Should an AI Agent Answer Your WhatsApp Inquiries?

WhatsApp is where many buyers in Saudi Arabia, the UAE and Egypt send their first question, and it is where a slow reply is most visible. An AI agent that replies at 11 pm, in the customer’s language, is an obvious attraction. Since June 2026 it has also become an easy one: Meta now offers its own agent inside WhatsApp, and since July the Business Platform bills it as a separate message category.

The question for a business owner or sales director is not whether the technology works. It is which inquiries an agent should answer, which it should hand straight to a person, and what that costs. This article sets out the facts Meta has published, the rule WhatsApp applies to all automation, and a decision guide by inquiry type.

What Meta’s Business Agent is, and how it is priced

Meta announced Business Agent on 3 June 2026 at its Conversations event. According to the WhatsApp Business announcement, the agent answers questions specific to the business, recommends products from a catalogue, books appointments, qualifies incoming leads and helps close sales, replying in the customer’s language. Meta said more than one million businesses were already using it on WhatsApp and Messenger, and that the business decides when a team member steps in. TechCrunch reported the agent was being made available globally after almost two years of testing in markets such as India and Mexico.

There are two routes, and they are priced differently. If you are not sure which one you are on, start with our comparison of the WhatsApp Business app or the Business Platform:

• Through Meta’s own business apps, getting started was free at launch. Meta said that “in the coming months” businesses of every size would access the agent through paid subscription offerings, without publishing prices.
• Larger businesses can build on the Meta Business Agent Platform, which Meta describes as offering enterprise-grade controls, guardrails and measurement. On the WhatsApp Business Platform, Meta’s pricing documentation lists a fifth message category, effective 1 July 2026: any non-template message powered by the Business Agent Platform is charged at one global rate of USD 2 per million tokens. Meta adds that a message typically consumes 20,000–25,000 tokens, roughly 4–5 US cents per message.

Two details matter for budgeting:

• The free ad window does not cover the agent. When a customer messages you from a click-to-WhatsApp ad, a free entry point window of up to 7 days opens, and most message categories are free inside it. Meta states it continues to charge for Business Agent messages in that window, and volume tiers and free messages do not apply to the category.
• Human replies are no longer free beyond a threshold either. Each business number on the Platform now gets 1,000 free service messages a month before charges apply (see service-message pricing from October 2026). So the comparison is no longer “AI costs money, people are free”. It is the agent’s token cost against your service-message rate plus staff time.

The three routes a GCC business is usually offered

Professional recommendation: ask any provider to put its fee and Meta’s charges on separate lines of the quotation, and to show you where the handover rules are edited, before you compare routes on price.

The rule every WhatsApp automation must follow

The WhatsApp Business Messaging Policy (last updated 23 September 2026) allows businesses to use automation when responding during the 24-hour customer service window, but requires them to have “prompt, clear, and direct escalation paths” available. The policy lists what counts: in-chat transfer to a human agent, a phone number, email, web support on the business website, in-store visits, or a support form.

Two points follow from that wording. First, escalation is not optional, and it has to be prompt and clear: an agent that loops a frustrated customer through the same three options does not meet the spirit of the rule. Second, outside the 24-hour window you can only send approved message templates, so an agent cannot chase a quiet lead with free-form follow-ups.

A separate rule sometimes causes confusion. Since 15 January 2026, WhatsApp’s business terms have barred AI providers from using the platform to distribute general-purpose assistants where the AI is the primary product. As TechCrunch reported, Meta confirmed that a business using AI to serve its own customers, such as a travel company’s customer service bot, is not affected. An agent that answers questions about your products and services is the intended use.

Decision guide: which inquiries should an agent answer?

The useful question is not “AI or human” but “what is the agent’s job in this type of chat, and what triggers the handover”. The table below works for most B2B and service businesses.

The pattern is consistent: agents are good at gathering, sorting and confirming; people should own commitments, such as prices, technical assurances, apologies and exceptions. A hypothetical HVAC service company, for example, could let the agent book routine maintenance visits end to end, while any message mentioning a leak, a warranty or a previous visit goes straight to a coordinator.

Saudi data rules are moving towards automated processing

For companies serving Saudi customers, the regulatory direction is worth watching. The Saudi Data & AI Authority (SDAIA) has opened a public consultation on amendments to the Implementing Regulation of the Personal Data Protection Law. As summarised by law firm Al Tamimi & Company, one proposal would apply additional safeguards “where automated processing results in decisions that affect the rights or interests of data subjects”, rather than only to decisions made exclusively by automated means. Al Tamimi’s note, published on 9 October 2026, says the consultation is open until 5 November 2026.

These are proposals, not final rules. But the direction is clear enough to design for now. An agent that only answers questions and books slots carries little risk. An agent that scores leads, decides who gets a call back, or declines service affects people’s interests, and that is exactly where wider safeguards would bite. Practical steps:

• Keep a record of what the agent is allowed to decide and what it must pass to a person.
• Tell customers they are chatting with an automated assistant and how to reach a person.
• Do not let the agent reject or deprioritise inquiries without a human reviewing the rule behind it.
• Check SDAIA’s current text, and ask your legal adviser to review the final regulation once it is published, and apply equivalent care for customers in the UAE and Egypt under their own data laws.

Design the handover before you switch the agent on

Most failures with WhatsApp automation are handover failures. The agent does its part, then the chat sits in a queue nobody watches. Before launch, settle five things:

1. Triggers. Write the list of words, topics and situations that end the agent’s turn: price negotiation, complaints, safety, legal terms, and any customer who asks for a person.
2. Owners. Name who receives each type of handover, by working hours and outside them.
3. Context. Make sure the person sees the full chat and the details the agent collected, so the customer is not asked the same questions twice.
4. Promise. Decide what the agent tells the customer at handover, such as “A sales engineer will reply within working hours”, and keep that promise.
5. Source material. Give the agent approved answers only: catalogue, price lists you are willing to publish, datasheets and FAQs. If you would not put it on your website, do not put it in the agent. Fix the language rules too: Modern Standard Arabic or a named dialect per market, and an Arabic reviewer for the approved answers (see our guide to checking AI-written Arabic). Writing approved replies first is the groundwork an agent needs: for the fire-safety brand GCS, RX prepared WhatsApp replies for the most common inquiries and message flows from first contact to quotation request, with tone and structure guidance the whole team could follow.

The reply flows, qualifying questions and handover scripts are as much a sales design task as a technology one; our WhatsApp sales script for quotation requests sets out the qualifying questions for price requests. RX’s sales enablement work covers the WhatsApp, email and LinkedIn messaging your team uses, and the website is often where the approved source material should live first.

When the answer is “not yet”

An agent is the wrong first step if any of these is true:

• You do not have written, approved answers to your twenty most common questions.
• Your team cannot respond to handovers within the time the agent promises.
• Most of your inquiries are complex, high-value quotations where every chat goes to a person anyway.
• You cannot yet see which WhatsApp chats became sales, so you will not know whether the agent helped.

In those cases, greeting messages, quick replies, labels and a clear qualifying sequence will deliver most of the benefit at lower risk. Add the agent once the basics are in place and you can measure what it changes. Before switching an agent on, list your twenty most common WhatsApp questions with approved answers and mark which ones end in a handover. If that list does not exist yet, a business diagnosis session is where we build it with you: inquiry types, handover owners and the wording the agent may use.