AI Agents & Automation
How AI agents work in enneo
An overview of the resolution plan, bot conversations and the roles of AI agents
Tip
AI agents handle recurring customer requests – such as new bank details, a change to the advance payment or a meter reading. How they go about it depends on the channel: for email and letter, an AI agent plans the entire ticket in a resolution plan that is executed with one approval. In chat and phone, the AI agent works live in the conversation and acts at the customer's request – without approval by service agents.
Three situations
| Situation | Who works on the request | Who confirms the actions |
|---|---|---|
| Email, letter and system tickets | A smart AI agent creates a resolution plan for the entire ticket. | A service agent confirms the plan once – or auto-processing releases it automatically. |
| Chat and phone with chatbot or voicebot | A smart AI agent conducts the conversation and calls other AI agents as tools along the way. | No approval by service agents – the AI agent acts at the customer's request in the conversation. |
| Chat and phone after handover to a service agent | AI agents are suggested to the service agent. | The service agent approves each AI agent individually. |
Email and letter: one plan for the entire ticket
An email often contains several requests – for example new bank details and a meter reading. Instead of approving a separate AI agent for each request, the ticket receives one resolution plan:
A ticket arrives
Enneo analyses the ticket and detects which smart AI agent is responsible.
The AI agent creates the plan
The AI agent reads the data it needs, for example from the contract. Anything that would change something – such as a change to the advance payment – is not executed but noted as a step in the plan. A prepared reply to the customer is created as well.
Review and confirm the plan
A service agent reviews the steps, adjusts them if needed and confirms the plan with Confirm resolution plan. Only then are the steps executed. If the plan is released for auto-processing, this happens automatically.
How to review a plan and decide on it is described in Review and confirm a resolution plan. How to trace why the AI agent suggests a step is described in Observability of AI agents.
Chat and phone: the AI agent in the conversation
As long as a chatbot or voicebot conducts the conversation, a smart AI agent works live with the customer. It uses the same tools and AI agents as for emails, but there is no resolution plan: it acts at the customer's request directly in the conversation.
- Chat: The action is executed if it matches the customer's request. If the AI agent has corrected a value, it asks first. If the customer declines or hesitates, nothing is executed.
- Phone: The action is executed immediately. The voicebot reads uncertain values back to the customer for confirmation first.
If a service agent takes over the conversation, suitable AI agents are suggested to them, which they approve individually – just as before the resolution plan was introduced.
The roles of AI agents
The lead smart AI agent. Every resolution plan is created by exactly one smart AI agent. Enneo decides which one in this order:
- the smart AI agent that was detected for the ticket,
- otherwise the associated AI agent of the subchannel (setting Associated AI Agent in the subchannel),
- otherwise the Base Agent.
Other AI agents as tools. The lead AI agent can call other AI agents – both smart and rule-based. A rule-based AI agent calculates its result without applying it; its action becomes a step in the plan. The lead AI agent decides when to call another AI agent based on that agent's detection prompt – see Tools of a smart AI agent.
The Base Agent. It takes over when no other AI agent fits and then creates a suggested reply based on similar requests and the knowledge base.
Smart and rule-based AI agents
When creating an AI agent, you can choose between two types:
| Smart reasoning logic (recommended) | Custom rule-based logic | |
|---|---|---|
| Setup | Instructions in natural language, without programming | Flow as code with fixed rules |
| Suitable for | Most requests; tickets with several requests; conversations in chat and phone | Strictly defined processes in which every step must be specified exactly |
| Role in the resolution plan | Can lead the plan | Is called as a tool by the lead AI agent; its action becomes a step in the plan |
Rule-based AI agents remain fully supported. Both types can be combined: a smart AI agent leads the request and calls a rule-based AI agent with a fixed flow as a tool.
Setting: ticket processing mode
Under Settings → Auto-processing, the setting Ticket processing mode determines how email, letter and system tickets are handled:
| Option | Behaviour |
|---|---|
| Solution plan (entire ticket, one approval) – default | One AI agent plans the entire ticket in several steps, which are executed together with one approval. |
| Compatibility mode (AI agents one by one) | As before: detected AI agents run individually and are approved individually. |
Info
Regardless of this setting, chat and phone always run in compatibility mode as soon as a service agent handles the conversation.