Skip to main content

How to get started with AI in Xentral

  • Updated

Finding the AI features in Xentral

Let’s get started - here you can see where to find the AI features and how to open them.

Steps:

  1. Log in to your Xentral instance.

  2. In the main menu on the left, you find an AI icon that opens the Agent Hub. You start on the Home page with your daily overview.

  3. In addition, you find another AI icon (Co-Pilot) at the top right next to your profile. Click on this icon to open the AI chat as a sidebar and work directly with the Co-Pilot (Chat). The sidebar is available everywhere in Xentral.

dashboard_AI_xentral_startseite_dashboard_002.png

The Agent Hub is divided into four tabs:

Tab

Description

Home

Your daily overview: Here you can see how many cases are waiting for your approval, open recommendations from the analysis of your requests, your Inbox address in the Ready in 2 steps area, as well as your AI colleagues and your custom agents.

Tasks

All requests your agents have processed. Here you review the agents' proposals, correct them if necessary, and approve them.

Knowledge base

The knowledge your agents work with, for example rules, product information, and reply templates.

Settings

Email integration and block rules for your AI agents. You can learn more in the article Agent-Hub: Setup & Settings.

In the Your AI colleagues area, you can see for each agent what it can do (CAN) and which data it accesses (SEES). The number on the card displays the agent’s open cases. Click on a card to display the tasks of this agent directly.

Getting started with the AI Co-Pilot (Chat)

Tip

Use the AI chat when you want to actively ask questions about your business, get analyses, or trigger tasks for your agents.

The AI Co-Pilot is your chat assistant in Xentral. It accesses your data in Xentral and your knowledge base, creates analyses, and answers questions about using Xentral. The Co-Pilot opens as a sidebar on the right-hand side of the screen and is available in all areas of Xentral – you do not have to leave the area you are currently working in.

Opening the Co-Pilot

Click on the AI icon at the top right next to your profile. The sidebar is opened with the Chat tab. Use the X icon to close it again.

chat_AI_xentral_startseite_dashboard_001.png

What the Co-Pilot can do

  • Retrieving data from Xentral: for example searching and summarizing customers, suppliers, items, orders, invoices, and other documents.

  • Creating analyses: for example revenue by time period or customer, as a table or chart if you like. For more in-depth analyses and KPIs, you need the Analytics module.

  • Answering questions about Xentral: The Co-Pilot uses the Xentral handbook for this, for example "How do I create a new shipping method?".

  • Using your knowledge: The Co-Pilot takes the entries from your knowledge base into account.

  • Creating tasks for your agents: for example having an order prepared (see Triggering actions via the Co-Pilot).

Asking the Co-Pilot questions

Steps:

  1. Open the Co-Pilot via the AI icon at the top right.

  2. Click in the input field "Ask me anything…" and phrase your request as specifically as possible, for example:

    1. "Give me an overview of my revenue for the last 14 days"

    2. "Which orders have not been shipped yet?"

    3. "Create a briefing for customer [Name]"

      Alternatively, select one of the suggestions below the input field or dictate your request using the microphone icon. Use the plus icon to upload a file, for example an order as a PDF.

  3. Send your request using the arrow icon.

    The Co-Pilot displays its answer directly in the chat. You can ask follow-up questions or refine your request in the same chat.

chat_businessuebersicht_inxentral_sidebar.png

Start with simple questions or clear tasks. The more specific your request, for example with a time period, customer, or item number, the more precisely the Co-Pilot can help you.

Triggering actions via the Co-Pilot

When you trigger an action for a customer via the chat, for example "Create an order for customer [Name] with item X", the Co-Pilot does not carry it out itself. Instead, it creates a task for it – just as if an email had come in. The matching agent prepares the action and you approve it in the Tasks tab.

You can also copy existing content, for example a customer email or an order, into the chat and have a task created from it.

When you tell the Co-Pilot a rule or piece of information your agents should remember, for example "Customer X is an internal customer and should be ignored in revenue analyses", it suggests saving it as an entry in the knowledge base. After your confirmation, the entry is displayed in the knowledge base with the source Co-Pilot.

Managing chats

  • New chat: Starts a new conversation. It is best to always start a new chat for a new topic.

  • History (clock icon): Displays your chats from today and the last seven days. Click on a chat to continue it.

  • Rename and Delete: Click on the arrow next to the title of the opened chat to rename or delete the chat.

Tasks in the sidebar

In the Tasks tab of the sidebar, you can see your tasks from the Agent Hub – with the same search and filter options as in the Tasks tab of the Agent Hub. This way, you can review your agents' proposals without leaving the area of Xentral you are currently working in.

Getting started with customer inquiries (emails)

Tip

Use the email agents when incoming customer inquiries should be automatically analyzed, structured, and prepared for you to process.

Tip

How proposals and recommendations are created

For each inquiry, you automatically receive a proposal, a recommended action, and matching actions, for example a reply draft. These are based on three central sources:

  • your customer’s specific inquiry

  • the data available in Xentral (for example orders, customers, status)

  • your company guidelines stored in the knowledge base

From this, the AI creates an assessment with a confidence score, a concrete recommendation for how to proceed, and proposed actions.

In the Ready in 2 steps area on the Home page, you can see how to get started: You forward a customer email to your hub address and the matching agent takes over automatically. You review the finished proposal and approve it. Only then are actions carried out.

mail_adresse_copy_xentral.png

How to get started:

  1. On the Home page, copy the Inbox address of your instance in the Ready in 2 steps area.

  2. Send a customer inquiry to this email address or forward an existing customer email.

    The email is automatically created as a task.

  3. Click on the Tasks tab at the top.

  4. Wait a moment until the inquiry has been processed and open the task.

  5. Review the agent’s proposal, correct it if necessary, and approve the proposed actions.

Tip

You can also create a task without an email: In the Tasks tab, click on the plus icon. In the Create New Task window, describe your task, paste a customer order, for example, and attach files if needed. Then click on Save & Start.

agent-hub-aufgaben-neue-aufgabe.png

Tasks at a glance

In the Tasks tab, you can see the list of your tasks on the left. For each task, the agent or category, the confidence score, and the risk, if applicable, are displayed. On the right, the Overview displays how many cases your agent needs you for, grouped by agent.

agent-hub-aufgaben-uebersicht.png

Element

Description

Search

Searches your tasks.

Filter icon

Filters the tasks by Confidence, Risk (Low, Medium, High), Agents, and Tags. Under the options for narrowing down, you can display only tasks with a comment or also display completed tasks.

Tag icon

Filters the tasks by tags. You assign tags in the opened task via Tag.

Overview icon

Displays the overview of open cases per agent on the right.

Refresh icon

Reloads the list.

Plus icon

Opens the Create New Task window.

Tasks / Done

Open and completed tasks.

More

Further lists: Queue (requests that are currently being processed), Errors, Spam, Not relevant, and Deleted.

Sort

Defines the sort order of the list.
agent-hub-aufgaben-filter.png

Reviewing and approving a task

When you open a task, the entire case is displayed on the right:

agent_handlungsempfehlung_002.png

  • Initial request: Your customer’s message. Use Original, DE, and EN to switch between the original language and a translation. Use Show full email to open the entire email.

  • Agent’s proposal: The agent describes what it proposes, summarizes the customer message, and gives a decision recommendation.

  • Agent steps: Displays which data the agent searched for and found in Xentral.

  • Proposed actions: The actions the agent has prepared for you, for example Send reply or Create sales order. Affected documents from Xentral are displayed as a link. Use the pencil icon to edit an action and the trash icon to remove it. For replies, select the language via Original, DE, or EN.

In the header of the task, you also find the Message History with all previous messages from this customer, the category of the task, Tag, the task number, the time of receipt, the confidence score, as well as icons for re-running and deleting the task.

Steps:

  1. Review the proposal and the proposed actions.

  2. If something does not fit, give the agent an instruction in the input field, for example "Max. 2% cash discount" or "Also refer to our terms and conditions". The agent then adjusts its proposal.

  3. Click on Run action or, if there are several actions, on Run all.

    Alternatively, use the arrow next to the button to select the manual option with a template. You then edit the email yourself before sending it and can use a template from your knowledge base.

    The actions are carried out in Xentral and the task is completed.

Important

The agents do not carry out any actions in Xentral without your approval. Only when you click on Run action or Run all are replies sent or orders created, for example.

Getting started with your knowledge base

With the knowledge base, you train your AI agents and continuously improve their responses.

  1. Open the Knowledge base tab in the Agent Hub.

    Below the heading, you can see how many entries you have created and how much space is still available (a maximum of 2,000 entries).

  2. Get an overview of your existing entries. Using the filters, you can filter by type (Expert knowledge, Company, Templates) and source (Manual, Website, Chat, Copilot, File, Suggestion).

    Learning-Loop.png

  3. Click on the edit icon to adjust an existing entry, or on Add entry to add new knowledge. You have three options:

    1. New entry – manually create an entry with Title / Question and Content / Answer

    2. Upload file – have knowledge extracted from a PDF file. You receive a preview of the entries found and decide which ones to accept.

    3. Import website – have knowledge extracted from a website. Enter the URL and the maximum number of pages to be searched and click on Analyze.

      agent-hub-wissensdatenbank-eintrag-hinzufuegen.png

  4. For manual entries, select the appropriate type:

    1. Company for general information (for example communication style, processes)

    2. Expert knowledge for product information, FAQs, or specific answers

    3. Templates for reusable reply templates

  5. Phrase your knowledge entry as a clear question and answer or instruction.

  6. Click on Create so that your agents can use the entry right away.

In addition, entries are created automatically from your chats with the agents: When you correct an agent in a task (for example "Max. 2% cash discount"), the system suggests saving this rule as a knowledge entry. You can recognize such entries by the source Chat. You can recognize entries from the AI Co-Pilot by the source Co-Pilot.

Use Export to download all entries of your knowledge base.

Recommendations from requests and knowledge base

The Agent Hub analyzes your previous requests and your knowledge base and displays where you can improve. You start the analysis on the Home page via the recommendations banner or in the Knowledge base tab via Analyze again. The analysis takes up to one minute. You can find the results in the Knowledge base tab in the Insights from requests & knowledge base area.

agent-hub-wissensdatenbank-empfehlungen.png

At the top, you can see how many cases were analyzed, how many of them had no reliable answer, and what share of these uncertain requests the recommendations address. The recommendations are divided into three categories:

  • Knowledge gap: A reliable knowledge entry is missing for one of your customers' questions. The Agent Hub proposes an entry to you. Fill in the placeholders and click on Accept to save it to your knowledge base.

  • Master data: Your master data in Xentral is incomplete or ambiguous, for example items that cannot be assigned to a product or duplicate customer accounts. Review the recommendation in Xentral and then click on Done.

  • Cleanup: Entries in your knowledge base contradict each other, are duplicated, or contain no usable information. Correct or delete the affected entries and then click on Done.

Use the X icon to hide a recommendation that is not relevant to you.

Note

Here are specific examples of company knowledge that is useful in the knowledge base:

Communication style:

  • "We always address customers informally."

  • "Responses should be friendly, brief, and solution-oriented."

  • "No emojis in a B2B context."

  • "Always start with empathy when handling complaints."

Company context:

  • "We are a D2C e-commerce company for furniture."

  • "Our main markets are Germany and Austria."

  • "Shipping is exclusively with DHL and DPD."

  • "Our brands are called [Brand1], [Brand2], [Brand3]"

  • "Our websites are [www.ourcompanyexample.com] and [www.ourcompanyexample2.com]"

Special rules & exceptions:

  • "No refunds for individually manufactured products"

  • "Goodwill decisions only for regular customers"

  • "Follow up proactively if tracking is missing after 3 days"

  • "Ignore shipping notifications from DHL"

Turn your past support emails into knowledge

Have you already answered many similar customer questions by email over the past few months? You can reuse this knowledge directly for your agents – instead of creating every knowledge entry one by one via New entry.

This approach is a good fit if:

  • Email threads already exist: from your support or customer service inbox.

  • Fill the knowledge base in one go: instead of capturing question-answer pairs one at a time.

  • Reuse answers that already work well: instead of writing them again.

To do this, you export existing support emails, have an AI tool condense them into a question-and-answer table, and upload that table to the knowledge base in one step. A first run typically takes 10 to 15 minutes.

  1. Export the relevant emails from your support inbox.

    In your email inbox, filter for emails from the last 90 days in your support or customer service folder where customers asked you questions and you replied. The fastest way to capture several complete conversations at once depends on your email program:

    • Outlook (desktop, Windows): Select the relevant emails in the list and drag them into an empty Word document while holding down the mouse button. Outlook automatically inserts the subject, sender, and full text of each email. Save the document afterward.

    • Outlook on the web: Select the relevant emails, use the print function, and save the result as a PDF file.

    • Gmail: Save individual conversations as a PDF via the print function. For larger volumes, use Google Takeout instead and limit the selection to Mail and the relevant inbox label.

    • Smaller volume (up to about 20 conversations), regardless of email program: Open each conversation individually, copy the full text, and paste it into a shared document.

      Tip

      The more real conversations you collect, the better the result. Even 15 to 20 email threads are enough for a good first run.

  2. Have an AI tool condense the emails into a question-and-answer table.

    Open an AI tool such as Claude, ChatGPT, or Gemini and copy the following analysis prompt into the chat window. Immediately afterward, paste your collected email text or upload the saved PDF or Takeout file, then send the message.

    You are analyzing a company's support email threads to build a knowledge
base for an internal AI agent. Real email conversations between the company
and its customers from the last 90 days follow this text.

Your task:
1. Identify recurring or typical questions and the matching answers from
   the emails.
2. Summarize each question as a clear, generally understandable question.
   Summarize each answer as a clear, general answer that would also fit a
   new, similar case.
3. Important: Remove all personal and customer-specific data from the
   answers (names, company names, order numbers, email addresses, amounts,
   phone numbers). The answer should be generally valid, not tied to a
   specific individual case.
4. Only include questions that occurred multiple times or are clearly
   typical for the business. Leave out one-off special cases with no
   repeat value.
5. If an email is unclear, contradictory, or incomplete, do NOT include it
   instead of guessing.
6. Output the result exclusively as a Markdown table with exactly these
   three columns: Question | Answer | Source
   In the "Source" column, always enter "Email support".
7. No explanations before or after, only the table.

Here are the email conversations:

[PASTE THE EMAIL TEXT YOU COPIED IN STEP 1 HERE]

    Important

    Before sending, check whether the pasted email text contains sensitive data that should not be shared with an external AI tool, for example contracts, internal prices, or health data. The prompt only removes personal data from the result. The raw text you paste in still contains everything you copied into it.

  3. Review the result.

    You get back a table with the columns Question, Answer, and Source. Check whether the answers are still accurate, whether customer-specific data has really been removed, and whether an important, frequent question is missing. Add any missing questions manually and delete rows that don’t fit. No rewording is needed at this point.

  4. Upload the reviewed table as knowledge.

    Save the reviewed table as a PDF file. Open the Knowledge base tab in the Agent Hub and click on Add entry > Upload file. Select your saved file and accept the suggested entries.

    Your agents use the uploaded content for their responses immediately.

Tip

Repeat this process whenever new support emails have come in, for example every few months. This way, your agents' knowledge grows along with your day-to-day business, without you ever having to type out a single email by hand.

Note

The knowledge base is limited to 2,000 entries. Only upload recurring patterns, not every individual email.

AI roadmap and your feedback

Note

The current Xentral AI roadmap and planned AI features can be found here: Xentral AI Roadmap →

Tip

Actively shape the AI future of Xentral!

We don’t just want to fill Xentral with features — we want to develop exactly the intelligent solutions that save you real time in your daily work.

Do you have processes that frustrate you every day? Are there data volumes that are barely manageable manually? Share your vision with us! The more detail you provide about your use cases, the better we can understand how AI needs to work for you.

Submit your use case now →

FAQ

What do the AI agents actually do for me?

The AI agents support you in automatically understanding, preparing, and processing incoming tasks — particularly customer inquiries.

They:

  • analyze content (for example emails and attachments)

  • recognize the concern (for example return, delivery status, product question)

  • access your ERP data and your stored knowledge

  • create a summary, a recommended action, and proposed actions, for example a reply draft or an order

You no longer have to manually gather information — you review and decide.

When do I use the AI Chat (Co-Pilot) and when do I use the email agents?

This is one of the most important distinctions:

AI Chat (Co-Pilot) → When you actively want to know or trigger something

Use the chat for:

  • Questions about your business (for example revenue, customers, orders)

  • Analyses and evaluations

  • Briefings (for example on customers or processes)

  • Questions about using Xentral

  • Manual actions (for example creating an order) – the Co-Pilot creates a task for your agents for this

The chat is your analysis and control tool.

Email agents → When a customer inquiry comes in

Use agents for:

  • Incoming customer emails

  • Automatic classification (for example return, delivery status)

  • Reply suggestions

  • Recommended actions and prepared actions

The agents work automatically in the background and take operational work off your hands.

Why couldn’t the chat answer my question?

Typical reasons:

  • The information is not available in the ERP or is not clearly assigned

  • The request lacks context (for example time period, exact name)

  • The Analytics module for in-depth questions and KPIs is missing

  • The use case is not yet covered

Important: The chat primarily accesses structured data + context.

How do I use the agents in my daily work?

Simply forward a customer inquiry to the provided email address (or set up automatic forwarding).

The process:

  • The email is created as a task

  • The inquiry is automatically classified

  • The matching agent takes over

You receive:

  • A summary

  • A decision recommendation

  • Proposed actions, for example a reply draft

Goal: Less manual processing, more focus on decisions.

How reliable are the results?

The agents are based on three central sources:

  • your customer’s inquiry

  • your ERP data

  • your stored knowledge (knowledge base)

In many cases, you already receive very good proposals. The confidence score and the risk assessment on each task display how certain the agent is.

Review the results before you approve them. Actions are only carried out when you approve them.

Quality continuously improves through:

  • your feedback and your corrections

  • your stored knowledge

  • the recommendations from requests and knowledge base

What happens if something is classified incorrectly?

It can happen that an inquiry is assigned to the wrong agent. For example, if a customer sent several requests in one email, or if it is not entirely clear which agent the inquiry should be assigned to.

In this case, you can:

  • review and correct the task manually

  • describe to the agent in the input field of the task what the inquiry is actually about

  • re-run the task using the corresponding icon in the header

This feedback is essential for improving routing and recognition.

Is my data used to train the AI?

No. The LLM providers used are contractually obligated not to use the input data for training or improving their own models or services. Consequently, the input data is neither used for their own model training nor stored. It is only processed to generate the output.

What should I definitely store in the knowledge base?

To ensure the AI responds reliably and in your style, you should maintain at least the following content:

  • Company data & signature (for example sender, closing phrase)

  • Website or relevant links

  • Tone of voice (for example formal/informal, tonality, example responses)

  • Product information (ideally as FAQs)

  • Rules & processes (for example returns, communication, special cases)

This is the most important foundation for good results.

Do I need to know how to "prompt" to use AI effectively?

No – but a few simple ground rules help enormously:

  • One task per request (don’t mix several topics)

  • Phrase clearly what you want

  • Provide context (for example customer, time period, goal)

  • Give examples if you expect a specific format

You don’t need technical knowledge – clear language is enough.

Can I submit feedback and feature requests for AI?

Your feedback plays a central role in the further development of the AI features. We welcome specific feature requests and use cases from your daily work. A lot is possible in the field of AI – however, our customers differ greatly in business model, company size, and technical affinity. This results in very different requirements and expectations. To set the right priorities, we use your feedback to better understand:

  • which features are needed most often

  • which use cases are relevant for different customer groups

  • where the greatest added value lies

On this basis, we develop the AI features in a targeted way.

You can share your ideas and specific use cases with us at any time via the familiar support or feedback channels.

Related articles

Visit the Academy - you'll find videos for most topics