An answer in two minutes, at three in the morning
Most customers don't walk away over price — they walk away over silence. A good bot answers in plain language, knows what it doesn't know, collects what's needed and hands over to a person at exactly the right moment, not a minute later.
A bot is not a menu of buttons
The previous generation of bots was a decision tree: press 1 for support, 2 for sales. Everyone learned to bypass it. A bot built on a language model answers what was actually written — including a half-formed question, spelling mistakes, and two questions in one sentence.
The secret isn't the model but what you feed it: your prices, your procedures, your hours, exactly what you do and what you don't. A bot that speaks beautifully and knows nothing does more harm than good, which is why half the work is building the knowledge base and the limits.
The other half is the handover. A good bot recognises an angry customer, an unusual request or a large deal — and passes it on with everything gathered so far, so the person picking it up doesn't start from zero.
What the bot can do
All of it built around your channel and the kind of enquiries you actually get.
Free-language answers
The customer writes as they speak, and the bot answers from your knowledge — not from a fixed script.
WhatsApp, web and more
The same bot with the same knowledge, wherever your customers already are.
Filtering and qualification
Understands who it's dealing with, collects the details that matter, and flags a worthwhile enquiry before it reaches your diary.
Booking
Offers open slots, books, sends a reminder and updates whoever needs to know.
Handover to a person
Clear handover rules and a summary of the conversation so far, so nobody asks for the same detail twice.
Measurement and improvement
You see what's asked, where the bot gets stuck and what keeps coming back — and that feeds the next version.
How it gets built
From real enquiries to a bot facing customers, usually in weeks rather than months.
- 01
Read real enquiries
We take hundreds of recent conversations and look at what's genuinely asked. That sets the knowledge base — not our assumptions.
- 02
Build knowledge and limits
Prices, procedures, hours, what may be promised and what may not, and when the bot must say "I don't know".
- 03
Run it on the side
The bot answers to the team before it answers a customer, and we compare. This is where every question nobody thought of shows up.
- 04
Go live and improve
Live with customers, reviewed weekly at first, and corrected on what actually happens in the field.
Rather than explain how a bot works, here's one
This bot knows me, what I build, the projects and how I work. Write to it in your own words, in English or Hebrew, and try to catch it out with something the site doesn't cover — it should tell you it doesn't know.
- Answers only from defined sources, and refuses to guess when it has nothing.
- Remembers the conversation, so a follow-up doesn't start from zero.
- Detects the language you wrote in and answers in it.
- Points you to WhatsApp the moment a question needs a person.
Shai's botOnline · answers in secondsHi! I'm Shai's bot. Ask me anything about him, what he builds, his projects or how a project starts — in your own words.
The bot answers only from what's on this site. For anything specific to your project, it's worth talking to Shai himself.
Why it works
A bot is the project that pays for itself fastest — provided it's built properly.
- A knowledge base built from your real enquiries, so the bot answers what's actually asked.
- Defined limits: it never invents a price, never promises what wasn't agreed, and can say it doesn't know.
- Handover with full context, so the bot shortens the conversation instead of adding a step to it.
- Measured from day one — how many were answered, how many went to a person, and where the conversation broke.
Questions that come up
Don't customers hate bots?
They hate a bot that pretends to be human and walks them in circles. They don't mind one that answers instantly, gets it right and hands over the moment it should. I always recommend saying it's a bot up front — it lowers false expectations and raises patience.
Can the bot make things up?
That's the real risk, which is why most of the work is about limits: it answers from defined sources, prices and promises are marked off-limits, and when there's no answer it says so and hands over.
How long does it take to launch?
A focused WhatsApp or website bot goes live within a few weeks. What stretches it is usually gathering the knowledge from inside the organisation, not the build.
Can it connect to our systems?
Yes — calendar, CRM, inventory, forms. That's what lets it book a meeting or check a status rather than only answering general questions.
What about customer data?
We define up front what's stored, for how long and who reaches it. Sensitive details can be kept away from the model entirely, in which case the bot collects them and passes them straight to whoever needs them.
Is there one I can try?
There is — on this page. The bot above is built on exactly the principle described here: a defined knowledge base, free-language answers, and a refusal to guess when it has nothing.
