is faster triage and a record of what was done, not replacing the manager.

Key takeaways

• Superintelligence is a research and marketing term, not a product you can buy. What you can run today are agents with clear limits.
• The first useful cases usually sit in tenant support and recurring tickets, not in contract negotiation.
• Contracts enter as lookup, date extraction and deadline alerts, validated by the legal team.
• Any action that changes billing, penalties, access or formal communication needs recorded human approval.
• Start with one real workflow, with logs, a metric agreed before launch, and a rule for whether to expand, adjust or stop.

Why the word superintelligence gets in the way of decisions

When a manager hears that AI will be superintelligent, the expectation is usually full autonomy, and that leads to bad decisions: handing a system tasks that require judgment, contracts and accountability. For mall management, the useful question is different. Which repetitive tasks with known rules can an agent handle, and which require a person who signs and answers for the result? If you want to separate the concept from the everyday tool, read Is ChatGPT Superintelligence? What It Is and What It Is Not.

Where AI helps mall management

Mall management handles many requests that repeat: questions about opening hours, rules in the internal regulations, the status of a maintenance ticket, meeting schedules, notices about works and events. These requests have documented answers and known paths, which makes them good candidates. Lease negotiation, grace periods, default, and formal complaints are different. They require context, history and authority that an agent should not take on.

Tenant relations: what the agent answers and what it does not

A simple rule helps the team define the scope before any setup:

• Can answer: published regulations, opening hours, standard documents, guidance on how to open a ticket, and the status of a ticket already logged.
• Can hand off with a summary: exception requests, requests to review a charge, complaints about another tenant, requests to change a unit.
• Must stay with a human: commercial negotiation, default, grace periods, penalties, termination, and any promise of a discount or deadline.
• Must not do: legal opinions, free interpretation of a clause, or commitments on behalf of management.

The agent needs to know when to stop. This is defined in the design, with explicit instructions and a handoff to a human agent with a conversation summary, so the tenant does not have to repeat everything.

Contracts: lookup and alerts, not decisions

Lease and mall management contracts contain many clauses, annexes and dates. An agent can help locate the adjustment clause, list obligations with deadlines, warn about upcoming expiries, and answer questions by pointing to the source passage. This works best when each answer cites the document and clause, and when the legal team reviews the document base before it goes into the system.

There are two cautions. First, reading a contract can go wrong, especially with amendments and outdated annexes, so the source must be the approved, current version. Second, tenant data such as company registration numbers, partner names and contact details is personal data in many cases and must follow the LGPD: collect only what is necessary, limit access by role, and record who consulted what. When an AI system makes an automated decision that affects a tenant, also review the obligations set out in the law, as explained in the content on LGPD automated decision-making and the right to review.

Tickets and maintenance: triage with a record

Illustrative scenario: a tenant reports water seeping through the back wall of the store at 7 p.m., by WhatsApp message. The agent asks for the exact location, whether there is an electrical risk, and whether there are photos, then classifies urgency using the criteria management has defined. If there is a sign of risk, such as a burning smell, exposed wiring or water near an electrical panel, the agent does not try to fix it. It immediately alerts the human on-call team and tells the tenant the emergency procedure. In common cases, it opens the ticket, gives the protocol number and the standard deadline, and records everything in the ticket system.

The value here is the complete record and consistent triage, not a clever reply. When tickets are closed, management can measure how many requests came through each channel and where the waiting time piles up, without depending on the memory of whoever answered.

Communication with human approval

General notices, works announcements, meeting invitations and documentation reminders are good candidates for an AI-drafted text, as long as they go through review before sending. A minimal flow has four steps:

• The agent drafts the message from a template approved by management.
• A responsible person reviews the text, date, audience and tone, then approves or returns it.
• The message goes out through the defined channel, with a record of the approver and the date.
• The tenant's reply returns to support with the history attached.

On WhatsApp, this discipline matters even more. Messages outside the customer service window require approved templates and proper permission, and the agent should not try to get around those rules. The details are in WhatsApp 24 Hour Customer Service Window: How an AI Agent Should Act.

Communication also carries a security risk. Messages from tenants, notices and documents can contain instructions that try to manipulate the agent, for example asking it to ignore rules or reveal internal data. OWASP guidance on prompt injection prevention covers this kind of attack and should be part of the pilot design.

A four-step pilot plan

The method we apply is map, limit, pilot and hand over. Applied to mall management, it looks like this:

• Map: list the most frequent tenant requests, who answers them today, how long they take, and where the process gets stuck.
• Limit: define which data the agent can access, which actions it can take alone, which require approval, and which are forbidden.
• Pilot: choose one real workflow, for example common-area maintenance tickets, with a log of every interaction and a success metric agreed before launch.
• Hand over: at the end of the period, decide whether to expand to other workflows, adjust the rules or stop, and document the operation for the team that will be responsible for it.

In this plan, superintelligence does not appear as a promise. It appears as a small, observable and reversible workflow.

Risks and governance management should require

Before expanding any agent, management should answer a few questions: who is responsible for the operation, where the records are kept, how long they are stored, who can see conversations containing personal data, and how the agent is shut down if it fails. The NIST AI risk management framework is a useful reference for structuring this discussion, because it covers governance, risk mapping and measurement without depending on one specific model. It is also worth following the guidance from ANPD, Brazil's data protection authority, on personal data processing, since using AI with tenant and customer data falls within that scope.

Frequently asked questions

Is superintelligence for shopping malls already available as a finished product? No, not in the sense of a system that runs the property without supervision. What exists and can be operated are AI agents with a defined scope for specific tasks, such as ticket triage and checking regulations, always with rules and records.

Can AI negotiate contracts with tenants? It should not. AI can look up clauses, list deadlines and prepare summaries, but negotiation, interpretation and the final decision are the responsibility of authorized people, with legal review when needed.

Which workflow should a mall pilot first? Usually a workflow with high volume, known rules and low risk, such as common-area tickets or questions about the regulations. Choose only one, define the metric before starting, and keep a record of every interaction.

How do you start without exposing tenant data? Begin with public documents and approved templates, limit access to personal data to the minimum necessary, and record who consults what. Before any expansion, review the legal basis, the notice to data subjects and the rules for data transfers.

Sources

• OWASP: LLM Prompt Injection Prevention Cheat Sheet
• NIST AI Risk Management Framework
• Anthropic: Building effective agents