Your team answers the same WhatsApp questions all day and evening messages get lost. What an AI agent can handle, and what should stay with a person.
Is your team answering the same WhatsApp questions all day?
It is Tuesday, 10 a.m. Your salesperson is putting together a quote and the phone keeps buzzing: "Is it in stock?", "How much is shipping to Medellín?", "What time do you close?", "Can you send me the location?". The quote gets left halfway, the same message gets pasted for the fifth time, and the work starts over.
At 7 p.m. another ten messages arrive. Nobody sees them until the next morning, and by then two of those customers have bought somewhere else.
This is not unusual. According to a HubSpot report on Colombia, 71% of sales teams use WhatsApp to prospect and reach customers, and 36% of service teams count it among their main support channels. Your customers are already there. What is missing is an orderly way to serve them.
Why does the company WhatsApp become a bottleneck?
It is rarely a lack of effort. It comes down to how the channel was set up:
- The number lives on one person's phone. When that person is at lunch, on vacation or quits, service stops.
- The answers are not written down. Everyone replies their own way, sometimes with prices or terms that have already changed.
- Everything arrives mixed together. The opening-hours question, the 5-million-peso order and the angry complaint land in the same chat with the same priority.
- Nothing is measured. Nobody knows how many messages come in per day, how long the first reply takes or how many go unanswered.
An AI agent helps with the first and third points, but only if you fix the second one first. If your answers are not written down, the AI will improvise them.
What can an AI WhatsApp agent answer?
A well-configured agent works for conversations that follow clear rules and rely on information you already have. These are the cases where it tends to pay off:
- FAQs: hours, location, payment methods, delivery times, return policies. It answers from a document you approve, not from what the model "thinks".
- Stock and catalog prices: as long as it checks a real source such as your inventory, a spreadsheet or your sales system.
- Lead qualification: it asks what the customer needs, by when, in which city and with what budget, then hands your salesperson a summary ready to close.
- Booking appointments or visits: it offers open slots from your calendar and confirms.
- Order status: it looks up the order or tracking number and replies with the data from your system.
- After hours: it covers the basics at night and on weekends and leaves the rest organized for the morning.
What should you not automate in WhatsApp customer service?
This is the part many vendors skip. In some conversations a fast but wrong answer costs more than a slow one. These should go to a person from the first message:
- Complaints. An upset customer needs someone who listens and has the authority to fix things. An automated reply, however polite, usually makes it worse.
- Money: refunds, discounts, price changes, payment arrangements or exceptions to your policy.
- Health, legal or financial questions where the answer depends on the specific case. The agent can take the request; someone qualified gives the answer.
- Negotiations and large sales. The agent qualifies and organizes; your team closes.
- Anything the AI cannot verify against your data.
This is not only good practice. The WhatsApp Business Messaging Policy requires that, if you automate service, customers have a quick, clear way to reach a person: in the same chat, by phone, by email or on your website.
Also, since January 15, 2026, Meta does not allow general-purpose AI chatbots on the WhatsApp Business Platform (the API), meaning assistants that chat about any topic. Agents focused on business tasks, such as customer service, orders or appointments, are still allowed. For a small business the takeaway is simple: your agent talks about your business and nothing else.
How can you start automating WhatsApp this week?
You do not need a full agent to start. You can take these steps over the next five days:
- Day 1: count the questions. Go through the last two weeks of chats and write down the ten questions that repeat the most.
- Day 2: write the approved answers. A short document with each answer, the current price or condition, and who approved it. This becomes the agent's knowledge base.
- Day 3: define when a person takes over. List the topics that are never automated (complaints, money, exceptions) and the hours when a real person replies.
- Day 4: measure the baseline. How many messages arrive per day, how long the first reply takes and how many are still unanswered after 6 p.m.
- Day 5: pick the pilot. One type of request, for example FAQs after hours.
Then you can choose the tool. If volume is low, the WhatsApp Business app with quick replies and away messages may be enough for now. If volume is high or you want an AI agent, you need the WhatsApp Business Platform (API), usually through a provider, connected to a model such as Claude, ChatGPT or Gemini, or to the agent Meta itself offers. The choice depends on your systems, your budget and which data it will touch.
During the pilot, someone on the team should read the conversations every day. It is the only way to see where the agent gets confused and to fix the answers before expanding it.
How much does Meta charge for WhatsApp Business messages?
If you go with the API, you need to understand how Meta charges, because it just changed. As of September 2026, Meta charges per message. Template messages (marketing, utility and authentication) are priced by country, and Colombia has its own rate.
The key change: starting October 1, 2026, replies inside the 24-hour customer service window, which had been free since November 2024, are charged per message. Meta includes 1,000 free service messages per month per phone number, and businesses without a payment method on file can no longer send those messages.
On top of that, add the platform provider's fee and the AI model usage. Before you commit, ask the provider for an estimate based on your real volume, the one you measured on day 4, and check the current rate in Meta's documentation.
One more detail: Meta requires the customer's opt-in before you message them on your own initiative. Someone asking you for a price is not permission to send them promotions.
What about the personal data that comes in through WhatsApp?
Every chat carries personal data: name, number, address and sometimes ID numbers or health information. Colombia's Law 1581 of 2012 applies just as it does on any other channel, and the SIC's External Circular 002 of 2024 asks that personal data in AI systems be handled in a way that is suitable, necessary, reasonable and proportionate, with privacy built in from the design stage.
On WhatsApp, that looks like this:
- The agent asks only for the data it needs to resolve the request.
- It never asks for card numbers or passwords in the chat.
- The first message links to your privacy notice.
- You know where conversations are stored, for how long, and what the AI provider does with them.
This is general information, not legal advice. If you handle sensitive data such as medical records or financial information, review it with your lawyer before connecting any agent.
Is it worth it for your company?
If your team loses hours every week answering the same things and evening messages go unanswered, probably yes. If you get twenty messages a day and each one is different, what you need first may be organized answers, not an agent.
Either way, the starting point is the same: know what customers ask, write down what you answer and decide what a machine should never answer.