AI chatbots and Microsoft Teams work together in Social Intents to create a support system where the AI handles routine questions instantly and human agents in Teams step in for complex issues. Below we cover how to configure AI chatbots when using the Teams integration, how escalation from AI to human agents works, and how to optimize the handoff experience.
How AI + Teams Works
When you enable an AI chatbot on a widget connected to Microsoft Teams, the conversation flow changes:
- Visitor opens the chat widget and sends a message
- The AI chatbot responds immediately using your configured AI engine (ChatGPT, Claude, or Gemini)
- If the chatbot can answer the question, the conversation continues without involving a human agent
- If the visitor asks to speak to a person, or the AI determines it cannot help, the conversation escalates to your Teams channel
- An agent in Teams joins the chat and takes over from the AI
The AI acts as a first-line filter. It handles common questions - product information, business hours, how-to guidance - and your team only gets pulled in when human judgment is needed.
Setting Up an AI Chatbot on a Teams Widget
Confirm Teams Connection
Make sure your widget is connected to Microsoft Teams. If not, follow the connection guide first.
Open Chatbot Settings
Go to My Widgets → select your widget → Chatbot tab.
Choose an AI Engine
Select your AI engine from the dropdown:
- ChatGPT (GPT-3.5 Turbo and GPT-4o free, or BYOK models up to GPT-5.2) - Best general-purpose option
- Claude (Anthropic, BYOK) - Strong with nuanced, detailed responses
- Gemini (Google, BYOK) - Good with factual, structured answers
Add Your API Key
Enter the API key for your chosen AI provider. See Connecting AI Engines for instructions on obtaining API keys from OpenAI, Anthropic, or Google.
Configure the System Prompt
Write a system prompt that tells the AI who it is and how to behave:
"You are a helpful support assistant for [Company Name]. Answer questions about our products and services. If you cannot answer a question or the visitor asks to speak to a person, let them know you will connect them with a team member."
Add Training Content (Optional)
Upload documents, paste URLs, or enter custom text to train the AI on your specific products, policies, and FAQs. The more relevant content you provide, the more accurate the AI's responses will be. See Training Your Chatbot for details.
Save and Test
Save your chatbot settings and open the chat widget on your website. Ask a few questions to verify the AI responds correctly, then ask to speak with a person to verify escalation works.
Escalation from AI to Teams
The most important part of the AI + Teams setup is ensuring smooth escalation. There are two ways escalation happens:
Automatic Escalation
Social Intents includes a built-in escalation mechanism. When the visitor says things like "I want to talk to a person," "connect me to a human," or "can I speak with someone," the system recognizes the intent and routes the conversation to your Teams channel.
Action-Based Escalation
For more control, create an Escalate Chat to Human AI Action (see Escalation Routing in Teams). This lets you:
- Route to specific channels based on conversation topic
- Customize the handoff message
- Add a confirmation step before escalation
- Include routing instructions for multi-department setups
What the Visitor Sees During Escalation
When escalation is triggered:
- The AI sends a message like "Let me connect you with a team member"
- The visitor sees a connecting state while Social Intents sends the invitation to Teams
- Once an agent joins in Teams, the visitor sees the agent's name and can continue the conversation
- The agent sees the full AI conversation history, so the visitor does not need to repeat their question
What Agents See in Teams
When a chat escalates to Teams, agents see:
- Visitor name (if provided)
- Page URL where the visitor started the chat
- Conversation summary - The AI's prior conversation gives the agent context
- Join button to enter the chat
This context eliminates the "how can I help you?" question - the agent already knows what the visitor needs.
AI Chatbot Behavior Settings
Several settings affect how the AI chatbot interacts with Teams:
| Setting | Effect |
|---|---|
| Chatbot enabled | AI responds to all incoming chats before involving agents |
| Chatbot disabled | All chats go directly to your Teams channel without AI interaction |
| Auto-close after chat ends | AI can close the conversation after resolution without needing an agent |
| Trigger once | Escalation action fires only once per conversation |
Common Configurations
AI-First with Human Backup
The AI handles all initial conversations. Agents only get involved when the AI cannot resolve the issue or the visitor requests a human. This is the most common setup and works well for teams that want to reduce chat volume.
- Enable AI chatbot
- Train it on your knowledge base
- Create an "Escalate Chat to Human" action
- Set your Teams channel as the escalation target
AI Triage + Department Routing
The AI identifies the visitor's need and routes to the correct department channel. Create multiple escalation actions, each targeting a different Teams channel, with clear routing instructions to help the AI choose correctly.
After-Hours AI Only
During business hours, chats go directly to Teams with no AI. After hours, enable the AI chatbot to handle conversations and collect contact information. You can configure operating hours in the widget settings to switch between these modes.
Best Practices
- Train the AI thoroughly - The better trained your chatbot is, the fewer chats need human intervention. Upload your FAQ, product docs, and common support answers
- Write clear escalation triggers - Tell the AI exactly when to escalate: "If the visitor asks about refunds, enterprise pricing, or anything you are not sure about, escalate to a human"
- Set expectations - Include in the AI prompt: "If you need to connect the visitor to a team member, let them know someone will be with them shortly"
- Monitor AI performance - Check the dashboard for chats where the AI escalated. Look for patterns - if the AI frequently escalates for the same questions, add that content to its training data
- Keep the AI prompt focused - Do not try to make the AI handle billing disputes or complex account issues. Let it answer informational questions and route everything else to your team
Preguntas frecuentes
Does the AI use the Teams channel history for training?
No. The AI chatbot is trained on content you provide - uploaded documents, URLs, and custom text. It does not learn from your Teams conversations.
Can I disable the AI for specific visitors?
There is no per-visitor toggle, but you can configure the chatbot to immediately escalate based on certain conditions. For example, if a returning customer's message contains "existing customer" or an account number, the AI can route them straight to a human.
Does the AI continue responding after an agent joins?
No. Once an agent joins the chat from Teams, the AI stops responding and the agent handles the conversation from that point forward.
Can I use different AI engines for different widgets?
Yes. Each widget has its own chatbot configuration. You could use ChatGPT on your marketing site widget and Claude on your support portal widget.
Next Steps
- Training Your Chatbot - Upload knowledge base content
- Escalation Routing in Teams - Set up department-based routing
- System Prompts and Behavior - Fine-tune your AI's personality