AI Agent and Chatbot Development for WhatsApp, Instagram, Web Chat and Voice
Clap Digital builds custom AI agents and chatbots that answer customers on WhatsApp, Instagram, website chat and the phone, and do real work in your systems: booking, qualifying leads, answering support questions and updating your CRM. We build them for businesses in the UAE, Turkey and worldwide, from Dubai and Istanbul, in English, Arabic and Turkish, and each agent is scoped around one job, connected to the tools you already use, and handed over with its rules written down.
What an agent does, by channel
One agent can answer on several channels from the same knowledge and the same rules, so a customer gets the same answer on WhatsApp as on your website. What changes by channel is how the conversation starts and what the platform allows:
- WhatsApp, through the WhatsApp Business Platform: the agent replies to customers who message your number. Messages a business sends first, such as reminders, must use templates Meta approves, and Meta charges for templates by category, so reminders are opt-in and planned into the running costs.
- Instagram DMs, on a professional account: the agent answers enquiries, prices and availability. Meta lets automated replies run only within a limited window after the customer last wrote, so follow-ups move to WhatsApp, SMS or email.
- Website chat: a widget on your own site, which you fully control, and the easiest place to start because no platform approval is needed.
- Voice: an agent that answers phone calls, handles the routine questions and bookings, and passes the rest to a person with a summary.
If most of your bookings come in on WhatsApp and Instagram, our WhatsApp and Instagram AI booking agent page covers that exact build for UAE clinics and salons in depth. For a plain-language view of what any agent build requires, read what it takes to build an AI agent for your business.
How we scope an agent
Scoping starts with one job, not with a model or a platform. On the discovery call we pick the job where an agent saves the most time or wins the most business, for example after-hours enquiries, first-line support or lead qualification, and agree what a good result looks like. Then we write down, with you:
- The channels it answers on, and the languages it is tested in.
- The questions it must answer, with the answers you want given, taken from your site, documents and past conversations.
- The actions it may take (book, quote a published price, create a lead, open a ticket) and the systems each action touches.
- The handoff rules: when it stops and passes the conversation to a named person.
- The topics it must never discuss or decide on.
This list is the scope, the quote and the test plan at the same time. Most of the accuracy of an agent comes from it, not from the AI model behind it.
Integrations
An agent that only talks is a FAQ page with extra steps. We are full-stack, so we build the connections that let it act: calendars and booking systems, CRMs, helpdesks and shared inboxes, order and inventory data, spreadsheets, and your own databases or APIs. When a system has no API, we use what it does offer, such as a prefilled booking link or a structured request your team approves, and say so plainly in the scope. If the agent needs a dashboard or portal of its own, that becomes a small web app inside the same project.
Data handling
- It collects only what the job needs, and never card numbers in chat.
- Customers are told they are talking to an automated assistant, with a link to your privacy notice.
- Conversation logs are kept where you approve, for as long as you set, with access limited to named staff.
- We name every AI and messaging provider that processes the messages, so the setup can be checked against the data protection law that applies to your licence before launch.
How the build runs
Every project follows the same order. First discovery and the written scope. Then the knowledge base and the integrations, starting any platform approvals (such as Meta business verification for WhatsApp) in the first week, because their timing is set by the platform, not by us. Then testing against real past conversations in each language, including the awkward ones. Then a supervised launch on one channel, with a person reviewing conversations daily before the next channel is switched on.
How long this takes depends mostly on how many systems the agent connects to and how fast platform approvals come back. You get the timeline in writing with the fixed quote.
Handoff: what you own at the end
- The agent running on your accounts (your WhatsApp number, your Meta business account, your hosting or ours, as agreed).
- The written knowledge base, actions and handoff rules, in plain language your team can edit.
- A short guide for the person who owns the inbox: how handoffs arrive, how to hand a thread back, and how to correct a wrong answer.
- A record of the test conversations it passed before launch.
What it costs
An agent is a one-time build plus running costs. The build price depends on the number of channels, languages and systems it connects to, and you get a fixed quote after a free discovery call. Running costs are separate: AI usage, hosting, and on WhatsApp the messaging fees Meta charges, which Meta changes from time to time, so we estimate them for your volume on the day we quote. Our guide to AI chatbot costs for small businesses compares what free widgets, off-the-shelf assistants and custom agents cost.
When not to buy an agent: if you get a handful of messages a day and someone already answers them within the hour, a WhatsApp click-to-chat link and a named inbox owner will do the job for far less. An agent pays off when messages wait, arrive after hours, or repeat the same questions every day.
To check whether an agent fits your business, book a free discovery call.
Frequently asked questions
How much does it cost to build an AI agent?
It depends on the channels, languages and systems the agent connects to, so we give a fixed quote after a free discovery call rather than a generic figure. Running costs are separate and depend on volume: AI usage, hosting, and the WhatsApp messaging fees Meta charges when the agent runs on WhatsApp.
Which channels can the agent answer on?
WhatsApp, Instagram DMs, your website chat and phone calls. One agent can cover several channels from the same knowledge and rules. We usually launch on one channel first and add the others once it is answering correctly.
Will the agent make things up?
It answers from the knowledge base you approve and is tested against real past conversations before launch. When it does not know, or a handoff rule applies, it passes the conversation to a named person instead of guessing.
Do you build agents in Arabic and Turkish?
Yes. We build in English, Arabic and Turkish, and the agent replies in the language the customer writes in. Each language is tested on real messages before it goes live.
Who owns the agent after launch?
You do. It runs on your accounts, and you receive the written knowledge base, actions and handoff rules in plain language so your team can update them. If you want us to keep it current, that can be a monthly retainer.
Want to talk it through?
Run your current site through our free audit, or book a discovery call and we will give you an honest read on what to build.
Or see transparent pricing for this kind of build.