In the Agent Hub, each AI colleague handles one type of request. Here you can find out for each agent when it becomes active, what it proposes to you, and where its limits are. You can find a short overview of all AI features in the article Available AI features.
The following applies to all agents:
-
Nothing without your approval: The agents only prepare actions and replies. Nothing is carried out or sent until you approve the task.
-
Your rules take precedence: Rules from your knowledge base always win over the agents' defaults.
-
Internal data stays internal: No agent passes on stock levels, purchase prices, internal notes, or data of other customers to your customers.
-
Open points are named: If an agent cannot clarify something reliably, it is noted in the summary of the task instead of the agent guessing.
General customer service is the intake point of your Agent Hub: It reads every inquiry, distributes it to the responsible specialist agents, and finally combines everything into one task with exactly one reply email.
|
When it becomes active |
|
|
What it proposes |
|
|
What it does not do |
|
What it does on its own
-
Sorts out spam and pure notifications, for example newsletters and out-of-office replies. Senders that exist in Xentral as a customer or supplier never end up in the spam filter.
-
Takes into account the CRM history of the last 30 days and matching rules from your knowledge base.
-
Recognizes attachments such as invoices, payment advices, or purchase orders. If an email contains hardly any text, the attachment decides who is responsible.
-
Assesses the confidence and risk of the proposal (low/medium/high), mainly based on the amount of money.
What it needs
-
Forwarding of your mailboxes to your Agent Hub address and a sender account for the replies.
Example
"Where is my order 4711? And please send me 5 more packs of espresso beans." The first question goes to the Delivery status agent, the second one to the Sales order agent. You get one task: a new sales order and a reply with the tracking link and the order confirmation.
The Sales order agent captures orders from the email text or an order PDF and prepares the sales order, offer, or order change so that you only have to approve it.
|
When it becomes active |
|
|
What it proposes |
|
|
What it does not do |
|
What it does on its own
-
Finds the customer via customer number, email address, name, or address. It does not book forwarded orders, for example from field sales, onto the intermediary.
-
Finds products via product number, EAN/GTIN, ISBN, or manufacturer number, and via the name only as a last resort.
-
Checks for duplicate orders within the last 5 days, delivery blocks, and availability. It reports shortfalls to you but does not reject the order because of them, unless your rule says so.
-
Applies your defaults for project, payment method, and shipping method.
What it needs
-
Well-maintained customer and product master data as well as prices and price groups in Xentral.
-
Optional: rules in your knowledge base, for example "new customers always pay in advance".
Example
A field sales representative forwards a garden center’s PDF order: 10 × birdhouse, 20 × feeding station, different delivery address. The agent finds the garden center, applies the delivery address, and sees that only 6 birdhouses are available. It proposes the full sales order including a confirmation email and points out the shortfall to you.
The Delivery status agent answers "Where is my order?" with the current shipping status and a tracking link, without you having to look it up yourself.
|
When it becomes active |
|
|
What it proposes |
|
|
What it does not do |
|
What it does on its own
-
Finds the sales order via document number, external order number, the customer’s order number, transaction number, or the customer’s name.
-
Reads delivery notes, shipping providers, and tracking numbers for the sales order.
-
If there is no tracking number yet, the reply says that the order is probably still being processed, and you get a note.
What it needs
-
Sales orders with recorded deliveries and tracking numbers in Xentral.
-
An order number or at least the customer’s name in the inquiry.
Example
"My order 200345 hasn’t arrived yet." The agent finds a DHL shipment with the status "Out for delivery, expected tomorrow" and proposes a reply with this status and the tracking link.
The Returns agent handles returns, complaints, refund requests, and cancellations. It checks the case against your rules and prepares the appropriate solution for your approval.
|
When it becomes active |
|
|
What it proposes |
|
|
What it does not do |
|
What it does on its own
-
Recognizes whether the customer wants a replacement or their money back. If no wish is stated, the agent proposes a solution.
-
If the customer does not name a sales order, the agent picks the most likely one from the recent sales orders.
-
Checks existing returns and credit notes and gives you the document number if applicable.
-
Checks deadlines, warranty, and exclusions and justifies its result in one sentence, for example "purchased 12 days ago, within the 30-day window".
What it needs
-
Sales orders, deliveries, returns, and credit notes in Xentral.
-
Your return and warranty rules in the knowledge base. Without your own rules, it decides based on general standards.
Example
A customer reports that her kettle broke after two weeks and wants her money back. The agent finds the delivery from 14 days ago, applies your "30-day return" rule, and proposes a return, a credit note, and a reply with the next steps.
The Product question agent answers customer questions about your products, for example about dimensions, material, variants, or availability, directly from your product data.
|
When it becomes active |
|
|
What it proposes |
|
|
What it does not do |
|
What it does on its own
-
Finds products by the meaning of the question (for example "waterproof hiking jacket in blue") or via product number, EAN, category, or manufacturer.
-
Understands photos and reads values from labels.
-
Only looks up the customer and sales orders if the question is about an earlier purchase.
-
Does the research itself instead of asking the customer and answers without ERP jargon.
What it needs
-
Well-maintained product data in Xentral (descriptions, properties, variants).
-
The product catalog for the semantic search. It is imported once.
Example
"Is the Trail 30 backpack also available in green, and does a 15-inch laptop fit in?" The agent replies that green is available and that the laptop compartment fits devices up to 15.6 inches.
The Incoming invoice agent reads supplier invoices, matches them against supplier and purchase order, and prepares the liability with pre-accounting, booked the way you have booked this supplier before.
|
When it becomes active |
|
|
What it proposes |
|
|
What it does not do |
|
What it does on its own
-
Reads supplier, invoice number, date, payment terms, amounts, taxes, and line items and compares the VAT ID with the master data.
-
Uses the supplier’s last 20 liabilities as a template for nominal account, cost center, and booking text. Without history, it uses your chart of accounts.
-
Matches the line items against the purchase order and deducts partial invoices already booked.
-
Ends with "ready to execute" or with exactly one question whose answer you can save as a rule.
What it needs
-
Suppliers and purchase orders in Xentral, ideally some liabilities already booked for the supplier.
-
Optional: rules such as "supplier X always on cost center Y".
Example
A packaging invoice of €1,190 gross with two tax rates arrives as a PDF. The agent finds supplier and purchase order, recognizes an earlier partial invoice and thus a remaining delivery, and proposes one liability with two pre-accounting lines.
The Payment advice agent matches every line item of a payment advice against your invoices, credit notes, and open bank transactions and prepares the payment bookings for your approval.
|
When it becomes active |
|
|
What it proposes |
|
|
What it does not do |
|
What it does on its own
-
Finds documents via the invoice or credit note number, otherwise via payer name, period, and amount. It skips documents that are already booked.
-
Assesses cash discounts: settled within the deadline, shortly after it your internal tolerance decides, beyond that the invoice remains partially paid.
-
Names discrepancies such as missing documents or underpayments and overpayments.
-
Ends with the note that you can execute or with exactly one question whose answer you can save as a rule.
What it needs
-
Imported bank transactions as well as invoices and credit notes in Xentral.
-
Optional: booking rules, for example tolerances or how a payer writes your invoice numbers.
Example
A payment advice covers 7 invoices, paid with one bank transfer. The agent proposes 7 bookings. On one invoice, €12.40 more was deducted than the payment terms allow. It books the amount paid, leaves the rest open, and asks you how such cases are to be handled in the future.
If none of the built-in agents covers your process, you create a custom agent, without any programming. As soon as you activate it, it is assigned matching inquiries and also only prepares actions for your approval.
You can find out how this works in the article Creating custom agents.