Skip to content
ZEISMO

WhatsApp bots and AI

A bot that answers, and knows when to stop.

Most bots are a phone tree in a chat window. Ours answers the question first, hands over the moment somebody needs a person, and never invents a price. Try the real routing below before you believe any of it.

Try the actual router

This is the routing logic from the bot that ran our own commerce operation, ported to your browser unchanged. Type an insult, or ask for a human in Arabic, and watch what it decides. Every branch below exists because it once got something wrong in production.

Or try one of these

The router decided

Asking for a person

Why

Matched a request to speak to a human being.

What happens next

Handed over at once, even mid-order. A form does not get to stand between somebody and a person.

This is the routing logic from the bot running our own shop, not a simplified demo. This runs entirely in your browser. Nothing you type is sent to us or stored anywhere.

What you get

Answers in seconds, at 3am

The median first reply on our own line is one second. A customer who waits an hour has usually bought elsewhere.

Knows when to stop

Anger, a request for a person, a price question, anything that commits you: handed to a human at once, mid-order if necessary.

Never invents anything

It answers from your real answers first and only calls a model when none fit. It does not guess a price or promise a date.

Speaks how your customers speak

Including dialect. Our production bot works in Egyptian Arabic daily; tone is part of training, not an accident.

Has a spending limit

A written daily cap. A runaway conversation costs one refusal, not a bill you find out about later.

You can switch it off

A kill switch in your own portal. A bot you cannot stop talking to your customers is not one you control.

When not to buy this

We would rather lose the sale than build something that does not work for you.

A bot will not fix a bad offer

If people are not buying because the price or the product is wrong, faster replies just get you to the no sooner.

Some work should stay human

Complaints, negotiation and judgement calls. Our free calculator will tell you to keep the person, and it is right.

It needs your real answers

The useful part of the build is extracting what you actually tell customers. Nobody can skip that step for you.

Work out whether a role should be automated at all with the AI versus human cost calculator. It is free, needs no email, and will tell you to keep the person when that is the right answer.

Talk to us on WhatsApp

Tell us what is eating your team

Three fields. We message you on WhatsApp, ask one or two things, and tell you honestly whether a bot helps. If it does not, we say so.

Any country. Include the code, for example +971 or +20.

We use this to message you and nothing else. No list, no resale.

How a build actually goes

01

We read your real answers

The useful part is extracting what you already tell customers. Nobody can skip this for you, and it is where most of the value is.

02

A narrow first version

One channel, the questions you answer most, and a written rule for when it hands to a person. Live in days, not a quarter.

03

Measured against silence

We watch what it gets wrong and what it escalates. A bot nobody measures drifts, and you find out from a customer.

04

Handed over

The accounts, the answers and the config are yours. You can stop it from your own portal without asking us.

Questions people actually ask

Will it sound like a robot?

Not if it is built from your own answers. The tell is not the technology, it is a bot answering a question nobody asked because it was written from a script instead of a real conversation.

What happens when it does not know?

It says so and offers a person. A bot that guesses is worse than no bot, because a wrong answer travels further than silence.

Can it work in Arabic?

Yes, including dialect. Our own production line runs in Egyptian Arabic daily, and the tone is part of the training rather than an afterthought.

How fast can it be live?

Days for a narrow first version, if you can get us your real answers. The extraction is the slow part, not the build.

What does it cost to run?

Model usage plus hosting and monitoring. There is a written daily cap, so a runaway conversation costs one refusal instead of a bill you discover later.

What if we want to stop it?

A switch in your own portal. Immediate, no approval from us. A bot you cannot stop talking to your customers is not one you control.

Tell us what your customers keep asking.

That is the whole first conversation. If a bot is the wrong answer we will say so.