Agent Studio: How to Create and Manage AI Agents in Scoreapps

Scoreapps Agent Studio is the workspace where you create, configure, train, test and improve AI agents for conversations, lead capture, support, automations and the channels available in your account.

Important: Do not think of Agent Studio as a standalone chatbot. It is the workspace where you design how your agents behave, review how they work and improve them so they match your business goals.

What Agent Studio is

Agent Studio brings together the tools you need to work with AI Agents in Scoreapps.

From this area you can:

• Create new agents with the copilot.
• Open the visual builder to design workflows and automations.
• Review your existing agents from My agents.
• Check each agent's status and reconfigure it when needed.
• Prepare agents for support, lead capture, sales, follow-up or internal processes.

When to use Agent Studio

Use Agent Studio when you want to turn a conversational or repetitive process into an AI-guided experience.

Customer support

To answer frequently asked questions, explain services, resolve basic doubts or direct users to the right resource.

Lead capture and qualification

To ask for key details, detect purchase intent, recommend the next step and prepare a more structured sales conversation.

Process automation

To listen to events, interpret information, generate replies and trigger actions through the connectors and channels available in your account.

Before you create your agent

Before you create an agent, define the objective. Agents work better when their mission is clear.

• Goal: what the agent should achieve: inform, qualify, book, route, sell, support or collect data.
• End user: who will talk to the agent and what they expect to solve.
• Knowledge sources: which documents, articles, websites, FAQs or instructions the agent should use.
• Tone: formal, friendly, technical, commercial, direct or consultative.
• Channel: website, app, PWA, messaging or any other channel available in your configuration.
• Escalation: when the agent should route the conversation to a person or say that human review is needed.

Important: The clearer the goal, the easier it is to train, test and improve the agent. Avoid creating a generic agent that tries to do everything at once.

How to access Agent Studio

1. Log in to your Scoreapps account.
2. Open the AI tools area.
3. Select Agent Studio.
4. You will see the main screen with options to create agents, open the visual builder and review My agents.

Create an agent with the copilot

The copilot helps you create a first version of an agent from a simple description. This is the fastest way to start and then adjust the details.

1. From Agent Studio, choose the option to create an agent with the copilot.
2. Describe what the agent should do, such as answering service questions, qualifying leads or guiding users to book a call.
3. Include the sector, target audience and expected outcome.
4. Review the initial proposal generated by Scoreapps.
5. Adjust the name, approach and instructions before publishing or testing.

Useful example:

I want an agent for an aesthetic clinic. It should explain treatments, answer common questions, ask for name, phone number and treatment of interest, and recommend booking an appointment when the user is interested.

Start with the visual builder

The visual builder is the right option when you want more control over the flow or need to design a more advanced automation.

From the builder, you can organize steps, conditions and actions so the agent follows a process instead of only replying freely.

Use it when you need to:

• Separate different conversation paths.
• Create flows for lead capture, support, bookings or follow-up.
• Define what should happen when the user meets a condition.
• Connect the conversation with available actions or integrations.
• Reduce improvisation in sensitive processes.

What to configure inside an agent

Identity and purpose

Define the agent name, main function and type of user it will assist. This helps the agent keep the conversation coherent.

Instructions

Instructions are the core of the agent's behavior. They tell the agent what to do, what information to request, how to respond and when to escalate.

A good instruction usually includes:

• The agent's role.
• The conversation goal.
• The data it should collect.
• The questions it should prioritize.
• Topics it should not answer when it lacks enough information.
• When it should recommend talking to a person.

Knowledge

Add or prepare the information the agent needs to answer correctly: services, prices, conditions, FAQs, policies, processes, links and documentation.

Important: The agent does not replace a clear knowledge base. If the source information is incomplete, ambiguous or outdated, the answers can be too.

Conversation experience

Configure the first message, tone, useful links, data capture and interaction style you want to offer users.

Channels and integrations

Review where the agent will be used and which channels or connectors are available in your account. Depending on your configuration, you can use it on websites, apps, messaging, workflows or specific integrations.

How to write effective instructions

Instructions should be specific. Saying “answer customers” is not enough. Explain what the agent should do in real situations.

Recommended template

• Role: you are an assistant for [business/service].
• Goal: help the user achieve [main outcome].
• Data to collect: request [name, email, phone, need, budget, date, website, etc.].
• Style: reply in a [friendly, professional, brief, consultative] way.
• Rules: do not invent prices, deadlines or conditions if they are not provided.
• Escalation: if the user asks for something complex, recommend contacting the team.

Example

You are a sales assistant for a marketing agency. Your goal is to understand whether the visitor needs a website, advertising campaigns or AI automation. Ask short questions and collect name, email and current website if they have one. If the user shows real interest, recommend booking a call. Do not promise specific sales results or fixed quotes without team review.

Test the agent before publishing

Before activating the agent on a real website or channel, test it with normal questions and difficult questions.

1. Ask as if you were a new visitor.
2. Check whether it understands the business goal.
3. Validate whether it collects the required data.
4. Test out-of-scope questions.
5. Review whether it escalates properly when it does not know something.
6. Adjust instructions and knowledge until answers are consistent.

Useful test questions:

• What services do you offer?
• How much does it cost?
• I want to talk to a person.
• I do not know what I need, can you guide me?
• Can you guarantee results?

Manage your existing agents

In My agents, you can reopen agents you have already created and review their status. Use this area to keep your agents up to date, not only to create them once.

Review regularly:

• Whether the agent still matches the current business offer.
• Whether the FAQs have changed.
• Whether new instructions, services or conditions should be added.
• Whether the agent is collecting the right data.
• Whether it needs a more structured flow in the visual builder.

Best practices for better answers

• Create agents with one clear goal, not overly generic agents.
• Write instructions in clear language and include examples.
• Update information whenever prices, services, schedules or policies change.
• Test the agent before publishing it on a website or channel with real traffic.
• Define when it should ask for data and when it should stop insisting.
• Include a human contact path for sensitive or high-value cases.

Important checks before publishing

AI Agents help automate conversations and processes, but they should be configured carefully.

• No invented information: if the agent does not have enough data, it should say so or escalate.
• Sensitive information: do not add private, confidential or unnecessary data if the agent should not use it.
• Regulated areas: for legal, medical, financial or similar topics, configure the agent to inform and route, not to make final decisions.
• Human review: in important sales or support processes, use AI to qualify and prepare the conversation, not to fully replace your team.

Common mistakes and how to fix them

The agent gives generic answers

Add more specific instructions, conversation examples and a more complete knowledge base.

The agent does not ask for important details

Add a clear list of required data and explain when it should request it.

The agent replies about topics it should not cover

Add scope rules. Explain which topics it should avoid and when it should escalate to a person.

The flow becomes confusing

Use the visual builder to separate conversation paths and define clearer steps.

Expected result

When finished, you will have an agent created from Agent Studio, with a clear goal, solid instructions, useful knowledge and a flow ready to test or publish in the available channels.

The key is not to create the agent once, but to review and improve it with real usage. The better the instructions, content and follow-up, the more useful the agent will be for your business or your clients.

Agent Studio and internal errors

Agent Studio and Chatbots are different products, so their channels, testing methods and publishing options are not interchangeable. If a new workflow shows role > 255, step type required, an HTTP 500 error or a Roles conversion problem, record the route, action, time and technical ID. Include the steps and a screenshot without sensitive data so support can investigate the issue.