Endelir
Start free
Use case

AI customer support on WhatsApp that hands over to your team

Endelir's AI teammate answers routine WhatsApp questions first, in seconds, from your own documents and policies; keyword rules and chat flows handle the fixed answers, and anything it cannot settle goes to a named person. Opening hours, where an order has got to, what the return policy says — a large share of what customers write is the same short list. Endelir answers that list without involving anyone, and passes the rest to a named person with the details already collected. Switch the AI assistant on and it answers first, in seconds; the plain limits are further down this page.

Start free for 14 daysTalk to us
A young woman sitting up in bed late at night, taking over a conversation on her phone
Live
Message to inbox in under a second.

The same twenty questions, arriving all day

A support line is mostly repetition. Check-in time, opening hours, where the parcel is, whether something comes in another size — the same short list, asked by different people, spread across the whole day. Each one takes a minute to answer and none of them can be left.

The cost is not the answering, it is the interruption. The conversations that need real attention — a complaint, a refund, a customer who is already annoyed — sit behind twenty easy ones. And the enquiry that arrives at eleven at night waits until morning, by which time the customer has asked someone else.

Routine questions answered, the rest sent to a person

Endelir splits the load. Keyword rules answer the questions whose answer never changes. Chat flows handle anything that takes a step or two — asking for an order number, recording a booking reference, putting the conversation in front of the right team with the answers attached. Both of those are live today, and both hand over to a person cleanly rather than looping.

The AI assistant sits behind them as the last resort, for a question worded in a way no rule anticipated. It is a switch in your workspace: turn it on and it answers first, in seconds, before anyone on your team has to; turn it off and the rules and flows carry on as before. By default it runs on Endelir's own AI connection, so there is no key to set up; you can connect your own AI provider instead.

How it works

  1. 1 A customer writes in The message reaches your business number and opens in the shared inbox, with whatever history that contact already has.
  2. 2 Your own rules are tried first A keyword rule can answer outright, and a chat flow can start on its own trigger. First match wins — two replies to one message reads as a broken bot.
  3. 3 A flow collects what is missing Question steps ask for an order number, a booking reference or what went wrong, and write each answer onto the contact record.
  4. 4 The AI takes what is left Once it is switched on, the assistant is offered only the messages nothing else matched. It answers from material you supply, not from general knowledge about your trade.
  5. 5 It stops rather than guesses If the assistant produces nothing usable, the customer gets the line you configured for that case, or nothing at all. The underlying error is never shown.
  6. 6 A person takes over The conversation is assigned to someone, and taking it over pauses automation on that chat so the customer is not answered twice by two different things.

What this looks like in practice

A hotel gets the same three questions all week: what time is check-in, can an airport transfer be arranged, and can we have extra towels. Two of them have a fixed answer. The third is a request that has to reach housekeeping while there is still someone on the floor to act on it.

Keyword rules answer the first two the moment they are asked, at any hour. The transfer request runs a short flow that asks for the arrival date, the flight number and how many people are travelling, saves all three against the contact, and assigns the conversation to the front desk with the answers already visible. With the AI assistant switched on, it takes the fourth question — the one nobody wrote a rule for — and hands over anything it cannot settle.

What can happen next

An answer goes straight back A rule replies inside the customer-service window, so no template is involved and nothing waits for a person.
A question is asked first A flow collects the order number or booking reference before anyone is involved, so the conversation arrives complete.
The answers are saved What the customer gives is written onto the contact record, where the next person to open the chat can read it.
It goes to a team A flow step assigns the conversation, so a billing question reaches billing rather than whoever happens to be free.
A person takes over Taking over pauses automation on that conversation. There is also a switch per contact, for a customer who should only ever hear from a person.
It reaches your own systems A flow step can call a URL you control, so a support case can open in your own tools while the conversation is still going.

What to know before you set it up

The AI assistant is a switch you turn on in your workspace, and once it is on it answers first. By default it runs on Endelir's own AI connection, so there is no key to find. If you already pay an AI provider, you can connect your own key or your own endpoint instead, and that usage is billed to you by the provider rather than by us.

The assistant answers from material you give it. You give it your own documents, price list and policies; a question is matched against that material and the closest sections are used to write the reply. It can still be wrong, so the material wants to be short, current, and written the way you would answer at the counter. Anything you never wrote down it cannot answer, and it should hand over instead.

Replying to someone who wrote to you first needs no template, inside the customer-service window Meta allows. Starting a conversation, or replying after that window has closed, uses a message template approved in advance, and automation does not change that. Nor does it remove the need for a way through to a person: a customer who asks for staff should get staff. What each plan includes is on the pricing page.

What to watch

  • How many conversations finish without a person being needed
  • Which questions the automation hands over, and at what point
  • How long a handed-over conversation waits for its first human reply
  • How often a customer asks for a person outright
  • Which questions come up most, so the answers behind them can be corrected
  • How many customers come back a second time with the same question

Questions people ask

Can AI answer WhatsApp customer questions automatically?

Yes. The AI assistant is a switch in your workspace, and once it is on it answers first, in seconds, from the material you give it. Keyword rules and chat flows answer automatically too, and between them they cover most of the repetition — opening hours, booking policies, standard support information, the first step of a return. A rule answers in seconds and asks nothing of your team. When the assistant is switched on it sits behind those rules and takes the messages they did not match, which is exactly what rules are bad at: a familiar question worded in an unfamiliar way.

How does the AI know the correct answer?

From material you supply, and nothing else. You give it your own material: documents, policies, hours, price lists, the answers you already give. The text is split into sections and indexed; when a question arrives it is matched against those sections and the closest ones are used to write the reply. That is deliberately narrower than a general chatbot. The assistant is there to repeat what your business already says, not to produce something plausible. It can still be wrong, which is why the material wants to be short, current, and written the way you would answer at the counter. Anything you never wrote down, it cannot answer.

What happens when the AI does not know the answer?

It stops. If the assistant produces nothing usable, the customer gets the line you configured for that case — a plain sentence saying someone will follow up — or nothing at all, and the conversation sits in the inbox waiting for a person. The underlying error is never shown to the customer. This is not only good manners. Meta expects a business using automation in customer service to leave a clear route to a person, so an assistant that will not stop answering is a problem rather than a feature. The same goes for subjects you would rather it never touched: complaints, medical questions, anything about money. Keep those out of the material and let them reach staff.

Can a customer ask to speak to a human?

Yes, and it should be easy. Every conversation sits in the shared inbox whether automation touched it or not, so a customer asking for a person is asking somewhere your team already reads. A keyword rule can catch the request and assign the chat straight to a team. When someone takes a conversation over by hand, automation pauses on that chat, so the customer is not talked over mid-sentence. There is also a switch per contact for someone who should only ever hear from a person — a difficult account, a case already in dispute — and it stays off until you turn it back on.

Can AI look up an order or reservation?

Not on its own. The assistant answers from text, not from your database, and the packaged shop integrations that would hand it order data are not switched on. What works today runs the other way round: your own system calls the Endelir API with the status, or a chat flow calls a URL you control and your system answers. That is enough to run order-status support properly — the customer gives an order number in the chat, the flow passes it to you, and the reply that comes back has real data behind it. It is set-up work rather than a switch, and it needs someone who can connect your system to ours.

Can AI collect information before transferring the conversation?

Yes, and this is the part worth building first. A chat flow asks for what your team always ends up asking for — order number, booking reference, what actually went wrong — and writes each answer onto the contact record before anyone opens the chat. The person who picks it up starts with the facts instead of three messages of questioning. It also makes the handover land in the right place: the flow can assign the conversation to a named team rather than a queue. None of this waits on the AI switch. Question steps, custom fields and team assignment are live now.

Can AI support customers outside business hours?

Automation runs whenever a message arrives, so yes. There is also a timing restriction if you would rather it did not: you set a start time, an end time and a timezone, and automated replies stay quiet outside that window. Which you want depends on the promise you can keep. Answering a stock question at midnight is helpful; implying that someone is sitting at a desk at midnight is not. The plain approach is to answer what has a fixed answer at any hour, and to say in the reply itself when a person will pick the conversation up. Anything started overnight is still in the inbox in the morning, with its history intact.

Give customers answers without making every question wait for a person

Keyword rules, chat flows and the AI assistant do this on your own business number, and we set the number up with you on a call.

Start free for 14 days Talk to us

14 days free. No credit card. We call you to set the number up.