WhatsApp AI agents connected to your ERP: what they can do, what they cost and how to start

In many markets, customers and suppliers already talk to your team on WhatsApp. They ask where their order is, whether an item is in stock, or what they owe. Someone opens SAP Business One, looks it up and types a reply. A WhatsApp AI agent connected to your ERP does that lookup and reply for the routine cases, and passes the rest to a person.

This guide explains what such an agent can do, how Meta charges for it and how to start without putting your number at risk.

The official route: WhatsApp Business Platform and the Cloud API

Businesses connect software to WhatsApp through Meta’s WhatsApp Business Platform. Today that means the Cloud API, hosted by Meta. The older On-Premises API was switched off on 23 October 2025, so new projects should use the Cloud API, either directly or through a Meta Business Solution Provider.

You need a Meta Business portfolio, a WhatsApp Business Account, a phone number that is not already active on the normal WhatsApp app, and business verification for higher sending limits.

How messaging and pricing work

Two ideas control everything: the 24-hour window and templates.

• Customer service window. When a customer messages you, a 24-hour window opens. Inside it you can send free-form replies. Each new customer message resets the clock.
• Templates. Outside that window, you can only send pre-approved template messages, and only to people who opted in. Templates fall into three categories: marketing, utility and authentication.

Since 1 July 2025, Meta charges per delivered template message instead of per conversation. Current rules, from Meta’s pricing page:

Rates vary by category and by the recipient’s country, and utility and authentication rates drop with monthly volume. Meta changes rates several times a year. In 2026 alone there were updates in January, April, July and October, including higher utility and authentication rates for Pakistan from 1 April 2026. Check current pricing on Meta’s rate cards for your markets before you budget.

For an ERP agent this is good news. Most of its work is answering questions customers send first, which falls inside the free window.

What the agent can do with ERP data

• Order status. “Where is my order 4512?” The agent checks the sales order and delivery in SAP B1 and replies with status and date.
• Stock check. A dealer asks if 200 units of an item are available. The agent checks available quantity by warehouse.
• Quotes. The agent collects items and quantities, applies the customer’s price list, and creates a draft quotation for a salesperson to approve and send.
• Payment reminders. A utility template reminds customers of overdue invoices, with the amount and a statement on request.
• Approvals. A manager gets a purchase order or discount request on WhatsApp and replies approve or reject. The agent records the decision in the ERP.
• Purchasing. The agent can ask suppliers for prices and delivery dates and log their replies against a purchase request.

Voice notes can be transcribed and handled the same way, which matters where many customers prefer speaking to typing.

How it links to SAP Business One or your ERP

 Customer / supplier / staff on WhatsApp
        |
        v
 WhatsApp Cloud API (Meta)  --webhook-->  Agent backend
                                          - identify sender by phone
                                          - AI model + fixed tools
                                          - rules, limits, logging
                                          - handover queue
                                                 |
                                                 v
                              SAP B1 Service Layer / read-only views
                              (orders, stock, balances, draft quotes)

The AI model never talks to the ERP directly. It can only call a fixed set of tools in the backend, such as get_order_status or create_draft_quotation. The backend matches the sender’s phone number to a business partner contact, and only returns that customer’s data. Writes go in as drafts or pending approvals, not posted documents.

Human handover

An agent that cannot hand over will annoy customers. Build handover in from day one:

• A clear phrase like “talk to a person” always works.
• The agent hands over on complaints, price disputes, large orders or when it is unsure.
• Staff see the full conversation in a shared inbox and reply from the same number.
• While a person is handling a chat, the agent stays quiet.

Data privacy

• Only answer a phone number linked to that customer in your ERP. Ask for an extra check, such as an invoice number, before sharing balances.
• Send the model the minimum data needed for the answer.
• Log messages and tool calls, and set a retention period that fits your local law and contracts.
• Check your AI provider’s terms on data use and storage, and put a data processing agreement in place where needed.
• Respect opt-outs. Someone who says stop should not get templates.

Realistic costs and steps

Costs come in four parts: Meta’s template charges, any Business Solution Provider fee if you use one, AI model usage per message, and building and maintaining the backend. Meta and AI model rates change, so check current pricing and estimate from your own message volume. A pilot with one use case keeps all four small.

1. Pick one use case. Order status or stock check is a good first one.
2. Set up the WhatsApp Business Account on the Cloud API with a dedicated number and business verification.
3. Write and submit templates for anything you send first, such as reminders.
4. Build the backend with fixed ERP tools, phone-to-customer matching and logging.
5. Add handover to a shared inbox.
6. Test with staff, then a few friendly customers. Read every conversation in the first weeks.
7. Expand to quotes, reminders and approvals once the first use case is stable.

Mistakes to avoid

• Unofficial WhatsApp libraries. Tools that automate the normal WhatsApp or WhatsApp Business app break WhatsApp’s terms. Numbers get banned, sometimes overnight, and you lose the number your customers know. Use the official API.
• Building a general chatbot. Meta’s terms from 15 January 2026 bar general-purpose AI assistants from the business platform. Keep the agent focused on your own business.
• Letting the AI post documents. Drafts and approvals first.
• Marketing through utility templates. Meta reviews categories. Misuse can lead to rejected templates or recategorisation.
• No handover. Customers stop using it after one bad experience.

We build these agents as part of our WhatsApp AI agents and AI agents for ERP services. See how one works in our WhatsApp AI sales and purchasing agent case study.

If you want customers to check orders, stock or balances on WhatsApp, contact us with your ERP and the first use case you have in mind.

Sources

• Meta: Pricing on the WhatsApp Business Platform
• Meta: On-Premises API sunset
• TechCrunch: WhatsApp changes its terms to bar general-purpose chatbots (Oct 2025)
• SAP Learning: Service Layer API terms and documentation

Related services

• WhatsApp AI agents
• AI agents for ERP