Chatbots that resolve the request or hand it to the right person

A rule-based chatbot follows fixed buttons and keywords, while an AI chatbot uses a language model to understand messages in whatever wording the customer chooses. That lets it handle typos, follow-up questions and mixed languages. We still use rules where they matter: what the assistant may promise, which actions need verification and when it must hand over. The combination gives flexible conversation with predictable outcomes.

Yes, on the official WhatsApp Business API, where the assistant can hold free-text conversations, accept photos and voice notes, and offer reply buttons or list menus when the choices are fixed. Meta's rules allow AI assistants that serve your own customers about your business but not general-purpose AI chatbots, so each assistant stays scoped to your services. Replies inside the customer service window are free-form, later follow-ups use approved templates, and staff can take over from a shared inbox.

Yes, and we usually recommend it. The assistant's logic, knowledge and integrations sit in one back end, and each channel is a thin connector that adapts formatting, buttons and media to what that channel supports. Customers get the same answers wherever they ask, your team maintains one set of content, and analytics can compare how each channel performs.

The assistant passes the conversation to a person when a request falls outside its scope, the customer asks for someone, or a rule such as a complaint or a high-value order requires it. The handover routes to the right team in your helpdesk, CRM or shared inbox, with the transcript, the details already collected and a short summary. Outside staffed hours, the assistant sets expectations and creates a ticket instead.

Yes, through APIs to the systems that hold the data, such as your store platform, booking tool, CRM or a custom back end. A small integration layer sits between the assistant and those systems and exposes named actions, such as order status or move booking, instead of open database access. Read-only lookups come first. Actions that change something, like cancelling a booking, ask the customer to confirm, and account details are shared only once the customer is identified by a method you approve.

Yes, current language models generally handle English, Urdu, Roman Urdu and Arabic, including messages that switch between them mid-sentence, but quality varies by model and dialect. We test each language on real messages from your customers before choosing a model. The assistant replies in the customer's language, website widgets support right-to-left layout, and WhatsApp templates are prepared in every language you serve. Fixed wording such as policy text can be reviewed by your team per language.

It can, which is why we restrict where its answers come from and test it against hostile conversations before launch. Prices, stock, policies and eligibility are read from your systems or approved documents, never from the model's general knowledge. Topic limits stop it discussing competitors, legal or medical advice and anything outside your services, and test scripts include customers trying to talk it into discounts or unrelated tasks. Transcript reviews after launch catch tone and accuracy problems for correction.

Yes. The assistant asks the qualifying questions you define, validates contact details and creates or updates the lead in your CRM with the source page, campaign and full transcript attached. Qualified leads can be booked straight into an adviser's calendar or routed to a salesperson on WhatsApp. Consent for follow-up messages is recorded with the lead, built with UK GDPR and PECR in mind.

Yes. We index your website pages, FAQs, help articles and documents such as price lists or brochures, and the assistant answers from them with a link to the page it used. The index re-syncs when pages change, so answers follow your site rather than a copy pasted into a prompt. When a question is not covered, the assistant says so and offers a person instead of guessing. Large or permission-restricted libraries call for a fuller retrieval system.

Use an off-the-shelf platform if your help centre answers most questions, one channel is enough and its pricing works at your volume. A custom assistant earns its place when it must look up orders or bookings in your own systems, serve website, WhatsApp and app from one back end, handle Roman Urdu reliably or follow handover rules a platform cannot express. Per-conversation or per-resolution fees also add up at high volume. We give that recommendation during discovery, even when it means not building.

Timelines depend on the number of channels, the integrations needed and how much existing content and conversation history we can work from. An assistant covering frequent questions and lead capture on one channel can often launch within a few weeks of discovery. WhatsApp approval steps, account integrations and additional languages add time, so larger assistants are usually released channel by channel.

Pricing depends on scope, and you receive a fixed quote after discovery. The build is shaped by the number of channels, conversation flows, integrations and languages. Running costs include model usage per conversation, hosting and, for WhatsApp, Meta's charges for template messages. We project the monthly figure from your expected conversation volumes, so you can compare the assistant with the staff time those same requests take today.

Yes, through an admin view we build for content changes: answers, opening hours, holiday messages, handover rules and the documents the assistant draws on. Changes to integrations, conversation flows or the model still go through developers and the test suite, because they can break things customers rely on. After launch, either your developers take over with our documentation, or we provide support covering transcript reviews, new flows, WhatsApp template updates and model upgrades, agreed in a written scope with a named lead.