How to Build a Cross-Border AI Chatbot with Coze: Automate Support & Slash Costs

Discover how Coze AI agents, now integrated with SaleSmartly, empower businesses to automate customer interactions on platforms like Facebook, WhatsApp, and more—no coding required. This guide walks through creating, customizing, and deploying AI agents to streamline workflows, reduce expenses, and boost efficiency. 

What is Coze?

Coze is a cutting-edge AI agent development platform with an intuitive interface and robust backend capabilities, making AI technology accessible to everyone. Even users without programming experience can effortlessly create AI agents powered by large models. These agents can understand and generate natural language, handle complex interactions, and be seamlessly deployed across social media, messaging apps, and websites.  

Coze stands out for its ease of use and high flexibility. With a visual interface, users can customize AI agents by selecting functional modules that suit their needs. These agents continuously learn and adapt, delivering personalized experiences. For businesses, Coze’s open and user-friendly design enables rapid AI implementation, boosting service efficiency while reducing operational costs.

Advantages of AI agents over traditional AI

Autonomy: AI agents are able to perform tasks independently without constant input or guidance from humans.

Interactivity: Agents can interact with humans or other agents using natural language. This interactivity makes the agents more friendly and easy to use.

Adaptability: The AI agent is able to continuously learn from past conversations and improve its performance and decision-making process.

Task decomposition and planning: The agent can decompose complex tasks into smaller, manageable subtasks and plan them effectively.

Memory: The agent has short-term and long-term memory, which enables it to remember key information related to the task and better understand the context.

Tool Usage: AI agents are able to operate and use external tools, such as Internet queries, code execution, etc.

Coze international vs. Chinese version: Key differences explained

Model compatibility: Currently, the Chinese version is only compatible with the "Skylark Large Model" as its dialogue engine, while the international version is compatible with large models such as "GPT 3.5/GPT 4" to drive dialogue. 

Integration capabilities:  Chinese users can directly integrate the software into popular platforms such as WeChat and Lark, while international users can seamlessly integrate it into widely used platforms such as Facebook Messenger, Slack, Telegram, etc.

Conditions of use: The international version requires the use of scientific Internet access tools, while the Chinese version does not.

How to build a Coze AI agent?

Since the operation pages for both the international and Chinese versions are identical, this article will use the Chinese version as an example for clarity. Any differences between the two versions will be highlighted with additional explanations.

1. Create an AI agent

Log in to Coze, click “➕” in the upper left corner of the page, enter the name of the agent and function description, and then click the Generate icon next to the icon to automatically generate an avatar.

Standard Creation

After entering the name, function description and icon of the agent, click [Confirm], and then you will be redirected to the editing page.

AI Created

You describe to the AI what kind of intelligent body you want, and the AI will help you generate a name, function description, and icon. AI will also help you configure some editors based on the configuration, which can be changed according to personal needs later.

2. Edit AI Agent

The edit page is mainly divided into the following sections:

Character setting and response logic

• Contextual understanding of natural language processing: Personality helps AI agents better understand and process natural language. Through preset personality traits and response logic, AI can generate more appropriate and natural responses in specific contexts, which is an important research direction in the field of natural language processing (NLP).
• Machine learning and adaptability: Response logic often involves machine learning algorithms, especially decision trees and probabilistic models when processing user queries. These algorithms need to be adjusted according to the parameters of the persona to ensure that the AI agent's response matches its established "personality" and expected behavior.

Skill

(1) Plug-ins

External APIs can be called to give the agent more new capabilities, such as searching for information, browsing web pages, generating images, etc., and the capabilities and usage scenarios of the agent can be expanded by adding different plug-ins.

The plug-ins of the Coze international version and the Chinese version are different. Coze International version has more software that can only be used overseas, such as Google Search, Dalle3 Drawing, Github, GPT4V, etc.

(2) Workflow

Workflow supports the combination of plug-ins, large language models, code blocks and other functions in a visual way, thereby achieving complex and stable business process orchestration, such as travel planning, report analysis, etc.

(3) Image Stream

The principle of image flow is the same as that of workflow, except that it is obtained from text instead of pictures.

In Coze (international version), both of them are in one column, while in Coze (Chinese version), they are separated into two columns.

(4) Trigger

Allow users to trigger corresponding processes based on time and events in the conversation.

Knowledge

Coze's knowledge base not only supports uploading and storing various external knowledge content, but also allows you to upload text and table data from a variety of data sources such as local documents, online data, Notion, and Lark documents when you cannot find the answers you need in existing AI resources or when you need to mine answers from specific information sources. Coze not only provides storage and management functions for external data, but also helps users efficiently find accurate information from their own customized knowledge base through enhanced retrieval functions, ensuring that the required knowledge and data can be obtained even outside the AI's standard answer library.

Memory

Coze’s memory refers to the software’s ability to store and recall external data to provide accurate responses when interacting with users. This function is achieved in three main ways:

(1) Knowledge base: It contains a large number of knowledge segments. For example, for vehicle model data, each segment of the knowledge base stores the basic data of a vehicle model. When a user asks for information about a specific vehicle model, such as "What is the price of a BMW X3?", the system can match the knowledge segment of the corresponding vehicle model.

(2) Variables: Record user preferences or information in the form of key-value pairs. For example, you can create a variable called "language" to remember the language used by the user and use this language as the preferred language in subsequent conversations.

(3) Database: This function is similar to the database in traditional software development. It stores information in the form of tables and operates through NoSQL. Currently, Coze provides a relational database. The database can be used in a variety of application scenarios, such as:

AI Notes: Allows recording of notes submitted by users, and supports submission and query operations. These operations are completed through natural language to SQL (NL2SQL) technology.

Vocabulary book: It can record the words learned by the user, including the words that have been remembered and the words that have not yet been mastered.

Conversational experience

The concept of this feature is relatively easy to understand. Users can customize the initial greeting and quick action buttons. When someone clicks on your robot, the system will automatically show them the preset conversation content, which can help them quickly grasp the main functions of the robot and enhance their understanding of how the robot can serve them. Users can convert frequently asked questions into quick buttons. With a simple click, you can activate the conversation on related topics and achieve fast and efficient communication.

3. Preview and debug

After configuring the above information, you can start debugging. If the agent does not respond as expected, is too slow, or even does not respond, you can also use the debugger to view the agent's execution details and troubleshoot the problem.

4. Click "Publish"

After previewing and debugging and finding no problems, you can click the [Publish] button.

How to effectively integrate Coze into your business?

After mastering the creation of a Coze robot, the next step is to explore how to seamlessly integrate this robot into existing business processes to maximize its potential for improving work efficiency and service quality. In daily operations, the areas that benefit most from robotic assistance are pre-sales and after-sales services, both of which involve significant communication. 

AI robots can alleviate communication pressure by handling common customer inquiries. However, relying solely on AI for these functions is not a good way, as robots may struggle to answer all questions and may find it challenging to facilitate conversions based on customer circumstances. Therefore, the optimal approach is to connect Coze's robot to SaleSmartly, an omnichannel private domain communication tool. This integration allows for routine inquiries to be managed by AI technology while ensuring a smooth transition to human agents when manual intervention is necessary. Additionally, during peak periods when human agents may be overwhelmed, AI can assist by addressing basic questions to reduce the overall workload. This hybrid communication strategy enhances the customer experience by making it more personalized and efficient. 

Recently, Coze was officially launched on SaleSmartly’s app store. SaleSmartly customers can connect a Coze AI to their chat plugin in a few simple steps: Tutorial on configuring Coze bot

Further reading

What is Coze? - Coze official documentation

Tutorial on configuring Coze bot - SaleSmartly - Help Center - SaleSmartly User Guide

Tutorial on configuring button robots - SaleSmartly - Help Center - SaleSmartly User Guide