AI support agent: a practical guide

If your team keeps repeating the same answers and missing enquiries, an AI support agent can respond using approved information and open a ticket when it needs data, a decision or human intervention.

In summary:

• An AI support agent responds using the company’s approved documentation, not an open collection of answers.
• The workflow must define which enquiries the agent can resolve and which require a person.
• Before opening a ticket, the agent must collect the information the team needs to continue handling the case.
• The knowledge base and escalation rules must be reviewed when products, processes or support terms change.

The starting point is to separate repetitive enquiries from cases that require judgement. The support vertical helps turn that separation into responses, data collection and escalations within a defined workflow.

What documentation does an AI support agent need?

An AI support agent needs up-to-date, clear and approved documentation about products, processes, service terms and the answers it can provide. It also needs explicit limits on what it must reject or refer to a person when the available information is not enough.

Prepare a knowledge base with the following content:

• Frequently asked questions and approved answers.
• User, installation and configuration manuals.
• Policies for changes, returns, warranties and support.
• Instructions for resolving common issues.
• Limits: topics that require human review.

Each document must state its scope and remain aligned with the current process. Avoid mixing outdated instructions with current ones: the agent could respond with a condition that no longer applies. If information is not documented or contains exceptions, mark it for review instead of filling in the gaps with an assumption. You can create the agent from your website and then organise the information it must use.

When can the agent respond without opening a ticket?

The agent can respond without opening a ticket when it finds an authorised instruction that matches the query, has the information needed to apply it, and does not need to make a decision, grant an exception or interpret information not covered by the documentation.

The practical approach is to separate straightforward queries from cases that require analysis. The agent can respond when:

• The answer is documented and matches the customer’s situation.
• The customer provides the information needed to follow the procedure.
• The request does not involve a complaint, exception or authorisation.
• The answer does not change the company’s terms, charges or commitments.

If information is missing, the documentation is contradictory or the customer needs a decision, the agent must not fill in the gaps based on intuition. It should ask for the missing information or prepare the handover. If the query arrives by email, the same approach can be applied with an email agent: respond only with supported information and hand the case over to a person when it is not covered by the documentation.

What data should an agent request before opening a ticket?

Before opening a ticket, the agent should collect only the data needed to identify the customer, understand the problem and allow a person to continue handling the case: identity, affected product, description, context, available evidence and preferred channel for receiving a response.

The conversation should follow a simple order:

• Identity: name, company and contact detail associated with the account.
• Affected product: service, order, contract or device related to the enquiry.
• Description: what is happening, when it started and what error message appears, if any.
• Context and evidence: steps taken, screenshots, documents or order number.
• Response channel: phone, email or WhatsApp support, according to the customer’s preference.

The agent must confirm the details before creating the ticket and avoid asking for information that does not help diagnose the problem. If an essential detail is missing, it must request it before passing the case to a person. The ticket must clearly state the request and the next step expected from the person handling it.

When should the case be handed over to a person?

The agent should hand the case over to a person when it cannot answer with confidence, the request requires a decision, authorised information is missing, an urgent incident arises or the customer asks for human support. Continuing with an automated response could leave the problem unresolved or create the wrong expectation.

Define clear rules for handing cases over:

• Complaints, billing disputes or requests for exceptions.
• Urgent incidents that could affect the service or the customer.
• Questions outside the available documentation or involving conflicting data.
• Cases where the customer asks to speak to a person.
• Situations that require reviewing private information or making a decision.

The agent should explain what will happen next and keep the context it has collected, so the team does not make the customer repeat it. If the enquiry comes by phone, the voice agent can be part of the first point of contact, with a clear rule for handing the case over to the team when necessary.

How do you review the support agent’s performance?

The team should regularly review conversations, open tickets, handed-over cases and unanswered questions. This helps correct the documentation, adjust handover rules and add instructions for repeated queries before the agent handles them again.

The review can follow this order:

• Group the questions the agent could not answer and decide whether documentation or a rule is missing.
• Check whether the responses resolved the query or led to another interaction.
• Review open tickets and handed-over cases to identify information that was missing at the start.
• Update the instructions when products, processes, terms or owners change.

It is not enough to measure how many conversations the agent handles. The team should read a sample of cases and correct the authorised source, rather than creating separate responses. It is also useful to record who approves each change and when it takes effect. If you need help setting up this process, see the implementation packs.

Frequently asked questions

Can an AI agent respond if the information is not in the documentation?

It should not make an assumption if the documentation does not contain the information needed. It should state that it cannot confirm the answer, collect the customer’s context and hand the case over according to the defined rules. This prevents incorrect instructions and creates an opportunity to expand the knowledge base with that query.

Can the customer ask to speak to a person at any time?

Yes. The support flow should treat this request as a handover rule. The agent can collect the reason and basic details before handing over the case, but it should not hold up the conversation or make the customer repeat information unnecessarily. The company must decide what happens if no person is available at that moment.

What is the difference between answering a query and opening a ticket?

Answering a query means providing an authorised instruction that resolves the question at that moment. Opening a ticket records a case that needs follow-up, a decision or human intervention. The ticket should include the context collected by the agent, the customer’s information and the reason why an automated response is not enough.

What types of issues should not be automated?

Complaints, sensitive requests, exceptions to the company’s terms and issues with financial or contractual consequences should not be automated without human review. Urgent, ambiguous or undocumented cases should also be handed over. The agent can collect the initial information, but a person must assess the response and decide on the next step.

Can the support agent be used on WhatsApp, email and phone?

The choice of channels depends on where your company receives queries and how the team will continue handling each case. Before setting it up, define what information the agent can access, what data it must collect and how a person is notified. The same criteria should apply across all channels to prevent contradictory responses.

Define your company’s response and ticket flow with UTIO’s support agent

Sources

• Vertical de soporte — utio.io