WhatsApp Business API for customer support is a different job from having WhatsApp at all. Most businesses that move to the API already answer customers on the app — the change is not the channel, it is what happens to a message once it arrives. On the free app, messages land on one phone and get answered when somebody picks it up. On the API, a message can be sorted before a human sees it, answered from your own documents, routed to the right person, handed over with its full history attached, and followed up after the conversation ends. That is the whole point of the API for support: not a bigger inbox, but a support operation that keeps working when the team is small and the queue is not. This guide covers the support side only — the use cases, the triage, the routing and handoff, the 24-hour window as a support constraint, the templates support actually uses, and how to measure whether any of it is helping. The API itself — what it is, how access works, and how to get set up — is covered in the
WhatsApp Business API guide, and it is not repeated here.
Which support jobs the API is actually for
Support on WhatsApp is not one kind of message. It is a handful of recurring jobs, and the API earns its keep on the ones that repeat. Sorting your queue by job, rather than by arrival time, is the first decision a support team makes — because the answer, the urgency, and the person who should handle it all depend on which job the message is.
- Where is my order? The most repeated support question for any business that ships. The answer lives in your order system, not in a person's head — so the API can fetch it, or the agent answering from your documents can, before a human is involved.
- Something arrived damaged, wrong, or late. The customer is upset and the answer is a process: an apology, a photo request, a replacement or a refund. The value of the API here is the handoff — the whole conversation, photos included, reaching the person who can authorise the fix.
- Returns and exchanges. A rules question with a fixed answer: your returns window, who pays postage, how a refund is issued. This is the job an agent answers best, because the policy document already contains the whole reply.
- Appointment changes. Bookings, reschedules, cancellations, reminders. The answer is a slot in a calendar, and the follow-up is a template reminder — the two halves of this job are exactly what the API's template messages are for.
- Account and billing help. Password resets, invoice copies, subscription changes, payment failures. These need identity checks before answers, so triage matters: the first reply should ask for the right detail, not guess at the account.
- Pre-sales questions that turn into support. A customer asking whether a product fits their need becomes a support conversation the moment they buy. The API keeps both halves in one thread instead of splitting sales and support across two tools.
Notice what is missing from that list: one-off complaints that need a manager's judgement, legal or safety issues, and anything that needs a phone call. Those still belong to a person, and the API's job for them is routing — getting the message to that person fast, with context, instead of letting it sit in a general queue.
Triage: sorting the queue before a human touches it
Triage is the step most small teams skip, and the one that saves the most time. Without it, the queue is first-come-first-served: a refund request waits behind three order-status questions, and the urgent message is answered last because it arrived last. With triage, every message is sorted the moment it arrives, on two axes — what it is about, and how urgent it is.
The first axis is intent. A message that says "where is my parcel" is an order-status job; "the box arrived broken" is a damage job; "can I move my appointment" is a booking job. An AI agent reading your own documents can make that call from the customer's own words, because the documents already describe the jobs — the returns policy, the delivery information, the booking rules. The customer does not have to pick from a menu, and nobody on the team has to read fifty messages to find the five that matter.
The second axis is urgency, and it is judged by the customer's situation, not by the words they use. A polite message about a delivery that was due yesterday is more urgent than an angry message about a delivery due next week. A payment failure is urgent because the customer's order is stuck; a question about opening hours is not, however it is phrased. Triage rules should be written around situations — delivery overdue, payment failed, appointment today — not around tone, because tone is the least reliable signal in the queue.
Triage also decides what never reaches a human at all. The order-status question, the opening-hours question, the returns-window question — these have fixed answers in your documents, and an agent can answer them one-to-one, around the clock, including the half of them that arrive after closing time. That is the real triage win: not sorting faster, but removing the messages that never needed a person. The human queue that remains is smaller, and every message in it is one a person actually adds value to.
Routing and handoff: getting the message to the right person
Routing is what happens after triage: the message goes to the person who can resolve it, not to whoever happens to be free. For a team of two or three, routing can be simple — orders to one person, bookings to another, everything damaged or refund-shaped to the owner. For a larger team, it follows skills: the person who handles billing does not need to see the delivery questions. The rule is the same either way: route by the job, and let the urgent jobs jump the queue.
The handoff is the moment routing usually fails. A customer explains their problem to an agent, the agent cannot resolve it, and the human who takes over asks the customer to explain it all again. That is the experience that makes customers hate automated support — not the automation, but the amnesia. A good handoff carries three things with the conversation: the full thread, what the agent already tried, and what the customer already told it. On Neksio, that is how the agent hands over: the human sees the whole conversation in one unified inbox and picks it up mid-thread, so the customer never repeats themselves. The full inbox shape is covered in the
Handoff needs triggers, not vibes. Decide in advance what sends a conversation to a human: the customer asks for one, the agent cannot find the answer in your documents, the message mentions a refund, a chargeback, or a complaint, or the same customer has written three times about the same order. Write the triggers down and check them monthly — the most common trigger tells you what to add to your documents next, which is how the agent gets better at the routine jobs without anyone rebuilding it.
After-hours handoff deserves its own rule. Half of support messages arrive when nobody is at the desk, and the wrong answer is to leave them until morning. The agent answers the routine jobs overnight; the jobs that need a human get a clear reply — what will happen, and when — so the customer is not left wondering whether anyone saw the message. A customer who knows their refund request is queued for 9am is a patient customer; a customer who hears nothing until 9am is not.
The 24-hour window as a support constraint
The 24-hour customer service window is the rule that shapes every support workflow on the API: once a customer messages you, you have 24 hours to reply freely; after that, you can only reach them with an approved template message. For a support team, that is not a technical detail — it is a deadline on every open conversation, and the queue has to be worked with it in mind.
In practice the window changes three things. First, it sets the real response target: not "when we get to it", but inside the window, because a conversation that slips past 24 hours becomes harder and slower to reopen. Second, it makes the follow-up a template job — a delivery update, an appointment reminder, or a "your replacement is on its way" message goes out as an approved template, which is why templates are a support tool and not a marketing one. Third, it rewards answering the routine jobs immediately: an agent that resolves the order-status question at 11pm keeps that conversation inside the window without anyone staying late. The rule itself, and how the window opens and closes, is explained in the
24-hour window guide — what matters here is that your support process treats the window as a deadline, not a surprise.
Templates for support, not for marketing
Templates are the messages you send outside the 24-hour window, and the ones support sends look nothing like the ones marketing sends. A support template is a utility: it carries information the customer asked for, or a next step in a conversation they started. The categories to build first are the ones your support jobs already produce.
- Order and delivery updates. "Your order has shipped" and "your replacement is on its way" — the message a customer is waiting for, sent the moment the status changes, not when somebody remembers.
- Appointment reminders. The day-before reminder and the one-hour reminder, with the time, the address, and how to reschedule. This one template prevents more support messages than any other, because it stops the "when is my appointment" question from being asked.
- Resolution follow-ups. "Your refund has been issued" or "your replacement has been dispatched" — the closing message that ends the conversation cleanly instead of leaving the customer to ask whether anything happened.
- Information requests. "Please send a photo of the damage" or "please confirm the delivery address" — the message that moves a stalled conversation forward when the customer has gone quiet.
What support templates are not: promotions, announcements, or anything sent to a list. Neksio is not a bulk or broadcast tool — it answers customers one-to-one, and its templates go out as 1:1 replies in conversations the customer started. A template sent to people who never asked is a marketing blast wearing a support costume, and WhatsApp's own rules treat it that way. The template categories, approval process, and examples are covered in the
Measuring whether support is actually getting better
A support operation without measurements is a queue with opinions. Five numbers tell you whether the API setup is working, and none of them needs a dashboard — a weekly count from the inbox is enough to start.
- First response time. How long from the customer's message to the first reply — from the agent or a human. This is the number customers feel most, and the one an agent moves most, because the routine jobs get a first reply in seconds, at any hour.
- Share resolved without a handoff. Of the messages the agent answers, how many end without a human stepping in. If the share is low, the fix is usually in your documents — the answer the agent needs is missing, not the agent.
- Repeat contacts on the same issue. A customer who writes three times about one order is a process failure, not a difficult customer. Count repeats per order or per appointment, and the worst offenders will point at the broken step.
- Conversations that slip past the 24-hour window. Every one is a support conversation that got harder to close. If the count is climbing, the queue is too big for the team, or triage is letting urgent jobs wait behind routine ones.
- Follow-up replies after a resolution message. Send a short "is everything sorted?" after resolving an issue and count the answers. It is the cheapest satisfaction check there is, and it catches the resolutions that did not actually resolve anything.
Review the five numbers together, weekly, for a month before changing anything big. A faster first response with more repeat contacts means the agent is answering quickly and wrongly — the documents need work, not the speed. A high resolution share with a climbing window-slip count means the agent is fine and the human queue is the bottleneck — the handoff triggers need tightening, not the agent. The numbers only help when they are read as a set.
A support day on the API: an illustrative example
Take an illustrative example: a small online shop selling home goods, two people answering support, about sixty WhatsApp messages a day. Before the API, the messages landed on one phone, the owner answered between packing orders, and the queue was worked in arrival order — the refund request waited behind the opening-hours questions, and the messages that arrived after 6pm waited until morning.
On the API, the day runs differently. Overnight, the agent answers the order-status and opening-hours questions from the shop's documents, so the morning queue is already smaller. At 9am, triage has sorted what remains: three delivery problems go to the owner, two appointment-style queries — delivery-slot changes — go to the second person, and one damaged-item report is routed straight to the owner with the customer's photos attached. The damaged item is the only conversation that needs a decision; the rest are process. At midday, the appointment reminders go out as templates for tomorrow's deliveries. At 6pm, the agent takes over again, and the two conversations still open — a refund awaiting the owner's call — get a template reply saying the refund will be confirmed by 10am. The next morning's queue starts with one job, not sixty. The numbers in this example are illustrative; run your own message volume through the free tier before deciding anything.
What running support on the API costs on Neksio
On Neksio, the plans follow the same shape as for any small business, and the unit is replies answered, not seats filled: Free covers 100 AI replies a month on WhatsApp, Starter at $12.99 a month covers 5,000 replies, Growth at $16.99 a month covers 15,000 replies and adds Facebook and Instagram, and Scale at $29.99 a month is unlimited replies with your own key, plus done-for-you onboarding. A two-person support team does not pay for two seats; it pays for the questions its agent answers. Start on Free — 100 replies a month covers the overnight order-status questions while you check whether the agent's answers are right — and move to Starter when the volume grows past that, or to Growth if you also answer customers on Facebook and Instagram. Full detail is on the
pricing page. For the full setup — connecting WhatsApp, uploading your documents, and letting the agent answer from them — the WhatsApp AI agent guide walks through it step by step.
Setting up support on the API, in order
The order matters more than the tools. First, list your five most repeated support jobs and write the answer for each from your own documents — that list is your triage map and your agent's knowledge in one. Second, decide the handoff triggers: what sends a conversation to a human, and who that human is for each job. Third, build the four support templates — delivery updates, appointment reminders, resolution follow-ups, information requests — and get them approved before you need them. Fourth, set the after-hours rule: what the agent answers overnight, and what gets a "we will confirm by 10am" reply. Fifth, run it on real customers for two weeks and count the five numbers above before changing anything. The businesses that struggle with support on the API are rarely the ones with the wrong tool; they are the ones that connected the API and kept working the queue the old way.
100 AI replies a month on WhatsApp, no credit card. Connect WhatsApp, upload your returns policy and FAQs, and let the agent answer the routine jobs from your own words. Full detail on the pricing page.
Frequently asked questions
- Can customer support run entirely on the WhatsApp Business API?
- The routine jobs can — order status, opening hours, returns windows, appointment reminders — because the answers live in your own documents and an agent can give them one-to-one, around the clock. The jobs that need judgement still belong to a person: refunds, damaged goods, complaints, and anything the agent cannot find in your documents. The API's value is the split: the agent clears the routine queue so the humans work the conversations only a human can resolve.
- What happens when a customer messages after the 24-hour window has closed?
- You cannot reply freely — you can only reach them with an approved template message. That is why support teams treat the window as a deadline: answer inside it, and use templates for the follow-ups that fall outside it, like delivery updates and resolution confirmations. The full rule is explained in the 24-hour window guide.
- Do I need a team inbox to run support on the WhatsApp Business API?
- You need somewhere the whole team can see the same conversations, with the full thread attached when a chat is handed over — which is what a shared inbox is. On a team of one, the API alone is enough. The moment a second person answers, or an agent hands off to a human, the shared inbox is what stops customers repeating themselves to each new person.
- How are support templates different from marketing templates?
- Support templates carry information the customer asked for, or a next step in a conversation they started — a delivery update, an appointment reminder, a refund confirmation. Marketing templates promote something to people who did not ask. Neksio is not a bulk or broadcast tool: its templates go out as 1:1 replies in conversations the customer started, never as blasts to a list.
- Can the AI agent handle refunds on WhatsApp?
- It can handle the routine half: explain the returns policy, ask for the order details and photos, and collect everything the person authorising the refund needs. The refund itself — the decision and the payment — stays with a human, and the handoff carries the full conversation with it, so the customer never repeats the story. Set that trigger in advance: any message mentioning a refund goes to a human with its history attached.
- What should I measure first when I move support to the WhatsApp Business API?
- First response time and the share of messages resolved without a human handoff. The first tells you whether customers are waiting less; the second tells you whether the agent is actually clearing work or just replying fast. Add repeat contacts on the same issue and conversations slipping past the 24-hour window in the second week — together, the four numbers show whether support is getting better or just getting faster.
