First Overview

Tip

The AI Control Centre is the central location for making the performance and control of enneo's AI agents transparent. It shows how reliable the AI agents are, how high the degree of automation is, and where further potential exists.

Automations speed up the handling of customer concerns, reduce process costs, and relieve service staff who can concentrate on complex cases.

The control centre shows which business transactions are already being taken over by the AI, where human approvals make sense, and which steps towards fully autonomous processing are possible.

It makes this development measurable, traceable and controllable, and thus provides the key to gradually expanding automation in customer service.

Prerequisites for Viewing and Use

General Access: Permission in the Data and Reports → AI Control Centre area
Automation settings: Access to configuration, ticket detection and triggering is controlled under Dark processing permissions
Quality control: Access to view, execute test runs, and their editing is controlled in the AI Quality Control area
Live Overview: Permission in the Tickets → Ticket Overview Page area
AI Agent Team: Access to view, create and edit is controlled in the AI Agent Management area

Tools Overview

The AI Control Centre consolidates the most important tools for automation in one place:

Live Overview & Metrics: Overview of live metrics, degree of automation and AI agent performance → learn more
Quality Check: Management of test cases and evaluation of test runs → learn more
Automation Settings: Control of automation levels with and without approval → learn more

Levels of Automation

The goal is the fully autonomous processing of business transactions by AI agents.
The journey to this point proceeds through several stages:

Manual Processing The AI agent fully recognizes the request, correctly reads the relevant data, deduces appropriate processing options and presents them. A human decides on the next action(s).
Dark Processing with Approval The AI agent takes over all steps of the processing: It recognizes the request, reads the data, derives the right action and chooses the appropriate processing option. The checker only needs to approve the AI agent's selection with a click, or reset the process to manual processing if deviations are recognizable.
Dark Processing without Approval The AI agent takes over processing fully autonomously: It recognizes the request, reads the data, derives the correct action, selects the correct processing option and carries it out independently.

Info

The automation levels build on each other.

Only when the first stage works reliably makes sense to progress to the next.

It is therefore crucial in the first two stages not just to complete cases, but to analyse and evaluate them consciously.

In this way, it can be determined whether and at which points the AI agent should be further adjusted, or whether the next level of automation can be unlocked.

Example: Bank Data Agent

A rule-based AI agent was created for the purpose of processing customer concerns about bank details changes.

It is supposed to check whether the IBAN given by the customer is already registered in the system, whether it is invalid, or whether it can be adopted as a new, valid bank connection.

Initially, an agent manually decides on the performance of the AI agent:

Is the customer's concern correctly identified (change of bank details)?
Are all relevant parameters correctly extracted (IBAN, validity date, account holder)?
Are the typical use cases covered correctly? (IBAN valid, IBAN invalid, IBAN already exists, validity date in the past)
Are the correct processing options offered? (Adopt IBAN, inform customer about error in the IBAN, adjust validity date)

If the agent finds that individual points on the checklist are not reliably met, this is an indication that the bank data agent needs to be adjusted.

This includes adjustments to the detection of input parameters, optimizing the business logic or enhancements to the output handling. Only when this feedback has been implemented and the results are stable can the next step in automation be taken.

If the bank data agent consistently shows correct results, it can be set to Dark Processing with Approval.

If reliable results are also confirmed here, the transition to Dark Processing without Approval, i.e. fully autonomous processing, follows.

Prerequisites and Risks

Before an AI agent is transitioned into higher automation stages, certain conditions must be met:

Reliable Customer Recognition
Accurate Issue Recognition
Low Error Rates in recognizing and reading parameters
Successful Test Runs in quality testing

Risks arise mainly when incorrect decisions go unnoticed in the dark processing. The dashboard and quality control help to minimize these risks and detect errors in time.

Role of Users in the Control Centre

Users of the AI Control Centre have various tasks:

Create and maintain test cases to verify the reliability of the AI agents
Observe results and document and/or communicate irregularities
Give an assessment whether an AI agent can be transitioned to the next level of automation.