How does the AI agent know what it's allowed to answer?

It answers only from the sources you approve, such as the website, your documents, the CRM or the project tool. Anything not approved, it doesn't know. For sensitive categories, approval gates apply on top: those answers only go out once someone on the team has checked them.

What happens when the agent can't answer a question?

It says so honestly and hands the conversation to your team, in the same chat and with the history so far. It makes no answers up. All the customer notices is that a person is now writing back.

Does the agent replace my support team?

No. It catches the recurring questions, such as those about status, documents, appointments and processes, so your team has time for the real work. Advice and anything substantive stay with people. Handoffs happen in the same chat.

How much upkeep does the agent need in operation?

The agent is as current as its knowledge base. When services, prices, responsibilities or processes change, the approved sources have to follow, or it gives outdated answers. We handle operation and monitoring; keeping the sources up to date stays a shared task.

WhatsApp customer support with AI that does more than a chatbot

Q: Is WhatsApp customer support with AI GDPR compliant?

Yes, with the right setup: WhatsApp Business Platform, a data processing agreement under Art. 28 GDPR, clear data boundaries and EU data storage. The private WhatsApp app on a company phone doesn't meet these requirements. Details like deletion periods and the list of approved sources are set during onboarding.

WhatsApp customer support with AI means this: an agent answers customer questions right in the chat, drawing on your company's knowledge, meaning exactly the sources you approve. Status questions, missing documents, appointment requests and the usual back-and-forth get answered on the spot, even at 7pm and on weekends. Anything it isn't sure about, it hands to your team, in the same chat. That is what sets it apart from classic chatbots, which run on decision trees and canned text.

Being wary of anything that sounds like a bot is fair enough. Almost everyone has been stuck in a bot menu, and so have your customers. The gap between that experience and an agent that knows your company comes down to one question: where does the answer come from?

Why classic chatbots frustrate customers

Classic chatbots frustrate customers because they don't know the company. They follow a decision tree someone built months ago and serve up canned text that rarely fits the actual question.

It always plays out the same way. The customer types their question. The bot spots a keyword and sends a link to the FAQ page. The customer rephrases, the bot offers three buttons, none of which fit. By the time "Was this answer helpful?" pops up on the screen, the mood has hit rock bottom. The customer is stuck in a menu, and their actual question is still open. That's the dead end that earned chatbots their reputation.

The reason runs deeper than clumsy conversation design. A decision-tree bot has no access to the company's knowledge. It doesn't know where an order stands or which documents are missing in a specific case. All it can do is repeat what someone stored as a text block. The moment a question strays from that, and real customer questions almost always stray, it's over. For the company, that means the bot saves time in theory and generates calls from annoyed customers in practice, customers who now had to ask twice.

What sets a support agent with company knowledge apart

A support agent with company knowledge answers from the sources you approve. That could be the website, a folder of service descriptions and prices, the project tool or the calendar. Nothing beyond that. It answers the question on the merits, because it can look up the specific case.

A chatbot waits for your input. An AI employee takes over the task.

On top of that, honesty is a rule. If the agent can't find the answer in the approved sources, it says so plainly and brings in a person, rather than guessing. It makes nothing up. That one trait separates useful AI customer support from the embarrassing kind.

The handoff happens in the same chat. The customer stays in the conversation, your team sees the history so far and simply picks up the thread. From the customer's side, only the sender changes, and the thread never breaks.

For sensitive topics, there are approval gates. In defined categories, say anything touching deadlines or commitments, the agent drafts a suggested reply. The message only goes out once someone on your team has approved it. That way you keep control over exactly the answers where control matters.

When people look for an alternative to a chatbot, this is usually the package they mean: answers with substance and a reliable path to a person. Our page on WhatsApp customer support with company knowledge shows how that works day to day, with real chat threads and the process behind them.

Which customer inquiries you can automate, and what stays with the team

You can automate customer inquiries whose answer sits in your systems or documents. Anything that needs judgment stays with people. In practice, four classes of question make up most of the volume:

Status questions: "Where do things stand for us?", "How far along are you?". The answer is in the project tool or the CRM, and the agent reads it off there.
Document questions: "Which documents are still missing?". The agent checks what's on file and names exactly what's missing. For example: "Two bank statements are still missing for the annual accounts, November and December."
Appointment questions: "When is the technician coming?", confirmations, reschedules.
The usual: services, processes, opening hours. Everything that's on your website and still gets asked fresh every week.

On its own, each of these questions looks harmless. Do the math: at 25 such inquiries a week and about six minutes per answer (check the system, write the reply), that's 150 minutes a week, and a good ten hours a month. A full working day that consists of nothing but looking things up and passing them on.

Everything with substance stays with the team. In a tax or accounting firm, that means the agent takes status and document questions, with a clear line at anything that counts as advice. Casework carries on without an interruption every few minutes.

In an agency, it answers "How far along are you?" from the project tools and escalates anything substantive to the responsible account manager. Customers feel informed, the team stays in the work. And in a service business, it also takes inquiries overnight, answers the usual questions and pre-qualifies the rest: Monday morning starts with a sorted list and an empty voicemail box.

Why WhatsApp as the channel for customer support

WhatsApp is the channel your customers already write on anyway, personally and for business. A support line there meets them where they already are: no new portal and no hold queue. The bar for asking a question is as low as texting someone you know.

The second argument is the time of day. Customer questions come when customers have time: after work and on weekends. In the classic setup, a message at 6:42pm ("Where do things stand for us?") waits until the next business day for a reply. Some customers wait it out. Some don't wait that long, especially with a new inquiry where they're comparing two providers at the same time.

Technically, WhatsApp Business automation runs on the WhatsApp Business Platform, Meta's official interface for companies. Through it, a system is allowed to receive and answer messages, and your team can step into the same chat at any point. Every conversation stays visible to you. That's what turns WhatsApp from a private messenger into a service channel you can actually run.

Yes, if the setup is right. The foundation is the WhatsApp Business Platform, a data processing agreement, clear data boundaries and EU data storage.

The private WhatsApp app on a company phone doesn't meet these requirements, if only because it syncs the entire address book with Meta. The Business Platform works differently. It sends messages over a defined interface, accesses no address book, and can be brought in cleanly by contract.

Cleanly by contract means: whoever runs the agent for you processes customer data on your behalf. Art. 28 GDPR requires a data processing agreement for that. It sets out what may happen to the data and when it gets deleted, and it binds the operator to your instructions.

Then there are two decisions that sit with you. First, the data boundaries: the agent gets access to the sources it needs for customer questions, and nothing else. Personnel files or internal costings stay out. Second, data storage: the knowledge base and conversation logs sit on servers in the EU. Both can be pinned down before you start and checked later.

How the rollout works

The rollout follows a fixed order: onboarding, then operation, with approvals and data boundaries in place from the start. There's no need for a big project with a formal spec.

Onboarding

You name the classes of question the agent should take on, and the sources it's allowed to answer from. We connect those sources and set the escalation rules together: which topics always go to people, and who handles handoffs. By the end of onboarding, the agent knows what it may answer, and just as clearly what it has to pass on.

Operation

The agent runs as a Managed AI Employee: we set it up and run it, hosting included, for €250 per agent per month. Your team keeps working in WhatsApp as before, sees every conversation and takes over where needed. When services or processes change, that feeds into the knowledge base.

Approvals

At the start, the agent runs on a tight leash. Answers in sensitive categories only go out once someone on your team has approved them. With each approval, the picture grows of which answer classes it handles reliably, and those are the ones you unlock, step by step. You hand over control at the pace that fits your risk.

GDPR and data boundaries

Before you start, the data processing agreement, the list of approved sources, EU data storage and the deletion periods are all set. With us that's a fixed part of onboarding, so the data protection question is answered before the first customer message arrives.

What the agent cannot do

It doesn't replace a team. And it's only ever as good as the knowledge base it answers from.

Where it comes to advice, or a complaint, you need a person with judgment and accountability. The agent is built to recognize those cases and hand them off. Anyone bringing it in to cut support headcount is planning against what will actually happen. The gain is in the time your team gets back for the real work, and in customers who get a proper answer even at 7pm.

The second limit is upkeep. If the agent answers from an outdated service overview, it passes on outdated information, flawless in wording and wrong in substance. So the knowledge base needs a fixed upkeep rhythm. The effort for that is small when it's planned in from the start. The damage to trust is large when it's missing.

The third limit is the escalation rules. An agent that hands off cleanly only helps if someone on the other side takes over. Before you start, set which topics go to people right away and how fast your team responds to handoffs. Otherwise you're swapping the hold queue on the phone for a hold queue in the chat.

And when is the whole thing not worth it? If hardly any recurring questions come in for you, because practically every inquiry is a one-off, there's little to automate. Whether your inquiries fit the pattern shows in a look at two typical weeks: if the same questions keep coming up, the conversation is worth having. Bring exactly those questions and talk the task through with us.